EPA
United States
Environmental Protection
Agency
Environmental Research
Laboratory
Corvallis, OR 97333
EPA/600/3-90/078
September 1990
Research and Development
Biospheric Feedbacks to Climate Change:
The Sensitivity of Regional Trace Gas
Emissions, Evapotranspiration, and Energy
Balance to Vegetation Redistribution
-- Status of Ongoing Research --
Edited By:
Hermann Gucinski, Danny Marks, and David P. Turner
-------
Biospheric Feedbacks to Climate Change: The Sensitivity of
Regional Trace Gas Emissions, Evapotranspiration, and
Energy Balance to Vegetation Redistribution
-- Status of Ongoing Research --
Edited by:
Hermann Gucinski, Danny Marks, and David P. Turner
The information in this document has been funded by the U.S. Environmental Protection
Agency as part of the Global Climate Research Program. The work presented here was
conducted at the Corvallis Environmental Research Laboratory. This report has been
subjected to the agency's peer and administrative review and it has been approved for
publication as an EPA document.
Mention of trade names or commercial products does not constitute endorsement or
recommendation for use.
n
-------
Acknowledgements
The successful completion of this report would not have been possible without the
support and assistance from a great number of individuals. We would like to ack-
nowledge a few of those here. The editors would like to particularly recognize and thank
Peter A. Beedlow, Technical Director, EPA Global Climate Research Program, Environ-
mental Research Laboratory-Corvallis, for his support and assistance in the development
of this report.
We also want to acknowledge the hard work and dedication of the Global Biogeo-
chemistry Team and associated staff:
Joseph V. Baglio
William G. Campbell
Jayne Dolph
James Lenihan
Kelly Longley
George King
Greg Koerper
Bruce McVeety
Ronald P. Neilson
Donald L. Phillips
Derek Pross
Richard Vong
Andrew G. Wones
We are grateful for the thoughtful criticism provided by the members of the public
peer review panel for this report:
Frank Davis
Ralph Dubayah
Upmanu Lall
Nikolas L. Nikolaidis
Phillip Sollins
Nathan Stevenson
Denis White
University of California, Santa Barbara
University of Maryland
Utah State University
University of Connecticut
Oregon State University
National Park Service
NSI Technology Services, Inc.
We also want to thank several other scientists for their direct and indirect contribu-
tions to this effort:
Bert Drake
Rik Leemans
Dennis Lettenmaier
Steve Running
Tom Smith
Starley Thompson
Smithsonian Environmental Research Center
National Institute of Public Health and
Environmental Protection, Netherlands
University of Washington
University of Montana
Oak Ridge National Laboratory
National Center for Atmospheric Research
in
-------
BIOSPHERIC FEEDBACKS TO CLIMATE CHANGE: THE SENSITIVITY OF
REGIONAL TRACE GASES EMISSIONS, EVAPOTRANSPIRATION, AND
ENERGY BALANCE TO VEGETATION REDISTRIBUTION.
—STATUS OF ONGOING RESEARCH—
TABLE OF CONTENTS
Executive Summary
I Introduction
Authors: Hermann Gucinski David P. Turner
II A Geographic Database for Modelling the Role of the Biosphere in Climate
Change
Authors: William. G. Campbell John K. Kineman
Danny Marks Robert T. Lozar
',111 The Sensitivity of Potential Evapotranspiration to Climate Change over
the Continental United States
Author: Danny Marks
IV Characterizing the Distribution of Precipitation over the Continental
United States using Historical Data
Authors: Jayne Dolph Danny Marks
V Evaluation of Geostatistical Procedures for Spatial Analysis of
Precipitation
Authors: Donald L. Phillips Danny Marks
Jayne Dolph
VI Effects of Global Climate Change on Global Vegetation
Author: George A. King
VII Toward a Rule-Based Biome Model
Authors: Ronald P. Neilson Greg Koerper
George A. King
VIII Effects of Climate Change on Carbon Storage in Terrestrial Ecosystems:
Equilibrium Analyses at the Global Level
Authors: David P. Turner Rik Leemans
IX Climate Change and Vegetation Derived Isoprene Emissions
Authors: David P. Turner Joseph V. Baglio
Derek Pross Andrew G. Wones
Bruce D. McVeety Richard Vong
Donald L. Phillips
IV
-------
EXECUTIVE SUMMARY
Implications of Global Climate Change
There is an emerging scientific consensus that global climate will
change over the next several decades in response to increasing
concentrations of greenhouse gases. However, there is less
agreement on the magnitude and timing of that climate change.
Among the major uncertainties are the role of atmospheric water
vapor and clouds and the potential positive or negative feedbacks
to climate change mediated by the biosphere. This document
provides an analysis of the sensitivity of biospheric processes,
including evapotranspiration, carbon storage in terrestrial
ecosystems, and biogenic trace gas emissions that could affect the
rate and extent of climate change through feedback processes.
Large Scale Data Requirements
Continental- and global-scale models require very large data bases
to analyze patterns of precipitation, soil moisture, and other
important processes which effect vegetation re-distribution. As a
step towards development of spatially distributed models, data
bases on terrain structure, soils, vegetation and land use,
precipitation and runoff, air temperature, humidity, wind, solar
radiation, and other parameters were assembled, evaluated for
accuracy and reliability, and incorporated into suitable Geographic
Information Systems (GIS).
Effects on Continental Scale Evapotranspiration
Simulation of potential evapotranspiration shows that climate
change will alter evaporative stress across the U.S.
Evapotranspiration affects the magnitude and spatial distribution
of soil moisture and the regional water balance, which determines
the condition and distribution of vegetation over the globe.
Historical climate data were used to calculate potential
evapotranspiration (PET) for the continental US by applying a
turbulent transfer model to the hydrologic/climatologic data set.
Results show that evaporative stresses are maximized in the
southwestern US during summer. These findings differ from previous
predictions based solely on air temperature. General circulation
models (GCM) predict increases in wind, temperature and humidity
for the continental U.S. Applying the GCM data to the
evapotranspiration calculation illustrates how minor differences in
predicted climate conditions can result in very different
distributions of evaporative stress. Predictions of temperature,
wind and humidity from the GFDL model lead to calculation of very
high PET in the midwestern U.S., while the GISS model, leads to
more modest increases in PET in the same region. These findings
have important implications to the future of agriculture in this,
the chief food-producing belt of the country.
-------
Large—Scale Precipitation Analysis
An assessment of the utility of historical data for providing
spatially distributed, large scale precipitation estimates
indicates a low elevation bias in mountainous regions. Historical
hydro-meteorological data were used to generate surfaces which
characterize the spatial distribution of precipitation and runoff
at regional- to continental-scales. A distributed water balance
was used to assess the utility and limitations of historical data
for providing estimates of precipitation across the continental
U.S. The precipitation and runoff surfaces indicate a deficit in
measured precipitation in comparison to measured runoff in
mountainous regions of the western U.S. The incorporation of snow
measurements into the precipitation record for these regions
significantly improves the water balance calculations and enhances
the utility of historical data for providing spatially distributed
estimates of precipitation at scales compatible with GCM analysis.
Geostatistical Procedures for Spatially Distributing Climate Data
Simple interpolation techniques for distributing climatological
data over the landscape fail to consider the effects of topography.
For precipitation, this may result in considerable error in
mountainous regions. Three geostatistical procedures, kriging,
elevation detrended kriging, and co-kriging, applied to a small
(300,000 km2) northwest US river basin were evaluated for the
precision and accuracy in the estimation of mean annual
precipitation at a grid of points across the landscape. The latter
two methods explicitly incorporate precipitation/elevation
relationships, and yielded estimates with improved accuracy and
precision. These methods appear promising for spatially
distributing climate data consistent with topographic influences at
this scale. The development of this approach to larger regions
will require piece-wise estimation of sub-basins, or the
development of an orographic precipitation model.
Re-distribution of Terrestrial Vegetation for a doubled CO, Climate
The potential redistribution of vegetation types in response to
climate change has been estimated in several analyses using
climate/vegetation correlation- systems such as that of Holdridge
combined with GCM climate scenarios. A review of results using
five different GCMs revealed a general agreement in terms of the
sign of the predicted change in the areal extent of specific
vegetation types, i.e., deserts, boreal forest, and tundra biomes
decrease, while grasslands and temperate and tropical forests
generally increase in area. These changes reflect the increases in
precipitation expected in temperate and tropical areas, and the
effect of rising temperatures in high latitudes that displace or
eliminate tundra and boreal biomes. The proportion of land surface
VI
-------
area changing from one vegetation type to another ranged from 16 to
56 % depending on the GCM used.
"Rule-based"- Approaches to Vegetation Modeling
The present climate/vegetation models or correlations are based on
relatively few climatic variables and in this study refinements are
made by relating the seasonal timing and magnitude of temperature,
precipitation and runoff patterns to the physiological reguirements
of specific vegetation types. For example, large winter
precipitation allows a closed canopy forest to exist, while reduced
winter precipitation produces a lower stature forest or savanna.
Using seasonal precipitation and threshold values for different
life-forms, the present model was calibrated and then tested for
the biomes of the US. The overall success in predicting US biomes
was 79%, with the best predictability at 94% for eastern forests,
and the poorest success (O%) for a small grassland biome in
northern California. Incorporation of potential evapotranspiration
into the rule base is anticipated and this will allow application
to future climates, e.g. doubled CO2 climate scenarios, while
"secondary" rules will improve model reliability by incorporating
disturbance, such as fire, and biotic interactions, including the
effect of CO2 enrichment and changes of water use efficiency of
plants.
Effect of Climate Change on Carbon Storage
An estimation of potential changes in storage of above- and below
ground carbon in response to climate change was made based on the
redistribution of vegetation types. Representative carbon pools
were assigned to each vegetation type and the distribution of those
vegetation types under the current climate and a double-CO2 climate
were used to predict changes in terrestrial carbon storage.
Results from vegetation redistribution using four different GCMs
and the Holdridge climate/vegetation correlation system were
compared. All GCMs predicted a net flux of carbon from the
atmosphere to the biosphere when considering just the aboveground
biomass. This flux reflected increases in the areal extent of
carbon rich vegetation types such as tropical humid forests. Two
of the GCMs predicted a net loss of below ground carbon because of
large decreases in the areal extent of tundra ecosystems with a
high level of below ground storage. These analyses suggest the
potential for a moderate negative feedback to global warming via
accumulation of carbon in terrestrial ecosystems.
Climate Change and Isoprene Emissions
A global model was developed for estimating spatial and temporal
patterns in the emission of isoprene from vegetation under the
current climate and doubled-CO2 climate scenarios. Current
emissions were estimated on the basis of vegetation type, foliar
biomass (derived from the satellite-generated Global Vegetation
vii
-------
Index), and global databases for air temperature and photoperiod.
Doubled C02 climate emissions were estimated based on predicted
changes in the areal extent of different vegetation types, each
having a specific rate of annual isoprene emissions. The isoprene
emissions under a doubled CO2 climate were about 25 % higher than
current emissions due mainly to the expansion of tropical humid
forests which had the highest vegetation-specific emission rates.
An increase in isoprene emissions could be expected to increase
atmospheric concentrations of ozone and methane which are important
greenhouse gases, however, detailed treatment of this question
awaits incorporation of these emission surfaces into 3-D
atmospheric chemistry models.
Research Directions
Fundamental questions remain about how ecological features can be
characterized across different temporal and spatial scales and be
integrated into predictive models of biosphere-climate interaction.
Current G€Ms do not include a realistic terrestrial biosphere, and
operate at scales that are not ecologically meaningful. While
these models are likely to improve over the next decade, it is
important that we provide a link between terrestrial ecology and
the atmospheric sciences in the context of climate change research.
The work presented in this report is an important first step in
this direction. Ecosystem models must be scaled up to regions
comparable to GCM analysis, and equilibrium approaches to
evaluating biospheric responses to climate change must give way to
more dynamic or transient approaches.
Evaluation of ecosystem responses to climate change will require
models which combine hydrology, nutrient cycling, and landscape
ecology. Short- to mid-term research objectives require progress on
three major fronts. The first is to improve the physical and
mechanistic basis of models of the climate effects on biospheric
processes so that they can be more effectively simulated for
different climate conditions. This will require laboratory, growth
chamber, and field experiments. A second is to develop simulation
models of these processes which can be run at large regional- to
global-scales. This will include an assessment of the effect of
scaling on simulation of ecological processes, and development of
"top down" models which treat features of the biosphere
hierarchically. The third involves integrating these methodologies
with efforts to improve both the spatial scale and characterization
of the terrestrial biosphere in the next generation of GCMs.
Vlll
-------
BIOSPHERIC FEEDBACKS TO CLIMATE CHANGE: THE SENSITIVITY OF
REGIONAL TRACE GASES EMISSIONS, EVAPOTRANSPIRATION, AND
ENERGY BALANCE TO VEGETATION REDISTRIBUTION
- STATUS OF ONGOING RESEARCH -
INTRODUCTION
Hermann Gucinski and David P. Turner
NSI Technology Services, Inc.
U.S. Environmental Protection Agency
Environmental Research Laboratory
200 SW 35th Street
Corvallis, OR 97330
(503) 757-4600
1-1
-------
There is an emerging scientific consensus that global climate will change over the next several decades
in response to increasing concentrations of greenhouse gases (IPCC 1990). However, there is less
agreement on the magnitude and timing of that climate change. Major uncertainties involve the role of
atmospheric water vapor-cloud water balance, and the potential positive or negative feedbacks to climate
change mediated by the biosphere.
The Global Change Research Program at the Corvallis Environmental Research Laboratory (ERL-C) was
initiated in 1987 with the goal of examining potential effects of climate change on terrestrial ecosystems.
The program has grown to focus in part on evaluation of biospheric feedbacks to climate change,,
especially as influenced by vegetation change. This area of research is of importance to EPA because of
its influence on the rate and magnitude of climate change, and hence on the formulation of national policy
regarding adaptation and mitigation. Early efforts at ERL-C have been on the development of
climate/vegetation correlations that could be used to predict redistribution of vegetation types in response
to climate change (Neilson et al. 1989). This work is in progress and more recent efforts have addressed
the development of databases and distributed models for simulating atmosphere-biosphere interactions.
The work described in this document was performed during the 1990 Fiscal Year and represents the initial
stages of a research effort that will extend into 1991 and beyond. The document is thus intended to report
on work in progress in the areas of (1) examining spatial and temporal patterns in the current interaction
between the biosphere and the atmosphere, (2) assembling and transforming the data bases required to
understand vegetation dynamics and the biophysical constraints acting on the biosphere, and (3)
1-2
-------
discussing approaches to quantifying biospheric feedbacks to climate change. The various chapters are
a sampling of the range of research topics being addressed by the global change program at ERL-C.
Climate-vegetation interactions have been the object of study for some years. Research has aimed to
clarify geographical patterns in vegetation distribution as they relate to climate, and to identify physiological
mechanisms by which plant species are adapted to particular climate regimes. The reciprocal question,
how plants or ecosystems influence the climate, has been addressed only recently. It is recognized that
significant changes in global climate are likely to drive changes in the distribution of vegetation types.
Major research questions must deal with how the flux in mass and energy from the biosphere influences
climate under current conditions and how the flux will potentially influence climate as anthropogenically
driven climate change begins to affect vegetation distribution.
The biosphere-atmosphere flux with which this document will be most concerned includes 1) water and
water vapor, and 2) radiatively and photochemically important trace gases, particularly carbon dioxide
(C02) and nonmethane hydrocarbons. A principal concern is how changes in vegetation in response to
climate change will influence the flux.
The structure of the document is as follows: The introduction briefly reviews background material relating
to several of the important feedback mechanisms and discusses research directions. We then begin with
an examination of the necessary global data bases required for assessing factors which influence or
produce biospheric feedbacks (II). There follows a series of chapters concerned with the hydrologic cycle.
A spatially distributed turbulent transfer model is developed to evaluate the sensitivity of potential
evapotranspiration to climate change (III). Continental-scale precipitation-runoff relationships for the U.S.
are characterized to examine the utility of the historical precipitation record for providing spatially distributed
1-3
-------
estimates of precipitation (IV). Thirdly, an evaluation of geostatistical procedures which could be used to
estimate regional precipitation distribution and volume is presented (V). A "top down" approach for
evaluating global vegetation redistribution in response to climate change is then presented (VI). The need
to base this approach on mechanistic principles is examined in the chapter on a rule-based biome model
(VII). The final chapters proceed from the estimates of current and future vegetation distributions to
evaluate the effect of climate change on carbon storage and carbon releases (VIII), as well as on emissions
of biogenic trace gases, which, by their importance in tropospheric chemistry, affect the concentration of
greenhouse gases such as methane (IX).
Biospheric Feedbacks
The application of the concept of feedback to biological systems began with the development of
cybernetics in the late 1940s (von Bertalanffy 1968). The concept arose in the field of engineering but it
has been recognized that feedback is a^basio. mechanism of self regulation in organisms and at higher
c^\ \_y
levels of organization. The basin idea is that a signal impinging on a component of a system initiates an
output or change in output from that component which amplifies (positive feedback) or dampens (negative
feedback) the magnitude of the original signal.
There are several applications for the feedback concept in the climate-biosphere system and it may be
useful to make some distinctions based on spatial and temporal scales. In climate dynamics, the signal
is a change in mean global temperature. In an equilibrium approach, one establishes what the climate
would be, for example, under a doubled-C02 scenario with no change in the biosphere. General circulation
models (GCMs), which account for the energy balance of the planet and atmosphere as well as the fluxes
of water, permit such an evaluation. One then applies the climate scenario to models of the biosphere,
determines how biospheric flux of mass and energy changes, and estimates how the new pattern of flux
1-4
-------
would further influence the climate, again using a GCM. A specific example would be a potential increase
in methane emissions driven by warming of high latitude wetlands. Since atmospheric methane would
increase, and methane is a "greenhouse gas", the associated warming would be a positive feedback to the
original doubled-CO2 warming. Lashof (1989) has recently reviewed a variety of potential feedbacks to
climate change using this approach and compared their relative magnitudes by reference to a gain factor.
The analysis of biospheric feedbacks using a dynamic or transient approach is considerably more complex.
Here a GCM must be coupled to biospheric processes and run over time intervals from decades to
centuries. At present, the biosphere is largely ignored in the GCMs. However, rapid strides are being
made in this area of much needed research. Detailed process models for the hydrologic cycle, nutrient
cycles and the exchange of trace gases and energy are being developed which will eventually be coupled
to GCMs.
A brief review of some of the major feedback mechanisms is included here in order to put some of the
material contained in this report in perspective.
Trace Gas Feedbacks
Water is the most important greenhouse gas, and is largely accounted for in terms of feedbacks in the
global radiation budget of GCMs. There are several important greenhouse gases besides water vapor,
including carbon dioxide (CO2), methane (CH,,), nitrous oxide (N2O), and ozone (O3). Concentrations of
these gases are not at chemical equilibrium in the atmosphere, an observation which has prompted the
notion that the continuous processing of materials by the biosphere maintains their steady state (Lovelock
1989). Global budgets of these trace gases reveal that biological processes do indeed account for most
1-5
-------
of the flux. Since the biosphere is expected to respond significantly to anthropogenically-induced climate
change, the concentrations of the trace gases might also be influenced. This produces the potential for
positive or negative feedbacks to climate change from the biosphere.
Carbon dioxide is the dominant trace gas among the non-water vapor greenhouse gases. Its concentration
is determined by many factors collectively characterized as the global carbon cycle. That cycle includes
large atmosphere-terrestrial biosphere and atmosphere-ocean flux. The flux is bi-directional, and the
atmospheric and surface pool sizes remain relatively constant. There is a seasonal oscillation of
approximately 10 ppm in the atmospheric C02 concentration in the northern hemisphere (Keeling et al.
1989) due to net photosynthesis in the summer and a release of CO2 via respiration in the winter. A
change in the flux rates to and from the atmosphere over longer time frames results in changes in
atmospheric pool size.
The C02 concentration has varied widely over geological time, but has had a range between two and three
hundred ppm over the last glacial-interglacial cycle, i.e., approximately 100,000 years (Barnola et al. 1987).
Long term changes in C02 concentration appear to be responses rather than initiators of climate change
and appear to be generally acting as positive feedbacks, i.e. changes in orbital forcings may initiate climate
change but biospheric or other factors which change atmospheric C02 amplify the original change. The
current global concentration is approximately 350 ppm, having risen from 280 ppm in the 1800's (Neftel
et al. 1985) and about 200 ppm during the last glacial maximum (Barnola et al. 1987). Further increase
in concentration over the next few decades is highly dependent on population growth and energy policy
and is expected to be large enough to strongly impact the climate.
1-6
-------
One of the mechanisms by which the atmospheric C02 concentration may change is via the storage of
carbon in vegetation and soils. The warming since the last glacial maximum induced large changes in the
distribution of vegetation types which may have influenced the rise from 200-280 ppm CO2. Changes in
above- and below-ground carbon pools may be a function of both changes in land surface area i.e. sea
level changes, and changes in the areal extent of different biomes, each having biome-specific amounts
of above- and below-ground carbon (Prentice and Fung in press). Evidence from the pollen record, for
example, reveals the northward migration of boreal forest since the last glacial maximum (Davis 1981). The
anticipated global warming of 3.5±1.5°C is likely to produce additional large changes in the areal extent
of different vegetation types. We examine the associated changes in terrestrial carbon pools and related
fluxes of CO2 in chapter VIII of this document.
Methane and nitrous oxide have relatively long atmospheric lifetimes, 10 years and 120 years respectively.
Atmospheric concentrations of both gases are also increasing currently, most likely due to anthropogenic
factors, and they are predicted to account for approximately 25% of the increase in global temperature due
to the "greenhouse effect" over the next several decades (Ramanathan et al. 1985). During previous
periods of warming, atmospheric concentrations of these gases increased, suggesting they were part of
a positive biospheric feedback to the warming (Khalil and Rasmussen 1989). However, there are large
uncertainties about future feedbacks associated with methane and nitrous oxide (Lashof 1989). Biogenic
sources may increase due to warming of high latitude biomes where there are large pools of potentially
decomposable belowground organic matter. There may also be release of large amounts of pre-existing
methane from marine sediments or from frozen high latitude soils.
Other biogenic trace gas emissions likely to increase with global warming include nonmethane
hydrocarbons (NMHC). These compounds are emitted by most types of vegetation, but emissions vary
1-7
-------
widely as a function of vegetation type and environmental variables such as temperature and light
(Zimmerman et al. 1978). Oxidation of NMHCs reduces the concentration of the hydroxyl radical, thus
serving to increase the atmospheric lifetime of methane and other important trace gases, and may result
in production of tropospheric ozone, a significant greenhouse gas (Logan et al. 1981). The exponential
rise of vegetative NMHC emissions with increasing temperature (Tingey et al. 1981) suggests that the
warming associated with climate change is likely to increase the global emissions. There may also be
increased emissions of NMHCs if the area! extent of forest ecosystems expands, as the modeling efforts
in this document suggest will occur with a doubled-CO2 climate. The corresponding increases in ozone
and methane concentrations will be positive feedbacks to the climate warming.
Water Vapor Feedback
As noted, the dominant greenhouse gas in the Earth's atmosphere is water vapor; it provides a large
potential feedback to climate change. About half the projected temperature increase in doubled CO2 GCM
runs is due to the positive feedback of increased atmospheric water vapor (Hansen et al. 1984). Much of
the flux of water vapor from the land surface is from the vegetation, perturbations of which are likely to
significantly alter the water vapor feedback (Shukla and Mintz 1982). Again, these biospheric processes
are poorly modeled in current GCMs and are a current research objective in the modeling community.
1-8
-------
Albedo Feedback
The Earth's albedo plays a significant role in its energy balance (Ramanathan et al. 1989) and hence its
climate. As climate warms there is a simple positive feedback in that the area of highly reflective ice cap
shrinks and more visible radiation is absorbed by the earth, a positive feedback. Much more complicated,
and as yet not fully understood, is the relationship between the global balance of atmospheric water vapor
and liquid water in the form of clouds, and the resulting balance in global albedo. Climate driven changes
in vegetation cover and the distribution of vegetation types would also result in changes in the global
albedo, although the magnitude of this effect may not be large (Dickinson and Hanson 1984). Future
treatments of potential changes in global albedo due to vegetation change must include anthropogenic
factors such as desertification (Sagan et al. 1979) and deforestation (Henderson-Sellers and Gornitz 1984).
Research Directions
Biospheric feedbacks to climate change operate across a wide range of spatial and temporal scales, and
fundamental questions remain about how information at these scales can be integrated into predictive
models, including present and future general circulation models. As a start, the current emphasis on
equilibrium approaches to evaluating biospheric responses to climate change must give way to more
dynamic or transient approaches. We have suggested, for example, that transitional releases of carbon
during vegetation redistribution may produce sufficient feedbacks to alter equilibrium end points (King et
al. 1990). Prediction of C02 fertilization effects on plants will likewise require an understanding of transition
dynamics.
1-9
-------
Integrating information at different spatial and temporal scales must remain a principal short- to mid-term
research objective. This objective will require progress on three major fronts. The first would seek to
describe biospheric and physical processes relevant to climate change at the plot or stand scale. A
second would concern the ability to simulation these processes at broader spatial scales. The third would
involve coupling of these simulations with GCMs such that there is two way interaction between the climate
and the biosphere.
Evaluation of local or site-specific responses to climate change will require ecosystem models which
combine hydrology, nutrient cycling and population biology. Driven by GCM climate scenarios, such
models will become increasingly useful tools for evaluating the mechanisms which will dominate the
response of particular sites to climate change.
The optimal approach to "scaling up" of responses from local sites to regional and global levels remains
problematical. The varied spatial and temporal scales of relevant biospheric processes do not allow
treatment of all processes simultaneously. Consequently, efforts will be required to develop models
appropriate to each scale and one of the major research tasks of the next decade is to determine the
• optimum way of coupling such models.
These problems emphasize the need for concrete, short term research objectives, such as developing
realistic capabilities for flux estimates at scales of hundreds of meters to kilometers. One approach used
successfully in the past has been to bring a number of investigators to a common site. These scientists
combine efforts to address questions such as the magnitude of biosphere/atmosphere fluxes of trace gases
and energy. The First International Satellite Land Surface Climatology Program (ISLSCP) is the prototype
for such efforts. In that case the surface/atmosphere flux of CO2 was measured using both chamber and
1-10
-------
eddy correlation techniques; open path FTIR or tunable dye laser now appear to be additional approaches
for flux measurements which could be added to the suite of existing techniques. In combination with
remote sensing of ecosystem processes, these measurement programs will contribute significantly to
advances in our capacity to scale up from sites to regions.
Lastly is the problem of incorporating models of biospheric processes into GCMs. Current representations
of the ground surface such as the Biosphere-Atmosphere Transfer Scheme (BATS) of Dickenson et al.
(1986) have significant limitations. These models of biosphere/atmosphere interaction need improved
treatment of physiological processes and will require high resolution global databases to be initialized and
run. The increasingly sophisticated hardware and software associated with Geographic Information
Systems will help in the development of appropriate databases for land surface features including
topography, vegetation, soils, climate and land use. However, increased attention to spatial and temporal
heterogeneity in relevant parameters is needed. In this regard, remote sensing offers considerable promise
for initializing and eventually validating these models.
A Note on the Format of this Report
To reach the widest scientific audience, the report was assembled from chapters written as stand-alone
manuscripts suitable for publication in the peer reviewed literature. While this introduces some unevenness
in the document, we feel that the state of the science requires considerably more research before a
comprehensive review of the role of biospheric feedbacks in climate change will be possible. This
juxtaposition of research addressing a variety of issues and taking a range of approaches also highlights
the need for increased inter-disciplinary collaboration and interaction to address the most critical, and in
many ways most controversial, research issues. We see our efforts contributing in this arena, adding light
1-11
-------
as well as heat to the debate. We well understand the incompleteness of the scientific record on the
problem of forecasting the effects of global climate change, and well know the limitations of the
contributions contained herein, but we see them as building blocks for improved future understanding.
Chapter VI, "Effects of global climate change on global vegetation," is included in this report as an integral
part of the development of ideas for a rule-based approach to modeling vegetation change (Chapter VII),
and as a basis upon which the assessment of the changes in terrestrial ecosystem carbon storage (Chapter
VIII) and in plant emissions of trace gases (Chapter IX) is built. The chapter is also included in another
report (King et al. 1990) because it fulfills a similar requirement there. We believe this duplication adds
continuity and cohesiveness to both reports, and minimizes disruptive cross-references.
1-12
-------
REFERENCES:
Barnola, J.M., D. Raynaud, Y.S. Korotkevich, and C. Lorius. 1987. Vostok ice core provides 160,000-year
record of atmospheric C02. Nature 329:408-414.
Davis, M.B. 1981. Quaternary history and the stability of forest communities. In: D.C. West, H.H. Shugart
& D.B. Botkin, eds., Forest Succession: Concepts and Application. Springer-Verlag, New York. 517
pp.
Dickinson, R., and B. Hanson. 1984. Vegetation-albedo feedbacks. In: J. Hansen and T. Takahasi, eds.,
Climate Processes and Climate Sensitivity. Geophys. Mono 29., Maurice Ewing, American
Geophysical Union, Washington, DC. 5:180-186.
Dickinson, R.E., A. Henderson-Sellers, P.J. Kennedy and M.E. Wilson. 1986. Biosphere-Atmosphere
Transfer Scheme (BATS) for the NCAR Community Climate Model. NCAR Technical Note,
NCAR/TN-274 + STR, National Center for Atmospheric Research, Boulder, Colorado, 69 p.
Hansen, J., A. Lacis, D. Rind, G. Russel, P. Stone, I. Fung, and J. Lerner. 1984. Climate sensitivity:
Analysis of feedback mechanism. In: Climate Processes and Climate Sensitivity, Geophys. Monogr.
Serv., AGU, Washington, DC 29:130-163.
Henderson-Sellers, A., and V. Gornitz. 1984. Possible climatic impacts of land cover transformations, with
particular emphasis on tropical deforestation. Climatic Change 6:231-257.
Holdridge, LR. 1947. Determination of world formulations from simple climatic data. Science 105:367-
368.
Intergovernmental Panel on Climate Change. 1990. Scientific Assessment of Climate Change. Report for
WGI Plenary Meeting.
Keeling, C.D., R.B. Bacastow, A.F. Carter, S.C. Piper, T.P. Whorf, M. Heimann, W.G. Mook, and H.
Roeloffzen. 1989. A three-dimensional model of atmospheric C02 transport based on observed
winds: Analysis of observational data. Geophys. Mono. 55:165-236.
.
King, G.A., J.K. Winjum, R.K. Dixon, LY. Arnaut (eds.). 1990. Responses and feedbacks of forest systems
to global climate change. Report to the US EPA, Office of Research and Development. USEPA
Environmental Research Laboratory, Corvallis, Oregon.
Khalil, M.A.K. and R.A. Rasmussen. 1989. Climate-induced feedbacks for the global cycles of methane
and nitrous oxide. Tellus 416:554-559.
Lashof, D.A. 1989. The dynamic greenhouse: Feedback processes that may influence future
concentrations of atmospheric trace gases and climatic change. Climatic Change 14:213-242.
Logan, J.A., M.J. Prather, S.C. Wofsy, and M.B. McElroy. 1981. Tropospheric ozone: A global perspective.
J. Geophys. Res. 96:7210-7254.
1-13
-------
Lovelock, J.E. 1989. Geophysiology, the science of Gaia. Rev. Geophys. 27:215-222.
Neftel, A., E. Moor, H. Oeschger, and B. Stauffer. 1985. Evidence from polar ice cores for the increase
in atmospheric C02 in the past two centuries. Nature 315:45-47.
Neilson, P.P., G.A. King, R.L DeVelice, J. Lenihan, D. Marks, J. Dolph, W. Campbell, G. Click. 1989.
Sensitivity of ecological landscapes and regions to global climate change. Report to the US EPA,
Office of Research and Development, USEPA Environmental Research Laboratory, Corvallis,
Oregon.
Prentice, K.C., and I.Y. Fung. Bioclimatic simulations test the sensitivity of terrestrial carbon storage to
perturbed climates. Nature. In press.
Ramanathan, V., R.J. Cicerone, H.B. Singh, and J.T. Kiehl. 1985. Trace gas trends and their potential role
in climate change. J. Geophys. Res. 90:5547-5566.
Ramanathan, V., R.D. Cess, E.F. Harrison, P. Minnis, B.R. Barkstrom, E. Ahmad, and D. Hartmann. 1989.
Cloud-radiative forcing and climate results from the earth radiation budget experiment. Science
243: 57-63.
Sagan, C., O.B. Toon, and J.B. Pollack. 1979. Anthropogenic albedo changes and the earth's climate.
Science 206:1363-1368.
Shukla, J., and Y. Mintz. 1982. Influence of land-surface evapotranspiration on the earth's climate.
Science 215:1498-1500.
Tingey, D.T., R. Evans, and M. Gumpertz. 1981. Effects of environmental conditions on isoprene emission
from live oak. Planta 152:565-570.
von Bertalanffy, L 1968. General Systems Theory. George Brazilla, Publ. New York. 295 pp.
Zimmerman, P.R., R.B. Chatfield, J. Fishman, P.J. Crutzen, and P.L Hanst. 1978. Estimates on the
production of CO and H2 from the oxidation of hydrocarbon emissions from vegetation. Geophys.
Res. Lett. 5:679-682.
1-14
-------
A GEOGRAPHIC DATABASE FOR MODELLING THE ROLE OF THE
BIOSPHERE IN CLIMATE CHANGE
WILLIAM G. CAMPBELL, DANNY MARKS,
NS1 Technology Services Corp., Environmental Research Laboratory, US
Environmental Protection Agency, Corvallis, Oregon
JOHN J. K1NEMAN,
National Geophysical Data Center, National'Oceanic and Atmospheric
Administration, Boulder, Colorado
and
ROBERT T.LOZAR
Construction Engineering Research Laboratory, US Army Corps of
Engineers, Champaign, Illinois
Abstract. Research into the effects of global climate change will require an
integrated geographic database for landscape characterization, visualization, and
modelling. An integrated, quality controlled, global environmental database.
however, is not yet available. To answer this need, the US Environmental
Protection Agency, in cooperation with National Oceanic and Atmospheric
Administration's National Geophysical Data Center and the Army Corps of
Engineers Construction Engineering Research Laboratory, is developing a global,
quality controlled geographic database for use in regional, continental, and global
modelling. This database will use an approach similar to that employed by the
International Council of Scientific Union's Global Diskette Project for Africa.
The database will use raster architecture for data storage and will rely on .
Geographic Information System technology for data access, retrieval, analysis,
and output. Data access will not be tied, however, to any specific software
package. Development of filters to reformat data will be actively encouraged.
Although designed specifically for the EPA's Global Climate Research Program,
the proposed database will, nevertheless, be available to the scientific
community in general and should be particularly valuable for global and
continental scale modelling.
II - 1
-------
1. Introduction
1.1. Climate Change Research
Increases in the atmospheric concentration of CO2 and other greenhouse
gases resulting from anthropogenic emissions are likely to alter forcing
functions that drive global climate, although uncertainties remain
regarding modelled distributions of temperature and precipitation
(Dickinson, 1989; Schneider, 1989). Analyzing climate change effects
will require a large scale interdisciplinary research effort and must
include a coordinated research component for collecting and managing
global data. To effectively analyze and model these large and complex
datasets, flexible data management and analysis strategies need to be
developed. This paper will present current efforts to develop global
databases and tools for climate change research in association with the
US Environmental Protection Agency's (EPA) Global Climate Research
Program (GCRP). The GCRP is a large, interdisciplinary research effort
designed to assess: 1) the role of the biosphere in climate, 2) the response
of the biosphere to climate change, and 3) mitigative strategies related to
management of the biosphere.
1.2. Data Needs
The research community has recognized the important role that geographic
data play in assessing global change (ESRI, 1984; Rizzo, 1988; Tomlinson,
1988; Campbell et al., 1989). To characterize the effects of climate
change requires analysis of a wide variety of spatial data. This must
include development of an integrated, quality controlled geographic
database. The need for an integrated global database was addressed
recently in the first meeting of the International Geographical Union's
Global Database Planning Project (IGU, 1988). Tomlinson (1988), in his
opening remarks at this meeting, stated "...there is a perceived need for
geographical analysis and spatial data on a global scale". Similarly, the
Committee on Earth Sciences (CES).has identified documentation of earth
system change through observational programs and data management
systems as the first integrating priority of the U.S. Global Change
Research Program (CES, 1989).
- 2
-------
Many different national and international organizations are currently
involved in planning for or collecting data on a global, scale, although
these projects are in preliminary stages (e.g., International Geosphere-
Biosphere Project, MINITOPO coordinated by the U.S. Defense Mapping
Agency, etc.) and/or are designed to address specific needs such as the
SOTAR (soils and terrain) working group (ISSS, 1986) and the World
Digital Database for Environmental Sciences (Bickmore, 1988). Currently,
a quality controlled, global geographic database for modelling does not yet
exist, although some work has been done for individual continents
(Kineman eta/., 1990) and ocean areas (Lozar, 1990).
A critical component of a global effects database, must be Digital
Elevation Models (DEMs). DEMs are a critical component because terrain
structure largely controls energy balance and hydrologic response of the
landscape as well as influencing the global distribution of vegetation. A
geographic global effects database will play a key role in four general
areas of research, including: 1) landscape characterization, 2) dynamic
measurements, 3) surface generation, and 4) modelling.
1.2.1. Landscape Characterization
To properly characterize static components of the physical landscape,
data on the distribution of soils, current vegetation/land cover,
political/continental boundaries, terrain (elevation, slope, aspect), and
geomorphic regions will be required. These data will be used to
characterize the physical landscape, to determine the status and extent of
resources, to provide information on the resources at risk, and to provide
baseline information for models.
1.2.2. Dynamic Measurements
Information will be required on dynamic characteristics of the landscape,
including temporal and spatial distributions of evapotranspiration,
reflectance, albedo, soil moisture, and vegetative stress. Because of the
need to characterize the temporal dynamics of large regions, remote
sensing will be a critical element of the characterization efforts. These
data will be used to assess the role of the biosphere in climate dynamics
II - 3
-------
as well as to assess potential effects of climate on the terrestrial
components of the biosphere.
1.2.3. Surface Generation
New techniques will be developed to spatially distribute data from either
point measurements or physically or empirically-based models.
Information will be needed on the spatial and temporal patterns of
precipitation, runoff, evapotranspiration, and temperature. These data
will be used to: 1) estimate the spatial and temporal distributions of
surface properties and their response to changing climate conditions, 2)
measure the relative sensitivities of regions (biotic and/or physical) to
changes in global climate, and 3) provide dynamic input to a variety of
analytical procedures and predictive models.
1.2.4. Modelling
Much of the emphasis within climate effects research must be
development and application of spatially distributed models, including
'models of regional energy balance, vegetation redistribution, and biogenic
emissions. These models will require various types of data as inputs
including measurements on static characteristics, dynamic surface
properties, and generated surfaces. Future research efforts will require
development and application of spatially distributed models at continental
and global scales.
2. Data Requirements
2.1. Spatial Data Handling Tools
There are a variety of hardware and software options available for
analyzing very large spatial datasets. We consider four primary options
to be critical: 1) "open" or non-proprietary operating systems, 2)
extensive use of "next generation" mass storage devices such as CD-ROMS
and optical read/write disks, 3) ability to pre-process or filter data to
conform to various format requirements, and 4) incorporation of
Geographic Information System (GIS) technology for database access,
retrieval, analysis, and output.
- 4
-------
Reliance on non-proprietary operating systems (e.g., UNIX) dramatically
increases flexibility of the proposed database by reducing the need to rely
on a specific hardware vendor for accessing and analyzing the data. UNIX
is also relatively hardware independent and is rapidly becoming the
industry standard for workstations.
Next generation mass storage devices (e.g., CD-ROM) will be used as the
primary distribution medium for the proposed database. Because of the
cost efficiency of these devices, their use can significant increase the
accessability and availability of large global databases to a wide variety
of researchers. These devices are available for personal computers,
workstations, as well as for mini-computer and mainframe environments.
This technology has already been employed for geographic database
management by the World Data Center-A in Boulder, CO under the auspices
of the International Council of Scientific Unions (Allen, 1988; Clark and
Kineman, 1988).
Development of filters to aid in the transference of data between various
software packages will be critical. The emphasis on filters rather than
"standard" software systems requires that "open" systems, particularly
for input/output routines, must be actively encouraged. Systems with
unknown or proprietary formats will be of limited use within this scheme.
This "open" philosophy has several advantages over the use of a standard
package including: 1) analyses can be application or need driven, 2)
analyses can take advantage of the inherent strengths of particular
software, and 3) the complexity of the chosen software can be varied
depending on user expertise and complexity of application.
GIS will be required for database access, retrieval, analysis, and output.
GIS have four essential components: 1) data input and verification, 2) data
management, 3) data manipulation and analysis, and 4) data output
(Aronoff, 1989). GIS are typically broken into two fundamental types
depending on the method used to store spatial data. Raster systems
(including hierarchical raster) store data as an array or hierarchical array
of grid cells, while vector systems store data as a series of distinct
spatial units called polygons (Burrough, 1986). Inherent strengths and
limitations of these systems are listed in Table I. Because the basic
- 5
-------
Table I Comparison of vector and raster systems
VECTOR
Advantages
Good representation of phenomenological data structure
Compact data structures
Topology can be completely described with network linkages
Accurate graphics
Retrieval, updating and generalization of graphics and attributes are
possible
Disadvantages
Complex data structures
Vertical aggregation of map layers creates difficulties
Simulation is difficult because each unit has a different topological
form
Display and plotting can be expensive
The technology is expensive
Spatial analysis and filtering within polygons are impossible
RASTER
Advantages
Simple data structures
The combination of mapped data with remotely sensed data is easy
Various kinds of spatial analysis are easy
Simulation is easy because each spatial unit has the same size and
shape
. The technology is cheap and is being energetically developed
Disadvantages
Volumes of graphic data
The use of large cells to reduce data volumes means that
phenomenologically recognizable structures can be lost
Crude raster maps are considerably less beautiful
Network linkages are difficult to establish
Projection transformation can be time consuming
(adapted from Burrough, 1986)
- 6
-------
purpose of this proposed database is spatial modelling, raster will be the
primary data type. Software that relies exclusively on the vector data
structure will not be supported.
2.2. Quality Assurance/Quality Control
Quality assurance and quality control (QA/QC) for spatial data is in its
infancy, particularly in GIS and spatial modelling environments (Chrisman,
1984; Burrough, 1986; Bailey, 1988; Goodchild, 1988; Campbell and
Mortenson, 1989). Because analysis of spatial data and the application of
spatial models do not lend themselves to classical QA/QC methodologies
(e.g., replication), QA/QC must focus on identification and documentation
of error associated with spatial data. At present, QA/QC of global and
regional spatial data is very poor, with inadequate documentation and
quality control, non-standard formats, and general inaccessibility.
In order to be a useful and viable component of global database efforts,
QA/QC efforts must explicitly address defining, documenting, and tracking
error and uncertainty of spatial data. The following elements must be
addressed:
1) Research must be directed towards a basic understanding of
errors associated with geographic analysis and spatial
modelling. Although much work has been done with assessing
and modelling errors associated with remotely sensed data,
this effort must be expanded to spatial data in general.
2) Techniques must be developed that automate the tracking
and/or documentation of spatial error through geographic
processing and analysis. This must include consideration of
the propagation of spatial error as it relates to overlay
analysis and spatial modelling.
3) Procedures that will document and track error associated with
entry of spatial data must be developed and/or refined. This is
needed to better document or control errors resulting from
digitization or scanning of base maps.
- 7
-------
4) Documentation procedures for identifying error within spatial
data, including tracking of map lineage, must be developed.
This must include consideration of the spatial components of
lineage, since global data will tend to be a horizontal
integration of vastly different source materials.
5) Analogous methods to the scientific peer review process must
be developed for review of spatial databases. This can include
a network of research and test sites (including data centers,
research institutions, U.S. and foreign agencies, universities,
UN organizations, and others) to aid in evaluation and in
developing data, systems, and methods.
Continuing research into the above-mentioned issues must be an integral
research component within global database efforts, particularly because
of the large reliance of the global change community on spatial data,
geographic processing, and spatial modelling.
3. Database Development
3.1. Approach
The approach used in developing this database will consist of a series of
steps, including:
1) Develop, in conjunction with other national and international
data centers, a coherent, standardized, and quality controlled
global environmental database of integrated regional and
global data sets, including biotic, physiographic, hydrologic,
edaphic, climatic, and other environmental factors derived
from existing observations and remotely sensed data.
2) Develop, in conjunction with the database, appropriate
analytical tools and methods for observational analysis and
comparative studies of ecosystem and climate patterns.
- 8
-------
3) Establish a peer review process and network for evaluation of
the database and software development, and to ensure the
scientific usefulness of these efforts in regard to the research
needs of the global change science community.
4) Disseminate original datasets and various derivations as
digital products for analysis, experimental design and
modelling support.
5) Work closely with other groups to link the technology, data,
and methods of spatial data analysis with research and
applications, including theoretical/mathematical modelling.
6) Develop products for inclusion with the International Space
Year and the International Geosphere-Biosphere Program.
3.2. Cooperative Agencies/Roles
Development of a geographic database will be a cooperative venture
between three federal agencies: the U.S. Environmental Protection
Agency's (EPA) Environmental Research Laboratory-Corvallis (ERL-C) in
Corvallis, OR; the National Oceanic and Atmospheric Administration's
(NOAA) National Geophysical Data Center (NGDC) in Boulder, CO; and the
Army Corps of Engineers' (ACE) Construction Engineering Research
Laboratory (CERL) in Champaign, IL. Data will be acquired, however, from
a number of additional sources, including other federal agencies
(particularly the US Geological Survey), cooperating universities, and
national and international research centers.
Specific responsibilities of the three federal agencies (EPA, NOAA, and
ACE) have been identified. The EPA, as the lead agency within the GCRP,
will be responsible for program and project management and addressing
the scientific questions regarding the role of the biosphere in climate
change. Additionally, the EPA will provide direction on data requirements,
geographic coverage priorities, and tool development. The NGDC will be
responsible for global data management, analysis of status and trends of
ecological resources, landscape characterization and visualization, and
development of specialized data handling tools. Additionally, NGDC's
- 9
-------
extensive contacts with other federal agencies and international research
centers will form a bridge between the EPA and other organizations
involved in global data collection. CERL is the developer of the GRASS
Geographic Information System (ACE, 1988) and will be. utilized for
software development, including integration of spatially distributed
models into the GIS environment, development of analytical tools within
GRASS, and global data management activities. The three agencies will
also perform cooperative research in a number of areas related to global
data, including the development of quality assurance and quality control
mechanisms, characterization and visualization techniques, spatial
statistical techniques, and data management strategies related to spatial
modelling.
3.3. North American Prototype
A North American prototype will be scheduled for immediate development,
windowing existing databases from NGDC's existing global database and
using a similar approach to that employed in the International Council of
Scientific Union's Global Change Diskette Project for Africa (Kineman et
al., 1990). A detailed listing of the North American prototype database is
presented in Table II.
4. Applications
The GCRP is using geographic data for a variety of analytical procedures
and predictive models related to assessing the role of the biosphere in
climate change. The following examples show the role geographic data
play in analyzing and modelling these environmental data.
4.1. Landscape Characterization
Landscape characterization involves parameterizing static components of
the landscape. To assess the role of terrain structure in controlling or
influencing climate change, we have characterized the distributions of
elevations across the continental U.S. using OEMs (Figure 1). The DEM data
was overlaid with water resources regions obtained from the US
Geological Survey (Figure 2 - USGS, 1978). Following overlay,
hypsometric curves depicting the distributions of elevations were
- 10
-------
Table II North American Prototype Database
Resolution
Monthly vegetation index for 1985 10 minute
through 1988
Monthly temperature and
precipitation anomalies
Average monthly temperature and 30 minute
precipitation
Monthly skin temperatures
Topography
Percent water and urban cover
Vegetation and reliability
classes, soil classifications
Vegetation classes and albedo
Cultivation Intensity
Ecosystem classifications
Soil classifications
Soil classifications
Geographic and political
boundaries
10 minute
10 minute
10 minute
1 degree
1 degree
1 degree
30 minute
1 degree
2 minute
10 minute
Source
NOAA, 1990
UNEP, 1990
Legates and
Willmott, 1990
GSFC, 1979
Kineman, 1989a
Kineman, 1989b
Wilson and
Henderson-
Sellers, 1985
Matthews, 1983
Matthews, 1983
Olson et a/.,
1983
Staub and
Rosenzweig,
1987
UNESCO, 1973
CIA, 1990
- 11
-------
ro
Fig. 1. Digital Elevation Model at 10 km resolution for the continental
United States.
-------
(02)
Water Resources
Regions
Fig. 2. Water resource regions for the continental United States.
-------
generated for water resources regions and subregions (Figure 3). This
data is being used in a quantitative analysis of the role of terrain
structure on the surface energy balance and potentiaJ sensitivity of the
landscape to climate change (see Chapter IX).
4.2. Dynamic Measurements
Measurements that characterize the dynamic properties of the landscape
are required to assess the spatial and temporal changes in
evapotranspiration, reflectance, albedo, soil moisture, and vegetative
stress. To assess dynamic changes in vegetation over the globe, we have
obtained information on the Global Vegetation Index (GVI - NOAA, 1990)
formatted for use in the GRASS GIS (Figures 4 and 5). Empirical
relationships between GVI and foliar biomass are being used to produce
estimates of the temporal and spatial dynamics of estimated biomass for
input to models predicting non-methane hydrocarbon emissions.
4.3. Surface Generation .
Techniques to distribute measurements taken at a point in space (e.g.,
runoff, precipitation) over large geographic regions are being developed to
analyze the spatial dynamics of selected variables and to provide input for
spatially distributed models. In particular, distributed estimates of
runoff and precipitation are required for areas ranging in size from large
water resource regions (50,000 - 500,000 sq. km) to continental and
global scales. This will include application of existing techniques (e.g.,
kriging) to larger areas to development and application of new techniques,
such as spatially distributed modelling and artificial intelligence.
To assess the spatial and temporal dynamics of rainfall and runoff over
large land areas, we have used historical data on precipitation (Quinlan et
al., 1987) and runoff (Wallace et al., 1990) and distributed these
measurements over the continental U.S. using an inverse distance
algorithm. Figure 6 shows the long-term (1948-1988) average annual
precipitation and runoff. The resulting data is being used in a preliminary
analysis of the adequacy of the historical record for defining a
distribution function of precipitation at the continental scale (Dolph,
1990; see also Chapter IV).
- 14
-------
Ul
CO
d~
CD
d~
CD
rt
c\j
d"
q
d"
~T
0
1000
2000
3000
elev
Fig. 3. Hypsometric integral for the Pacific Northwest water resource
region.
-------
Fig. 4. Global Vegetation Index for January, 1988 (see also Chapter IX).
-------
Fig. 5. Global Vegetation Index for July, 1988 (see also Chapter IX).
-------
GRASS GRAPHICS WINDOW
00
Precipitation
Annual Surfaces
1948-1988
• 0)
1-50 TOO.
51-100 too.
Runoff
O ) 101-350 am
i! _j 4 ) 351-600 ran
5 ) 601-850 ran
!| | Q } I 851-1100 ran
7) 1101-1350 ran
8 ) ! 1351-1600 ran
III 9 ) | 1601-1850 ran
10) 1851-2100 ran
11) 3101-3550 ran
|12 ) 2551-3000 ran
113 I > 3000 ran
Fig. 6. Long-term (1948-1988) average annual precipitation and runoff
surfaces for the continental United States (from Dolph, 1990; see also
Chapter IV).
-------
In addition to analyzing these data using simple interpolation techniques,
we are also investigating the application of spatial statistical techniques
to large areas. Phillips et al. (1990; see also Chapter V) investigated the
application of three spatial statistical techniques for estimating the
distribution of precipitation across the Willamette Valley in Western
Oregon. The three techniques were (1) kriging, (2) kriging elevation-
detrended data, and (3) cokriging with elevation as an auxiliary variable.
Both detrended kriging and co-kriging showed improved estimates of
precipitation when compared to simple kriging. The distribution of
precipitation over the Willamette Valley using co-kriging is depicted in
Figure 7.
4.4. Modelling
GIS provides a unique tool for modelling of spatial data, primarily through
the development and application of spatially distributed modelling for
energy and water balance, biogenic emissions, and vegetation
redistribution. Marks (1990; see also Chapter III) has applied a spatially
distributed turbulent transfer model to estimate the surface energy
balance across the continental U.S. Critical inputs include Digital
Elevation Models (e.g., see Figure 1), temperature, wind, and humidity. To
create spatially distributed estimates.of temperature, long-term monthly
estimates were obtained from the Historical Climatological Network
(Quinlan et al., 1987). These data were then converted to potential
temperatures (1000 mb surface), interpolated across the continental U.S.
using an inverse-distance algorithm, and then re-mapped onto the DEM grid
(Figure 8). The resulting temperature surface more accurately reflects
the distribution of elevation across the U.S. than a simple interpolation
scheme. This data can then be combined with distributed estimates of
terrain structure, wind fields, and humidity, to provide model input for
estimating Potential Evapotranspiration (Figure 9).
5. Discussion
An important consideration in designing a long-term plan for management
and analysis of geographic data is the ability to adapt to new technology
and adjust for future growth, in terms of data, hardware, and software.
- 19
-------
Longitude (deg)
90
210
EC
2V1
270
Fig. 7. EsuT,;.ies of precipitation for the Willamette Valley. OR using
cokricirvj ;frcm Philips e; ai. 1990: 3ee also Chapter V).
M - 20
-------
-25C
Fig. 8. Monthly air temperature for the continental United States (from
Marks, 1990; see also Chapter III).
-------
ro
25mm
250mm
1500mm
Fig. 9. Monthly Potential Evapotranspiration for the continental United
States (from Marks, 1990; see also Chapter III).
-------
Two general developments will guide the incorporation of new data. First,
given the long timeframe of the global effects research, newer and better
data will likely become available and can be added to. the existing
database. This will include data with better geographic coverage, and
increasing spatial and temporal resolution. Secondly, given the large
increases in the capabilities of workstations to effectively handle large,
CPU-intensive applications, it is likely that future analyses and modelling
will be able to more effectively handle applications such as fine
resolution spatially distributed modelling and analysis of medium
resolution remotely sensed imagery on a global scale. The reliance on
open architecture should allow better adaptation to new technology.
Future software development efforts are likely to focus on two general
areas. First, better analytical methods are needed for analyzing data in
the spatial domain. Although GIS technology has increased the importance
of spatial data in environmental analysis, much of the previous work has
been in data management and cartographic applications, rather than
spatial analysis and modelling. The specific needs of the GCRP require
that better .analytical tools be added to the list of current GIS
capabilities, including but not limited to spatial statistical techniques,
spatial and statistical queries, surface generation tools, and spatial
modelling in the GIS domain. A second general area of development is
system integration and/or filter development. Because of the complex
problems facing researchers in the global arena, researchers must be able
to draw on a variety of tools and software rather than be constrained by
individual vendors and software packages. This emphasis on flexibility
will aid in the analysis on climate change effects by focusing on the
application while reducing software limitations.
- 23
-------
References
ACE: 1988, GRASS Users and Programmers Manual, US Army Corps of
Engineers Construction Engineering Research Laboratory, Champaign, IL.
Allen, J.H.: 1988, The World Data Center System, International Data
Exchange and Global Change', in H. Mounsey and R. Tomlinson (eds.),
Building Databases for Global Science, Taylor and Francis, Philadelphia,
pp. 138-152.
Aronoff, S.: 1989, Geographic Information Systems: A Management
Perspective, WDL Publications, Ottawa.
Bailey, R.G.: 1988, 'Problems with using Overlay Mapping for Planning and
their Implications for Geographic Information Systems', Environmental
Management 12, 11-17.
Bickmore, D.P.: 1988, 'World Digital Database for Environmental Sciences
(WDDES)', in H. Mounsey and R. Tomlinson (eds.), Building Databases for
Global Science, Taylor and Francis, Philadelphia, pp. 181-191.
Burrough, P.A.: 1986, Principals of Geographical Information Systems for
Land Resources Assessment, Clarendon Press, Oxford.
Campbell, W.G., Church, M.R., Bishop G.D., Mortenson, D.C., and Pierson, S.M.:
1989, The Role for a Geographical Information System in a Large
Environmental Project', International Journal of Geographical Information
Systems 3, 349-362.
Campbell, W.G., and Mortenson, D.C.: 1989, 'Ensuring the Quality of
Geographic Information System Data: A Practical Application of Quality
Control', Photogrammetric Engineering and Remote Sensing 55, 1613-
1618.
CES: 1989, Our Changing Planet: The FY 1990 Research Plan. The U.S. Global
Change Research Program, Committee on Earth Sciences, Executive Office
of the President, Washington, DC.
- 24
-------
CIA: 1990, World Data Bank - II, Central Intelligence Agency, National
Technical Information Service, Washington, DC.
Chrisman, N.R.: 1984, The Role of Quality Information in the Long-term
Functioning of a Geographic Information System', Cartographica 21, 79-
87.
Clark, D.M., and Kineman, J.J.: 1988, 'Global Databases: A NOAA Experience',
in H. Mounsey and R. Tomlinson (eds.), Building Databases for Global
Science, Taylor and Francis, Philadelphia, pp. 216-232.
Dickinson, R.E.: 1989, 'Uncertainties of Estimates of Climate Change',
Climatic Change 15, 5-13.
Dolph, J.E.: 1990, 'Characterizing the Distribution of Precipitation and
Runoff over the Continental United States using Historical Data', Journal
of Geophysical Research (submitted).
ESRI: 1984, 'Map of Desertification Hazards', in Geographic Information
Systems for Resource Management: A Compendium,. American Society for
Photogrammetry and Remote Sensing, Falls Church, VA.
GSFC: 1979, Skin Surface Temperature (SST) Data from HIRS2/MSU for
1979, Goddard Space Flight Center, Greenbelt, MD.
Goodchild, M.F.: 1988, The Issue of Accuracy in Global Databases', in H.
Mounsey and R. Tomlinson (eds.), Building Databases for Global Science,
Taylor and Francis, Philadelphia, pp. 31-48.
IGU: 1988, Building Databases for Global Science, H. Mounsey and R.
Tomlinson (eds.), Taylor and Francis, Philadelphia.
ISSS: 1986, World Soils and Terrain Digital Database at a Scale 1:1M,
International Society of Soil Scientists,. Wageningen, The Netherlands.
Kineman, J.J.: 1989a, Improved USNAVY/NCAR Global 10 Minute Terrain
Dataset in Image Format, National Geophysical Data Center, Boulder, CO.
- 25
-------
Kineman, J.J.: 1989b, NOAA Monthly Vegetation Imagery on a 10 Minute
Grid, National Geophysical Data Center, Boulder, CO. .
Kineman, J.J., Clark, D.M., and Croze, H.: 1990, 'Data Integration and
Modelling for Global Change: An International Experiment', in Proceedings
of the International Conference and Workshop on Global Natural Resource
Monitoring and Assessments: Preparing for the 21st Century, Venice, Italy.
Legates, D.R., and Willmott, C.J.: 1989, Global Surface Air Temperature and
Precipitation, Long-period Grid Means, National Center for Atmospheric
Research, Boulder, CO.
Lozar, R.C.: 1990, 'CERL's Global Modelling Capability', presented at the
North American NOAA Polar Orbiter User's Meeting, Washington, DC.
Marks, D.: 1990, The Sensitivity of Potential Evapotranspiration to
Climate Change over the Continental United States', Journal of Geophysical
Research (submitted).
Matthews, E.: 1983, 'Global Vegetation and Land Use: New High Resolution
Data Bases for Climate Studies', Journal of Climatology and Applied
Meteorology 22, 474-487.
NOAA: 1990, Global Vegetation Index User's Guide, ed. K.B. Kidwell,
National Oceanic and Atmospheric Administration, Washington, DC.
Olson, J.S., Watts, J.A., and Allison, L.J.: 1983, 'Carbon in Live Vegetation
of Major World Ecosystems', Environmental Sciences Division Publication
No. 1997, Oak Ridge National Laboratory, Oak Ridge, TN.
Phillips, D.L., Dolph, J.E., and Marks, D.: 1990, 'Evaluation of Geostatistical
Procedures for Spatial Analysis of Precipitation', Water Resources
Research (submitted).
Quinlan, F.T., Karl, T.R., and Williams, C.N.: 1987, United States Historical
Climatology Network Serial Temperature and Precipitation Data, US
- 26
-------
Department of Energy, Carbon Dioxide Information Analysis Center, Oak
Ridge, TN.
Rizzo, B.: 1988, The Sensitivity of Canada's Ecosystems to Climatic
Change', Newsletter of the Canada Committee on Ecological Land
Classification 17, 10-12.
Schneider, S.H.: 1989, The Greenhouse Effect: Science and Policy1, Science
247, 771-781.
Staub, B., and Rosenzweig, C.: 1987, Global Digital Data Sets of Soil Type,
Soil Texture, Surface Slope, and other Properties, National Center for
Atmospheric Research, Boulder, CO.
Tomlinson, R.: 1988, 'Opening Remarks', in H. Mounsey and R. Tomlinson
(eds.), Building Databases for Global Science, Taylor and Francis,
Philadelphia, pp. 1-9.
UNEP: 1990, Monthly Temperature and Precipitation Anomalies for 1987
through 1988, United Nations Environment Programme, Global Resource
Information Database, Geneva, Switzerland.
UNESCO: 1973, UNESCO/FAO Soil Map of the World. United Nations
Environment Programme, Global Resource Information Database, Geneva,
Switzerland.
USGS: 1978, The Nations Water Resources 1975-2000, US Geological
Survey, Washington, DC.
Wallace, J.R., Lettenmaier, D.P., and Wood, E.F.: 1990, 'A Daily
Hydroclimatological Data Set for the Continental United States',Water
Resources Research (submitted).
Wilson, M.F., and Henderson-Sellers, A.: 1985, 'A Global Archive of Land
Cover and Soils Data for use in General Circulation Models', Journal of
Climatology 5, 119-143.
- 27
-------
A Continental-Scale Simulation of Potential Evapotranspiration
for Historical and Projected Doubled-CO2 Climate Conditions
DANNY MARKS
NSI Technology Services, Inc.
U.S. Environmental Protection Agency,
Environmental Research Laboratory
200 S.W. 35th St.,
Corvallis, OR 97330,
(503) 757-4657;
III-l
-------
TABLE OF CONTENTS
2. TURBULENT TRANSFER MODEL 7
3. GEOGRAPHIC DATA SURFACES FOR CURRENT CLIMATE CONDITIONS 9
4. GEOGRAPHIC DATA SURFACES FOR PREDICTED 2xCC>2 CONDITIONS 19
5. POTENTIAL EVAPOTRANSPIRATION (ETJ>) OVER THE U.S 28
6. DISCUSSION 36
7. CONCLUSIONS 38
III-2
-------
LIST OF FIGURES
1. Shaded relief map of the U.S. [[[ 11
2.b Location of HCN data measurement sites. [[[ ...................... 12
2.b Locations of U.S. airport measurement sites 12
3. Monthly air temperature Ta surfaces, historical climate data 14
4. Monthly vapor pressure ea surfaces, historical climate data 16
5. Seasonal wind speed u\Qm surfaces, historical climate data 18
6. Synopsis of historical climate data and GCM-predicted 2xCC>2 Ta, ea, and u i om 20
7. Monthly air temperature Ta surfaces, GFDL model, 2xCC>2 22
8. Monthly air temperature Ta surfaces, GISS model, 2xCC>2 23
9. Monthly vapor pressure ea surfaces, GFDL model, 2xC02 24
10. Monthly vapor pressure ea surfaces, GISS model, 2xCC>2 25
11. Seasonal wind speed M !(>„ surfaces, GFDL model, 2xCC>2 26
12. Seasonal wind speed u\Qm surfaces, GISS model, 2xCC>2 27
13. Calculated potential evapotranspiration £T,P> historical data 29
14. Thornthwaite estimate of Eftp, average July conditions, historical climate data 30
15. Calculated potential evapotranspiration ETJ> for the GFDL model, 2xCC>2 32
16. Calculated potential evapotranspiration EJJ> for the GISS model, 2xCO2 33
-------
A Continental-Scale Simulation of Potential Evapotranspiration
for Historical and Projected Doubled CC>2 Climate Conditions
DANNY MARKS
NSI Technology Services, Inc.
U.S. Environmental Protection Agency,
Environmental Research Laboratory
200 S.W. 35th St.,
Corvallis, OR 97330,
(503) 757-4657;
ABSTRACT
Potential evapotranspiration ETJ> was calculated over the continental U.S. for current climate conditions
and for predicted conditions for a doubled CC>2 climate, using a turbulent transfer model presented by
Marks [1988]. The simulation was done over a grid of 138,650 geographically referenced points
representing a digital elevation model (DEM) of the U.S. at 10km spacing. Data for air temperature Ta,
vapor pressure ea, and wind speed at 10 m above the surface u, were corrected for topographic effects, and
used to generate surfaces at the same 10km resolution. These surfaces represent a long-term average
monthly time-series for the period between 1948-88. Relative changes in these parameters were estimated
from 2xC02 scenarios from the GFDL and GISS GCMs. E^p surfaces were calculated for both current
and 2xCC>2 conditions.
The ET^P surface for current conditions shows evaporative stress maximized in the western U.S. in sum-
mer. This is significantly different from EJJ> estimated from a Thornthwaite [1948] temperature-
regression method which incorrectly shows evaporative stress maximized in the southeastern U.S. Both
GCMs predict increases in wind, temperature, and humidity. The GFDL model predicts drier, windier
conditions than the GISS model, and shows E^p as very high in summer in the mid-western U.S. The
GISS model predicts moister but less windy conditions, and also shows a modest increase in E^p in the
mid-western U.S. This analysis illustrates the importance of physically based estimates of evaporative
stress for climate change analysis. It also shows that it is possible to evaluate the sensitivity of surface
processes, such as ETJ at resolutions which are ecologically meaningful.
III-4
-------
1. INTRODUCTION
Projected changes in the global climate driven by increases in atmospheric CO2, indicate a
significant change may occur in climatic conditions during the next 50 to 100 years [Keeling, 1973;
Manabe and Wetherald, 1975, 1980; Keeling and Bacastow, 1977; Thompson and Schneider, 1982;
Ramanathan, el al., 1985; Dickinson and Cicerone, 1986; Broecker, 1987; Hansen, et al., 1988]. This
would affect the terrestrial biosphere through changes in the regional energy balance [Dickinson, 1983]
which would alter the regional water cycle [Strain, 1985; Eagleson, 1978, 1982, 1986; Lettenmaier and
Surges, 1978; Lettenmaier and Can, 1990; Lettenmaier and Sheer, 1991; Smith and Tirpak, 1989], and
have a profound effect on vegetation distribution [Holdridge, 1947; Mather and Yoshioka, 1968; Botkin, el
al., 1972; Shugart and West, 1977; Solomon, 1986] and condition [Perrier, 1982; Gates, 1983; Eagleson
and Segarra, 1985]. Feedbacks between vegetation and climate, described in detail by Hansen, et al.
[1984], Dickinson and Hanson [1984], and Rind [1984], would influence both the magnitude and timing of
climate change by altering the surface albedo and radiation balance, soil moisture storage, evapotranspira-
tion, and the water balance. Evapotranspiration, ET, is the link between terrestrial hydrology, vegetation,
and the climate system.
ET is the combination of direct evaporation from the surface and transpiration from vegetation.
They are combined because, for practical purposes, it is difficult to separate them in a natural environment.
Direct evaporation from the soil is usually only a small percentage of total water loss [Kramer, 1983],
while ET has been estimated to be 50% or more of the annual precipitation at continental to global scales
[Budyko, 1974; Baumgartner andReichel, 1975; Korzoun, et al., 1977; Brutsaert, 1986]. These studies of
large-scale evapotranspiration have estimated ET from regression relationships between air temperatures
and measured evaporative loss [see, for example Thornthwaite, 1948; Holdridge, 1967; Mather and
Yoshioka, 1968; Budyko, 1974; Whittaker, 1975; Eagleman, 1976; Woodward, 1987; Stephenson, 1989].
The limitations of this approach are pointed out by Larcher [1980], who notes that plants respond to solar
radiation, humidity, temperature, wind, and soil moisture. Any of these can provide the stress to close the
stomata and eliminate or reduce transpiration. An estimate of evaporative stress based solely on air tem-
perature assumes that air temperature is functionally related to all of these other parameters, and is thus an
effective surrogate for their interaction. Over a heterogeneous region, there are a multitude of combina-
tions of humidity and wind which can occur at a given temperature. Under these conditions, evaporative
stress is not adequately specified by air temperature.
This experiment is aimed at regional to continental scale analysis of evaporative stress. In this con-
text, a region in the U.S. is on the order of one of the U.S.G.S. Hydrologic Regions, such as the Great
Basin, the Upper Colorado, or the Missouri. This is an area the size of several states, or about a million
km2. At these regional scales, ET has a strong influence on precipitation. Shukla and Mintz [1982] state
that evaporation will influence local precipitation, but that this effect will vary regionally. Early estimates
of the portion of regional precipitation derived directly from regional ET were as little as 10-15% [e.g.:
Benton, et al., 1950; Budyko andDrozdov, 1953] butLettau, et al., [1979] reported as much as 71% of the
precipitation in the Amazon basin is from local ET, and Salati and Vose, [1984] estimated 48% of Amazon
III-5
-------
basin ET becomes local precipitation. An improved understanding of the current distribution and magni-
tude of ET is needed to evaluate how its distribution and magnitude might be altered under projected
2xCC>2 climates.
Potential evapotranspiration ETJ is the atmospheric demand for water. Eftp = ET over water, snow,
or a moist vegetated surface; in most natural environments ET is less than ET,P. Thus EJJ> can be used to
estimate water stress or drought by determining the precipitation (P) excess or deficit, from the evapora-
tive demand: P~ET,P.
Rind, el ai, [1990] used a variation of the Palmer drought severity index (PDSF) [Palmer, 1965],
called the supply and demand drought index (SDDf), with climatic conditions from the GISS GCM to
show how drought conditions are likely to increase during the next 50 years. Their analysis was carefully
done, but at the very coarse resolution of the GISS GCM (7.83°xlO°); it is very difficult to evaluate how
increased ETJ will affect regional ecosystems, all of which occur at sub-grid resolution.
In this study, the distribution of ETJ> is evaluated over the continental U.S. at a spatial resolution of
10 km, which accounts for the effects of topography at continental or regional scales, and for regions or
ecosystems as small as 100km2. An aerodynamic turbulent transfer model, presented by Marks [1988]
and Marks and Dozier [submitted] to calculate energy and mass flux from a snow cover was modified to
accept geographically distributed input data to calculate ETJ> over large regions. Historical data from a 30
to 40 year period between 1948 to 1988 for the U.S. were used for the analysis. Air temperature Ta, vapor
pressure ea, and wind speed u were aggregated into long-term monthly averages, corrected for elevation
effects using methods detailed below, and interpolated to a 10 km grid over the continental U.S. These dis-
tribution surfaces are used as input to the turbulent transfer model to calculate a monthly time-series of
ET,P for current, or lxC02 conditions.
Ta, ea, and u data from two GCMs (GISS and GFDL) t for both Ix- and 2xCC>2 runs of are con-
verted to change ratios Atf :
"-SOOT ">
The A/? values are interpolated, using an inverse distance squared method [Isaaks and Srivastava, 1989]
to the same 10 km grid to form a change distribution surface. These are used to modify the Ta,ea, and «
distribution surfaces generated from the historical data to approximate the 2xCC«2 distribution surface, and
to calculate a predicted ET,P distribution surface time-series for 2xCC>2 for both GCM predicted condi-
tions. This experiment shows how GCM scenarios can be better integrated into known distributions of
properties and processes at spatial resolutions which are fine enough to provide ecologically significant
f: Data from the GISS and GFDL GCMs were selected because they included air temperature, wind and humidity, and
among the GCM scenarios from the group developed for the EPA report The Potential Effects of Global Climate Change
on the United States [Smith and Tirpak, 1989] they generally represent the most extreme range of 2xCC>2 conditions
predicted for these parameters over the U.S.
III-6
-------
information. It also shows the importance of utilizing GCM predicted changes in humidity and wind, as
well as air temperature, to evaluate the effects of climate change on biophysical processes such as ETJ>.
The use of the physically-based turbulent transfer model to predict ETJ> at this resolution, over the U.S.,
illustrates the importance of utilizing information about humidity gradients and wind to estimate evapora-
tive sensitivity to changing climate forcings.
2. TURBULENT TRANSFER MODEL
The energy balance of the surface is expressed as
AQ=Rn+H +LVE + G +M (2)
where A<2 is change in surface energy, and Rn, H, LVE, G, and M are net radiative, sensible, latent, conduc-
tive, and advective energy fluxes, respectively. In temperature equilibrium, AQ = 0; a negative energy
balance will cool the surface, decreasing it's temperature, while a positive energy balance will warm the
surface, t
In most terrestrial environments, G and M are relatively small. Turbulent energy and mass flux at
the Earth's surface is second only to radiation in importance for the energy balance of terrestrial ecosys-
tems. Over land, net radiation /?„ is of approximately equal magnitude to sensible H and latent heat
exchange, LVE [Budyko, 1974; Baumgartner andReichel, 1975; Korzun, et at., 1978]. However, because
LVE is usually negative (a heat loss), and H is usually positive (a heat gain), the sum of H + LVE is gen-
erally much smaller than Rn when integrated over a day or longer [Marks andDozier, submitted].
The turbulent transfer of momentum, heat, and water vapor described by H and LVE are the most
complicated forms of energy exchange, and are not easily measured in a natural environment. The data
required to calculate them are difficult to measure at a point, and they have a highly variable distribution
over a topographic surface. Not only does significant energy transfer occur by turbulent exchange, but in
most environments significant water loss can occur from sublimation, direct evaporation, or transpiration
by vegetative cover [Budyko, 1974; Baumgartner andReichel, 1975; Korzun, et al., 1978; Beaty, 1975;
Stewart, 1982; Davis et al., 1984].
Tractable approaches to calculating sensible and latent fluxes have been summarized [Fleagle and
Businger, 1980; Brutsaert, 1982]. The method used for this experiment was adapted from Brutsaert [1982]
by Marks [1988] and Marks and Dozier [submitted]. A similar approach was used by Martin, et al. [1990]
to estimate the sensitivity of ET to climate change over several different types of vegetation in North
America.
The data required for calculating turbulent transfer at each grid point are air density p, air and sur-
face potential temperatures 0a, &s, air and surface specific humidities qa, qs, wind speed u, and surface
roughness ZQ. The methodology for estimating these parameters over the U.S. will be discussed in a later
f: The sign convention used in this analysis is that a flux of energy or mass is negative if it is away from the surface, and
positive if it is toward the surface.
III-7
-------
section.
The equations to be solved at each grid point are:
Obukhov stability length:
M*3 P
L =
H
TaCp
Friction velocity:
«* =
+ 0.61E
uk
(3)
In
zu-dQ
zo
V.rm
zu
L
(4)
Sensible heat flux (positive toward the surface):
(8a - 6,) aHku*p Cp
H —
In
ZT~ do
zo
-V**
zr
(5)
Mass flux (positive toward the surface):
(a^ ft) as k M* p
E =
In
Zq -dp
ZQ
~V*w
12.
L
(6)
The latent heat flux is Lv x E, or LVE, where Lv is the latent head of vaporization (=2.5xl06 J kg l),
or subh'mation (=2.8xl06 Jkg"1). Cp is the specific heat of dry air (1005 Jkg"1 K"1). Measurement
heights are zu for wind, 27- for air temperature, and z? for humidity. A measurement height of 10 m was
used for all three in this experiment a// and OE are the ratio of eddy difiusivity and viscosity for heat and
water vapor, respectively; while there is some uncertainty associated with the value of these ratios, Brut-
saert [1982] suggests that for most natural surfaces, a// = ag = 1.0. k is von Karman's constant (dimen-
sionless), k = 0.40. g is the acceleration of gravity (9.80616m s~2). do is the zero-plane displacement
height; Brutsaert [1982] also suggests do = (2/3) 7.35 ZQ. ZQ is the surface roughness length (m). For
fairly smooth surfaces, ZQ ranges from 0.0001 to 0.005 m, though vegetation cover and terrain features can
constrain the value to be much higher.
The Y stability functions, v|/jm for mass, \|/s/, for heat, and \\rsv for water vapor, are:
Stable (C = 7- > 0):
= 5
(7)
1II-8
-------
Unstable (C =
-------
U.S. Army Corps of Engineers as a Beta release [USAGE, 1988].
Digital Elevation Grid
The base-layer of the GIS system used in this analysis is a digital elevation model (DEM) of the con-
tinental U.S. sampled at a 5 ' latitude and longitude spacing. It was acquired from the NOAA National
Geophysical Data Center in Boulder Colorado on CD-ROM as part of the data distribution Geopyhysics of
North America [NOAA, 1989]. This database was re-projected into an Albers Equal-Area Conic projection
[Snyder and Voxland, 1989] at a uniform 10km grid spacing. This is a grid of 295 rows by 470 columns,
with 138,650 grid points. When the boundary of the U.S. is used to mask the file, 77,857 grid points are
within the area used for analysis. The minimum elevation is -121 m located in Death Valley, California,
and the maximum elevation is 3790m, located in the Sierra Nevada, California. Figure 1 is a shaded
relief map of this data layer, showing the distribution of relief and topography across the U.S.
Elevation-Corrected Air Temperature Surfaces
Monthly average air temperatures were extracted from the Historical Climatology Network (HCN)
database for the years 1948-87 [Quinlan, et al., 1987; Karl, et ai, 1990]. Data from 1211 stations within
the continental U.S. were used to construct long-term monthly averages for this 40 year period. Figure 2.a
shows the distribution of measurement sites across the U.S. The spatial distribution is adequate across
much of the U.S., except for Nevada, the southwest deserts, Texas, and the Dakotas. In general, however,
high elevation areas particularly in the western U.S., are not well represented by these data. Regional vari-
ation in air temperature is largely controlled by variations in air pressure caused by topographic structure.
From the equation of state for an ideal gas and the hydrostatic equation [Byers, 1974], we can show how
air temperature changes with elevation and pressure:
C0 dTa = -
RT,
a
gpdz
mP
This reduces to:
efr°=-\ih\dz
[LP\
where Ta is the air temperature (K), P is the air pressure in Pascals (Pa), p is the air density (kgm"1), z is
the elevation (m), R is the gas constant (8.3143Jmor' K"1), m is the molecular weight of dry air
(28.9644 kg m~3), g is the acceleration due to gravity (9.80616 m s~2 ) and Cp is the specific heat of dry air
at constant pressure (1005 J kg-1 KT1).
Ill-10
-------
FIGURE 1: Shaded relief map of Hie U.S. computed from l()kiu};ri(l DRM (77,857 |X)ints).
-------
FIGURE 2.a: Location of Historical Climatology Network (HCN) data measurement sites.
FIGURE 2.b: Locations of U.S. airport measurement sites. Climate Data for Planet Earth database.
111-12
-------
Equation 12 defines the adiabatic temperature change with height in the atmosphere. The decrease in tem-
perature at higher elevations is caused by a combination of adiabatic cooling, and local or regional surface
interaction with the atmosphere. To extend air temperature data from the irregular measurement network
to a uniform grid across the U.S. we must systematically account for topographic effects in the interpola-
tion procedure.
To achieve this, topographic effects were removed from the measured air temperatures by converting
them to their sea-level equivalent prior to interpolation, and then the sea-level air temperatures interpo-j
lated to the 10 km grid were re-converted to the appropriate air temperature for the elevation from the
DEM. Air Temperatures Ta were converted to their sea-level equivalent, or potential temperatures @a:
ea = Ta[-^-] 13
*z
where PQ and Pz are air pressures in Pa at sea level, and elevation z, respectively. The HCN measurement
site elevations were used to derive the air pressures using the hydrostatic equation (equation 11, above).
This procedure effectively eliminates the elevation effects on the measured temperatures, but retains
latitude/longitude effects by scaling air temperatures to their equivalent at the 100 kPa level (approxi-
mately sea level), making them directly comparable. They were then interpolated to the geographic grid
spacing of the DEM data using a simple linear inverse distance squared algorithm \Jsaaks and Srivastava,
1989] t
resulting in a geographic surface of 0a for the U.S. Eq. 13 was then inverted using the 6a surface and the
DEM elevation surface to map Ta back onto the elevations of the DEM grid.
Figure 3 shows the long-term average monthly time series of elevation corrected Ta for the U.S.
Though the elevation correction was fairly simple, it gives a result which is quite different from other
attempts to distribute air temperature over large regions [e.g.: Legates and Willmott 1990]. These efforts
have been able to show the general spatial variation of air temperature, and some latitudinal trends, but
have made no attempt to correct for topographic effects. The temperature surfaces used in this analysis
represent the first large regional-scale temperature database which is elevation-corrected.
f: The inverse distance squared interpolation is:
where VGP is the interpolated value at a grid point, v; is the measured value of the i* nearest neighbor, d; is the distance to
the i* nearest neighbor, and n is the number of nearest neighbors considered. While the interpolation is setup to search up to 12
nearest neighbors, there is little impact on the result beyond 4 to 6, unless they are tightly spaced so that the distances are all similar.
This algorithm was selected over other more complex methods such as spline interpolation [Eubank, 1989] because it is numerically
simple, computationally relatively fast, and is an exact interpolation method (the interpolated surfaces are constrained to go through
the input data points). Other methods will be investigated in future experiments
111-13
-------
s
FIGURE 3: Monthly air temperature Ta surface under historical climate conditions.
-------
Persistent local features such as valley inversions, are not accounted for, but the temperature sur-
faces show significant detail in the monthly average temperatures. Both topographic and latitudinal tem-
perature features are illustrated. A surprising number of relatively small regional features such as the
Columbia Gorge in the Pacific Northwest, the Sierra Nevada, White Mountains, and the Owens Valley in
eastern California, the Grand Canyon, and the southern Appalachians show clear differences in tempera-
ture at various times of year.
Elevation Corrected Humidity Surfaces
Daily dew point temperatures were extracted from National Climatological Data Center Archive
(NCDC) data supplied on CD-ROM by WeatherDisc Associates, Inc. [WeatherDisc Assoc., Inc. 1990].
These data are described in detail by Spongier andJenne [1988]. Dew point temperatures from the World-
wide Airfield Summaries database for 1565 airport measurement sites in the continental U.S. were aggre-
gated into a long-term average monthly time series for 1960-87. Figure 2.b shows the distribution of air-
port measurement sites across the U.S. As with air temperature, in most regions the distribution is ade-
quate, except for Nevada, the southwestern deserts, Texas, and the Dakotas, and high elevation sites are
severely under represented. Topographic correction for dew point temperature is more uncertain than for
air temperature. Dew point temperature not only diminishes with elevation, but it is constrained to be
equal to or less than the local air temperature. Dew point temperatures were converted to vapor pressures
ea (Pa), and then combined with the appropriate Ta from the temperature surfaces to calculate
temperature-corrected relative humidity RH:
RH = -^— 14
&a,sat
where eatSat is the saturation vapor pressure at Ta. Ta from the temperature surfaces was used in this cal-
culation because Ta derived from the airport database was frequently inconsistent with the HCN derived
temperature surfaces. This was due to differences in observation time and the relatively coarse resolution
of the 10km grid when locating measurement stations. Air temperatures measured at airports also tend to
be biased because they are usually measured over tarmac.
Temperature-corrected RH's were then interpolated to the geographic grid spacing of the DEM data
and the Ta surfaces, using the same inverse distance squared method [Isaaks and Srivastava, 1989], used
for air temperatures, resulting in a geographic surface of RH for the U.S. RH was used for the interpola-
tion, rather than vapor pressure ea because RH tends to be more stable over mountainous regions than ea
[Marks and Dozier, 1979]. Eq. 14 was then inverted, using the Ta and RH surfaces to map ea onto the
DEM grid.
Figure 4 shows the long-term average monthly time series of elevation corrected ea for the U.S.
While this correction is not as physically-based as the Ta correction, the spatial distribution of humidity is
much smoother than that of temperature, and the elevation-corrected ea surfaces represent the only humi-
dity database of this type. The values ea are generally low during winter across the U.S., with higher
humidities during the summer.
111-15
-------
I
5
Pa
0-500
500-750
750-1000
1000-1250
1250-1500
1500-2000
2000-2500
2500-3000
3000-3500
3500-5000
FIGURE 4: Monthly vapor pressure ea surfaces under historical climate conditions.
-------
The Southeast is consistently humid, with very high humidities during the summer. High humidities occur
along the eastern seaboard during summer, but are absent during colder months. Many of the same
regional features that were evident on the Ta surfaces are also evident on the ea surfaces. High elevations
are consistently drier, throughout the year. The "summer monsoon" in the Southwest is also evident from
the incursion of higher humidities during July and August
Wind Surfaces for the Continental U.S.
Synoptic wind data for the continental U.S. are nearly non-existent at space and time scales required
for this analysis. National Meteorological Center (NMC) Grid Point Data are available that provide twice
daily estimates of wind speed at two or three pressure heights (850, 500, and 250 mb) over a 10 minute
(= 20km) grid which covers the U.S. [e.g.: Jenne, 1970 Jenne, 1990]. While it may be possible to derive
wind surfaces from these data, the data were not available in a usable format for testing at the time of this
experiment
The Department of Energy had developed a wind database described in Wind Energy Resource Atlas
[Elliott, et al., 1986]. These data are available in digital form for seasonal averages for the continental
U.S. The averaging periods of the data are variable, and unspecified, as the database was designed for
locating wind energy systems, not hydro-meteorological analysis. As Elliott, et al. [1986] explains, they
are based on a combination of surface measurements, and upper air data such as radiosonde and the NMC
data described above. They have been topographically corrected to account for higher wind speeds over
ridges and mountains. The data were resampled from a 1/3x1/4 ° latitude-longitude grid to the same 10km
DEM grid as the Ta and ea surfaces, using the same inverse distance square method. No attempt was made
to account for the greater topographic detail in the DEM grid.
Figure 5 shows the long-term seasonal time series of wind speed at 10m (uiOm) for the U.S. The
data surfaces show the expected increases in wind speed in areas of high topographic relief. Higher wind
speeds occur in winter, and lower during summer. The data used to develop the wind surfaces were com-
piled to estimate wind power potential over the U.S. and are based on conservative estimates of wind
speed [Elliott, et al., 1986]. In many areas, actual monthly wind speeds are probably higher. While these
wind surfaces are imperfect, they are the best synoptic wind data available at this time, and are adequate
for this analysis.
111-17
-------
X
FIGURE 5: Seasonal wind speed UK^ surfaces, under historical climate conditions (DOE wind database).
-------
4. GEOGRAPHIC DATA SURFACES FOR PREDICTED 2xCO2 CONDITIONS
Data from the NASA Goddard Institute of Space Studies (GISS) and the NOAA Geophysical Fluid
Dynamics Laboratory (GFDL) GCMs were extracted from the doubled CO2 climate scenario data assem-
bled for the EPA assessment of potential climate change effects [Smith and Tirpak, 1989]. These were
selected because they included wind and humidity, as well as temperature and because they represent the
most extreme range of 2timeCO2 conditions predicted for these parameters over the U.S. A detailed
description of the GCM data is given by Jenne [1990]. The GFDL resolution is 4.5x7.5°, and the GISS
resolution is 8 x 10° latitude-longitude. Values for air temperature, humidity, and wind for present condi-
tions (lxCO2), and for 2xCO2 were used to generate ratios of the relative change in these parameters A/?,
as shown in Eq. 1. This approach to using GCM scenario data was presented by Parry, et al. [1987] and
Parry and Carter [1989], and was used for precipitation analysis in the EPA report to congress [Smith and
Tirpak, 1989]. The ratio approach is used because the actual values predicted by the lxCO2 GCM runs
are inconsistent with historical data distributions in both time and space. This is caused by the coarse reso-
lution of the GCMs, the lack of a realistic topography which influences parameters such as temperature,
wind, humidity, and precipitation, and the relatively crude parameterization of surface properties within
GCMs [Rosenzweig and Dickinson, 1986; Schlesinger, 1988].
Developing ratios for wind and humidity, which are based on absolute scales, is straightforward.
Air temperature in °C must be first transformed to K. When temperature is transformed to K, the change
ratios AR are generally between 1.01 and 1.02, and are not very different from predicted absolute tempera-
ture changes. Because of this, the absolute temperature change Ara (K) is used in place of A/? for the
estimating GCM-predicted Ta, to avoid rounding errors.
Values of A/? for humidity and wind, and &Ta for air temperature, were interpolated from the GCM
grids to the 10 km DEM grid using the inverse distance squared algorithm discussed above [Isaaks and
Srivastava, 1989]. A/? surfaces for humidity and wind are multiplied by the ea and u\Qm surfaces
described in the previous section. The ATa surfaces are added to the Ta surfaces, to generate Ta, ea, and
u iQm surfaces for the predicted 2xC02 conditions for the GFDL and GISS GCMs.
Figure 6 summarizes the temperature, humidity, and wind data surfaces for current conditions as
predicted by historical data, and for 2xCO2 conditions, as predicted by the GISS and GFDL GCMs. In
general, both GCMs predict increases in all three meteorologic parameters. There is little difference
between GFDL and GISS predicted Ta and ea except during June, July, August, and September. Over this
interval, the GFDL model predicts higher temperatures, and lower humidities than the GISS model. Com-
pared to the GFDL model, wind speeds u \Qm, are higher for the GISS model in winter and fall, about the
same in the spring, and then substantially lower in summer. Though this temporal trend is not as great in
the base-case historical data, it is generally similar.
111-19
-------
FIGURE 6: Synopsis of historical climate data and GCM-prcdicted 2xCO2 Ta,ea, and
based on mean values for each parameter, over U.S. DEM grid.
. Comparison
t?
Ill
LLJ
cr
3
H-
<
cr
UJ
Q.
2
LU
OO
30
25
20
15
10
=
1 1 1 1 - 1 1 1 >
^-v^_
/T"~T~Y^V
^? • "**" ^\ \.
«*^ / • ^r
/T/* \ \
xT /* *\T\"
^ / ^ -
T "^ ^ * ^ \.
n t- -• *
FMAMJJASOND
MONTH
0
Q.
UJ
cr
ZJ
to
C/l
1. 1
cr
Q.
cr
o
Q.
<
^C3UU
2000
1500
1000
•
500 jrj—
0
!
J F
9
^^s
n
I3'
"a 7.
LU
UJ
a 6
T' ' —
_
10 • • .-
§ 5
=e
^
-
WINTER
, I , . . i :
T-—T
^f ^^ ^7 ~' ^^ ^\
.^-^"" N. _
* + *\v\
. ™ X" 0 ^
^ /* •\
^i^ ^ •*. ^f
I'* ^'^
I 1 ! I1'!!'
WAMJJASOND
MONTH
i i
- - - '
__^^- — "~~--^/
\, X"5
\. ./
^^^ .^
^^T— -^
• -^_^^^
1 1
r
7
•
SPRING SUMMER FALL
SEASON
• BASE T GlSS V GFDL !
i
111-20
-------
The GFDL model predicted substantially higher winds in summer. The GFDL surfaces are drier, warmer,
and windier than the GISS surfaces in summer, when the maximum EJJ> occurs. As shown in Eq. 6, the
combination of higher humidities, and lower wind speeds during summer will lead to lower E?j> for the
GISS-predicted conditions, while the lower humidities and higher wind speeds during summer will lead to
much higher predictions ofE^p for the GFDL-predicted conditions.
Figures 7 and 8 show the distribution of Ta for the GFDL and GISS 2xCO2 conditions. These show
that warmer conditions prevail throughout every month of the year in comparison to the historical climate
conditions shown in Figure 3. During the winter months there is little difference between the GISS and
GFDL temperature predictions, but during summer the GFDL predictions are substantially warmer. For
the base-case (Figure 3), air temperatures gradually increase from winter through spring to summer, and
then decrease in the fall. In the GISS prediction (Figure 8), the temperatures are higher, but this same pro-
gression occurs. For the GFDL prediction (Figure 7), winter and fall temperatures basically agree with
the GISS temperatures. However, during the months of June, July and August, the GFDL model predicts
substantially higher temperatures for the entire country.
Figures 9 and 10 show the distribution of ea for the GFDL and GISS 2xCC>2 conditions. More
humid conditions prevail throughout most months of the year in comparison to the historical climate con-
ditions shown in Figure 5. During the winter months there is little difference between the GCM predic-
tions and the historical data predicted base-case. During summer the GFDL model predictions are only
slightly more humid, while the GISS model predicts higher humidities throughout most of the country,
especially the Southeast. This condition persists into the fall.
Predicted wind surfaces are shown in Figures 11 and 12, for the GFDL and GISS 2xCC>2 conditions.
These surfaces show increased wind across the U.S. during all seasons of the year in comparison to the his-
torical climate conditions shown in Figure 6. They also display^ interesting artifacts caused by the coarse
resolution of the GCM computational grids. Wind appears to be somewhat less stable in the GCMs than
are air temperature or humidity, showing a much more_ejrati^progression_from one season to the next as
shown in Figure 6. In particular, the GFDL wind surfaces (Figure 11) show relatively low wind speeds
just west of the Great Lakes region during spring, but extremely high wind speeds in the same region dur-
ing summer. The GISS wind surfaces (Figure 12) show a similar transition in wind speeds in around the
Great Lakes region from summer to fall.
111-21
-------
FIGURE 7: Monthly air temperature Ta surfaces for conditions predicted by the GFDL model for 2xCO2.
-------
1
degrees
• -40 -10
-10 - -5
1-5-0
0 - 5.
5-10
10 - 15
n 15 - 20
20 - 25
25 - 30
30 - 50
FIGURE 8: Monthly air temperature Ta surfaces for condidtions predicted by the GISS model for 2xCO2.
-------
Pa
0-500
500-750
750-1000
I 1000-1250
1250-1500
1500-2000
2000-2500
2500-3000
3000-3500
3500-5000
I
FIGURE 9: Monthly vapor pressure ea surfaces for conditions predicted by the GFDL model for 2xCC>2.
-------
Pa.
0-500
500-750
750-1000
1000-1250
1250-1500
1500-2000
2000-2500
2500-3000
3000-3500
3500-5000
FIGURE 10: Monthly vapor pressure ea surfaces for conditions predicted by the GISS model for 2xCO2.
-------
m/s
20-26
26-35
FIGURE 11: Seasonal wind speed M10m surfaces for conditions predicted by the GFDL model forlxCCV
-------
FIGURE 12: Seasonal wind speed ulftn surfaces for conditions predicted by the GISS model for2xCO2.
-------
5. POTENTIAL EVAPOTRANSPIRATION (ET)P) OVER THE U.S.
The turbulent transfer model presented above was used to calculate evaporation E surfaces for the
historical data predicted base-case and the GCM predicted 2xC02 conditions. For this experiment, evapo-
transpiration Ej was considered equivalent to E. Because potential evapotranspiration Ejtp is being
modeled, the surface is assumed to be saturated, so the surface vapor pressure es = eStSat, the saturation
vapor pressure at the surface temperature Ts; therefore Ej- as calculated by the model is actually ETJ>.
A surface temperature Ts database from which long-term monthly average Ts could be derived was
not available for this analysis. The monthly average surface temperature can be approximated from
monthly average air temperature over most surfaces. A notable exception is over snow or ice, which can-
not exceed 0°C. For this experiment, the average air temperature Ta, adjusted for topographic effects as
discussed in the previous section, was assumed to be equal to the average monthly surface temperature 7^.
If actual surface temperature data were available, it is unlikely that the average monthly Ts would differ
significantly from average monthly Ta, except under the circumstances listed above. Measured, or satellite
derived surface temperature would allow us to test the model at a time-step of days to weeks, rather than
for average monthly conditions.
Vapor pressure at the surface es was derived from the saturation vapor pressure at Ts. The roughness
length ZQ scales the turbulent interaction between the atmosphere and the surface.
For a given set of climatic conditions, a smooth surface such as bare soil or grassland, will have a
small ZQ, which will reduce turbulent interaction and reduce evaporation; a rough surface such as a forest
canopy, will have a larger ZQ, which will increase turbulent interaction and increase evaporation [Brut-
saert, 1982]. Numerous estimates of ZQ over smooth surfaces such as grass, soil, snow, etc., have been
made, [e.g.: Sutton, 1953; Kondo, 1962; Chamberlain, 1966 and Businger et al, 1971] Deacon [1973]
estimated ZQ over mixed vegetation in England, and Fichtl andMcVehil [1970] estimated ZQ over tropical
vegetation and forests in Florida. Few measurements of ZQ exist, however, over mid-latitude forests. Brut-
saert [1982] shows that ZQ can be derived for different vegetation and other porous surfaces from canopy
height, density, and leaf-area-index (LAI). These data, however, were also not available for this experi-
ment.
In the absence of detailed monthly foliar biomass, leaf-area-index (LAI) data, or vegetation height
and density, the surface roughness length ZQ was assigned a constant value related to a grassland or prairie
over the grid. This was selected because it was a conservative value, that should under-estimate E^p in
most instances. ETJ> should be overestimated over snow, water, and some desert regions, but it will
underestimated over forested regions, particularly mountainous regions where winds speeds may be high.
Because the purpose of this experiment is to estimate the distribution of ETJ> over the U.S. within a rea-
sonable range, the uncertainty in the estimate of ZQ should not detract from the results.
Ta, Ts, ea, es, HIO^, and ZQ data surfaces were assembled into a six-band data image. The model
was run at each DEM grid point within the continental U.S., to derive the average monthly ET,P surfaces
from historical average monthly climate conditions as shown in Figure 13.
111-28
-------
E
mm
0-50
50-100
100-150
150-200
200-250
250-300
300-400
400-600
600-1000
1000-2100
y
__J
FIGURE 13: Calculated potential evapotranspiralion ETtP, for historical climate data.
-------
I
FIGURE 14: Thomthwaite estimate of ETf, using the temperature-regression method for average July
conditions, historical climate data.
-------
Winter ETJ> is relatively low throughout the U.S. By March it begins to increase in the southwest
and by late spring, has increased in the southwest, and the inter-mountain west In summer it is highest in
the southwest, but substantially higher throughout the west and the grain-belt in comparison to the east and
the southeast.
This is significantly different from the result obtained when Ej,p is estimated by regressing air tem-
perature against measured evaporation [e.g.: Thornthwaite, 1948]. Figure 14 is a July ETJ> surface
derived using the Thornthwaite [1948] temperature-regression method which has been commonly used to
estimate potential evapotranspiration over large areas [e.g.: Thornthwaite and Mather, 1955; Holdridge,
1967; Mather and Yoshioka, 1968; Budyko, 1974; Whittaker, 1975; Eagleman, 1976; Woodward, 1987;
Stephenson, 1989]. It predicts the high values in the southwest deserts and also in the southeast, and the
along the eastern seaboard. The inconsistencies inherent in predictions of E-f^p based solely on air tem-
perature, without consideration of wind and humidity, are clearly apparent. As the data presented in the
previous section show (and as anyone who has ever been to those regions in July will attest) high humidi-
ties and low wind speed dominate the southeastern U.S. in summer, which reduces the humidity gradient
q - qs, and the friction velocity u*, as shown in Eq. 5.
Figures 15 and 16 show ET during spring and sum-
mer. The GFDL EJJ> surfaces show substantial increases, especially in the southwest and midwest, but the
GISS surfaces show more moderate increases. This is primarily because the GFDL model predicts a
warmer, windier, and drier environment, with large increases in temperature and wind, but only moderate
increases in humidity, as shown in Figure 6. The GISS model predicts a warmer, more humid environ-
ment, with relatively uniform increases in temperature, wind, and humidity.
Figures 17 and 18 show difference surfaces (2xC02 - HistoricalClimate) for ETJ> calculated from
conditions predicted by the GFDL and GISS 2xCC>2 GCM runs and ETJ> calculated for historical climate
conditions. Both models predict increases in spring and summer EJJ> for the 2xC(>2 conditions. The
GFDL model shows increased ET,P in California's Central Valley, while the GISS model shows increased
Eif in southern Florida. Though the GFDL model shows large increases in ETJ> during summer, while the
GISS model show only modest increases, both models show increase in E-pp in the midwestem grain-belt
during spring and summer growing seasons.
111-31
-------
I J
H
mm
0-50
50-100
100-150
150-200
200-250
250-300
300-400
400-600
600-1000
1000-2100
FIGURE 15: Calculated potential evapotranspiration ETJ, for conditions predicted by the GFDL model for
2xC02.
-------
1
n
mm
0-50
50-100
100-150
150-200
200-250
250-300
300-400
400-600
600-1000
1000-2100
FIGURE 16: Calculated potential evapotranspiration ETiP for conditions predicted by the GISS model for
2xCO2.
-------
-100 - -150
-500 - -1500
FIGURE 17: Difference Between 2xC02 and historical climate preditions of ETJ> for the GFDL model.,
-------
FIGURE 18: Difference Between 2xC02 and historical climate predictions for the GISS model.
-------
6. DISCUSSION
The distribution of air and surface temperature, humidity, and wind are strongly affected by topo-
graphic structure [Marks, 1988; Dozier, et al., 1988]. The detail of topographic structure shown in Figure
1 illustrates variability of topography across the continental U.S. The 10km grid resolution for DEM
shows significant topographic detail for continental-scale analysis, while keeping the number of grid
points (=140,000) computationally acceptable. The Ta,ea, and u\om surfaces presented in Figures 3, 4,
and 5 clearly show that even relatively simple corrections for topographic effects show the topographic
influence on the distribution of climatic parameters, and substantially improve our ability to use historic
data to estimate these distributions at ecologically meaningful resolutions.
Wind data over the U.S. are limited. The UK^I surfaces shown in Figure 4 are long-term seasonal
averages derived from unspecified periods during the last 40 years. Monthly wind surfaces would be more
compatible with the Ta and ea surfaces. Hopefully, such data will be available in the near future.
ETJ> calculated from these data surfaces, using the turbulent transfer model show increases into the
southwestern U.S. starting in spring, with maximum Eftp occurring during summer, and dominating the
western U.S. This is contrary to the result when E-j-tp is estimated using a temperature-regression approach
such as Thornthwaite [1948] (Figure 14) which doesn't account for either wind or humidity gradients. At
large scales, where wind and humidity vary considerably, the temperature-regression approach shows a
strong temperature bias, and gives an incorrect result showing E^p maximized in the southeastern U.S.
during July. The relationship between temperature and Ejf is weak, at best, as shown in Eqs. 3,4, and 6.
Clearly, for analysis at large regional to continental scales, where humidity and wind will vary consider-
ably, it is important to use a model of Ejf which is based on the variation of humidity and wind.
Temperature-regression methods were developed primarily for application over agricultural areas,
where the assumption of uniform humidity and wind conditions were reasonable. They were not designed
for application under changing climatic conditions and should not be used for large regional or continental
scale analysis, where these assumptions are unrealistic. In the late 1940's, climatological databases did
not exist, nor did the the technology to process and display them. The analysis presented in this paper
required 109 bytes of data storage, and computer processing capabilities approaching that of a super com-
puter. Even a few years ago, this would have required affiliation with a major computer research facility,
such as NCAR, GFDL, or GISS, where most GCM research has been conducted. The combined power and
storage of networked, desk-top workstations, has made this type of analysis possible at relatively low cost,
without the specialized technical support required for GCM runs.
Figures 7 thru 12 show Ta,ea, and u iQm surfaces for predicted 2xCC>2 conditions for the GFDL and
GISS GCM models. As is shown in Figure 6, Ta and ea differences between the models are slight, but
differences in u\Qm are large. The GFDL model predicts warmer, drier, and much windier conditions dur-
ing summer, while the GISS model predicts warmer, wetter, and less windy conditions during the same
time of year. These differences have a profound effect on the calculated summer EJJ>, as shown in Figures
15 thru 18. ET,P calculated from the GFDL model conditions is extremely high in summer, while GISS
model predicts Eff that is only slightly higher than the base conditions predicted from the historical data.
111-36
-------
Similar results were reported by Martin, et ai, [1990] from their analysis using very similar methods to
predict actual ET on a grassland, a wheat field, and a forest, for current climate conditions, and 2xCC>2
conditions predicted by the GISS and GFDL GCMs. If the GISS scenario were realized, there would be an
impact on vegetation and agriculture in the U.S. If the GFDL scenario were realized, much of the agricul-
tural grain-belt of the mid-western U.S. would be desicated by increased evaporative demand. This could
have profound implications on the economy of the United States, as reported by Smith and Tirpak, [1989].
Until more realistic feedbacks between terrestrial ecosystems and the climate system are built into the
GCMs, it will be impossible to achieve better agreement between the models, or to use GCM predictions
to estimate the likely ecological consequences of climate change.
There is considerable uncertainty in the absolute magnitude of these calculations. There is no way
to verify the accuracy of the Ta, ea, or u\Qm surfaces for either the current or 2xCC>2 conditions. The
elevation-correction methods and interpolation method discussed above maintain the original data points,
however, so only the distribution between points is uncertain. The turbulent transfer model has been
tested [Stewart, 1982] and verified [Marks, 1988] over snow where both the surface temperature and vapor
pressure, and the scaling parameters can be measured or approximated. For continental scale applications
over heterogeneous surfaces that are not well characterized, uncertainty in the scaling parameters, ag, k,
and ZQ, over a large region, could cause uncertainty in the magnitude of ETJ>, as could approximating the
surface temperature from the average monthly air temperature. The basic trends in the spatial distribution
ofETj> should not change, however.
In general, the magnitudes ofE^p shown for the base case in Figure 13 are in agreement with other
estimates over regions of similar size cited below, f If similar methods can be used to estimate the distri-
bution of surface temperature, net radiation, precipitation, and soil moisture over the U.S., then it will be
possible to compute a water balance surface for the U.S. which can be used verify the evaporation calcula-
tions. For purposes of this analysis, I have emphasized discussion of the spatial distribution of E^p shown
in the surfaces, rather than the predicted magnitude.
Potential evapotranspiration EJ-J> has been used to estimate vegetation stress and soil moisture
depletion by many investigators, using the regression of Ta and measured evaporation [Thornthwaite,
1948; Holdridge, 1967; Mather and Yoshioka, 1968; Woodward, 1987; and drought [Palmer, 1965;
Shugart, 1984; Solomon, 1986]. This approach may work when it is limited to a brief time period, and to a
small field or region, where wind and humidity can be assumed constant, but it is not appropriate for appli-
cation over a large region. It is particularly problematic for climate change analysis, where, as shown in
Figure 6, variations in wind and humidity may in sharp contrast to variations in air temperature. Based on
differences in temperature alone, we would not expect EJJ> to be very different for the GISS and GFDL
predictions.
f: This isn't a valid comparison, however, because the other estimates are based-on temperature-regression methods of
calculating Eff.
111-37
-------
7. CONCLUSIONS
This analysis shows the importance of topographic structure in estimating the distribution of
meteorological parameters over large areas. Elevation corrected surfaces of the temporal and spatial dis-
tribution of Ta, ea, and u\$m, determined from historical data for the U.S. are generated at a 10km grid
spacing. Using these surfaces as a base condition, changes predicted by the GISS and GFDL GCMs can
also be mapped over the U.S. The predicted GFDL and GISS surfaces for Ta and ea are representative of
what may be expected to occur under a 2xCO2 climate.
Wind surfaces are more difficult to estimate from historical data, and GCM predictions of wind
appear less stable and reliable than temperature or humidity. Erratic variations in wind from one season to
the next cause uncertainty in the predicted 2xCO2 u \0m surfaces. It is difficult to determine if this is
because of the seasonal aggregation, or because GCM wind predictions are more closely related to the
atmospheric dynamics of the models than to surface conditions. A monthly time-series of wind data would
be desirable for this type of analysis, as would improved GCM wind predictions.
The calculations of Ej-^p by the turbulent transfer model are substantially different from previous
estimates of E-j-^p based on temperature-regression approaches. Base case calculations of ETJ> show max-
imum evaporative stress during summer over the western U.S., in strong contrast to the maximum shown
over the southeastern U.S. by the temperature-regression methods. This discrepancy illustrates the poor
relationship between air temperature and evaporation shown in Eqs. 3,4, and 6. It also points out that a
regression relationship between two physically unrelated parameters is likely to be inconsistent over an
area as large as the U.S. It is essential that regional to continental-scale estimates of ETJ> be based on esti-
mates of wind speed and humidity gradients.
The ET,P surfaces calculated for predicted 2xCC»2 conditions show an increase in evaporative stress
in the western and midwestern U.S. The GFDL model predicts conditions that would severely impact the
agriculture of the midwestern grain-belt of the U.S. To verify these calculations, and improve our
confidence in the magnitudes of the estimated E-j-f, we must 1) improve our ability to estimate the distri-
bution of climatic parameters over large regions, accounting for topographic effects (this is particularly
important in regard to development of a synoptic wind database which is at a monthly time-scale); 2)
include surface temperature Ts, derived from satellite data, to make the estimate ofE^p more realistic, and
to allow evaluation of sensible H and latent LVE as well as ETJ>\ 3) incorporate synoptic vegetation pat-
terns into the turbulent transfer model, to account for the distribution of surface roughness, stomatal resis-
tance, and leaf area index (LAI) [Martin, et al., 1990]; 4) test the model and a daily time-step for selected
years of particular climatic interest, such as the 1988 drought, or selected flood years.
This analysis of ETJ> sensitivity is the first step toward a comprehensive evaluation of the energy
and mass balance over a continent at a spatial resolution that is ecologically and hydrologically meaning-
ful. It will lead to a better understanding of the interactions between climatic conditions and surface
processes, and feedbacks between regional processes and vegetation and the climate system under current
and 2xCO2 conditions.
111-38
-------
8. REFERENCES
Baumgartner, A. and E. Reichel, The World Water Balance, 179 pp., Elsevier Scientific Publications,
Amsterdam, 1975.
Beaty, C. B., Sublimation or melting: observations from the White Mountains, California and Nevada,
U.S.A., Journal ofGlaciology, 14,275-286, 1975. „
Benton, G. S., R. T. Blackburn, and V. O. Snead, The role of the atmosphere in the hydrologic cycle, EOS
Transactions of the American Geophysical Union, 31,61-73,1950.
Botkin, D. B., J. F. Janak, and J. R. Wallis, Some ecological consequences of a computer model of forest
growth, Journal of Ecology, 60, 849-872,1972.
Broecker, W. S., Unpleasant surprises in the greenhouse?, Nature, 329,123-126,1987.
Brutsaert, W., Evaporation into the Atmosphere, 299 pp., D. Reidel, Dordrecht, 1982.
Brutsaert, W., Catchment-scale evaporation and the atmospheric boundary layer, Water Resources
Research, 22,39S-45S, 1986.
Budyko, M. I., Climate and Life, 508 pp., Academic Press, New York, 1974.
Budyko, M. I. and O. A. Drozdov, Characteristics of the moisture circulation in the atmosphere, Izv. Akad.
NaukSSSR,4,5-l4,l953.
Businger, J. A., J. C. Wyngaard, Y. Izumi, and E. F. Bradley, Flux-profile relationships in the atmospheric
surface layer, Journal of the Atmospheric Sciences, 28,181-189,1971.
Byers, H. R., General Meteorology, vol. Fourth Edition, 461 pp., McGraw-Hill, New York, 1974.
Campbell, W. G., D. Marks, J. J. Kineman, and R. T. Lozar, A geographic database for modeling the role of
the biosphere in climate change, submitted to Climatic Change, (1990).
Center, National Oceanographic and Atmospheric Administration, National Geophysical Data, Geo-
physics of North America: Users Guide, U.S. Department of Commerce, Boulder, Colorado, 1989.
Chamberlain, A. C., Transport of gasses to and from grass and grass-like surfaces, Proceedings of the
Royal Society of London, A290,236-265,1966.
Davis, R. E., J. Dozier, and D. Marks, Micrometeorological measurements and instrumentation in support
of remote sensing observations of an alpine snow cover, Proceedings of the Western Snow Confer-
ence, 57,161-164,1984.
Dickinson, R. E., Land surface processes and climate-Surface albedos and energy balance, Advances in
Geophysics, 25,305-353,1983.
Dickinson, R. E. and R. J. Cicerone, Future global warming from atmospheric trace gasses, Nature, 319,
109-115,1986.
Dickinson, R. E. and B. Hanson, Vegetation-albedo feedbacks, in Climate Processes and Climate Sensi-
tivity, edited by J. E. Hansen and T. Takahashi, Geophysical Monograph, vol. 5, pp. 180-187,
111-39
-------
American Geophysical Union, Washington, D.C., 1984.
Dozier, J., J. M. Melack, D. Marks, K. Elder, R. Kattlemann, and M. Williams, Snow deposition, melt,
runoff, and chemistry in a small alpine watershed, Emerald Lake basin, Sequoia National Park, Final
Report, CARB Contract A3-106-32, 367 pp., Computer Systems Laboratory, University of Califor-
nia, Santa Barbara, CA, 1988.
Eagleman, J. R., The Visualization of Climate, Heath, Lexington, Mass, 1976.
Eagleson, P. S., Climate, soil, and vegetation I. Introduction to water balance dynamics, Water Resources
Research, 14,705-712,1978.
Eagleson, P. S., Land Surface Processes in Atmospheric General Circulation Models, in Proceedings of
Greenbelt Study Conference, p. 560, Cambridge University Press, New York, 1982.
Eagleson, P. S.,The emergence of global-scale hydrology, Water Resources Research, 22,6S-14S, 1986.
Eagleson, P. S. and R. I. Segarra, Water-limited equilibrium of savanna vegetation systems, Water
Resources Research, 21,1483-1493,1985.
Elliott, D. L., C. G. Holladay, W. R. Barchet, H. P. Foote, and W. F. Sandusky, Wind Energy Resource
Atlas, Solar Technical Information Program, 210 pp., U.S. Department of Energy, 1987.
Eubank, R. L., Spline Smoothing and Nonparametric Regression, Statistics: Textbooks and Monographs,
vol. 90,455 pp., Marcel-Dekker, New York, 1989.
Fichtl, G. H. and G. E. McVehil, Longitudinal and lateral spectra of turbulence in the atmospheric boun-
dary layer at Kennedy Space Center, Journal of Applied Meteorology, 9,51-63,1970.
Fleagle, R. G. and J. A. Businger, An Introduction to Atmospheric Physics, 2nd ed., 432 pp., Academic
Press, New York, 1980.
Frew, J. E., The Image Processing Workbench, Ph.D. Thesis, 382 pp., Department of Geography, Univer-
sity of California, Santa Barbara, CA, 1990.
Gates, W. L., The use of general circulation models in the analysis of the ecosystem impacts of climatic
change, in Report of the Study Conference on Sensitivity of Ecosystems and Society to Climatic
Change, World Climatological Impact Study Program, World Meteorological Organization, Geneva,
1983.
Hansen, J., I. Fung, A. Lacis, D. Rind, S. Lebedeff, R. Ruedy, and G. Russell, Global climate changes as
forecast by Goddard Institute for Space Studies three dimensional model, Journal of Geophysical
Research, 93D, 9341-9364,1988.
Hansen, J., A. Lacis, D. Rind, G. Russell, P. Stone, I. Fung, R. Ruedy, and J. Lerner, Climate sensitivity,
analysis of feedback mechanisms, in Climate Processes and Climate Sensitivity, edited by J. E. Han-
sen and T. Takahashi, Geophysical Monograph, vol. 5, pp. 130-163, American Geophysical Union,
Washington, D.C., 1984.
111-40
-------
Holdridge, L. R., Determination of world formulations from simple climatic data, Science, 105, 367-368,
1947.
Holdridge, L. R., Life Zone Ecology, Tropical Science Center, San Jose, Costa Rica, 1967.
Isaaks, E. H. and R. M. Srivastava, Applied Geostatistics, 561 pp., Oxford University Press, New York,
1989.
Jenne, R. L., The NMC Octagonal Grid, National Center for Atmospheric Research, Boulder, CO, 1970.
Jenne, R. L.,Data From Climate Models; the CO2 Warming, National Center for Atmospheric Research,
Boulder, CO, 1990.
Jenne, R. L., D. Joseph, C. Mass, E. Recker, M. Albright, and H. Edmon, National Meteorological Center
Grid Point Data Set, Version II, Department of Atmospheric Sciences, University of Washington,
and National Center for Atmospheric Research, Seattle, WA, 1990.
Karl, T. R., C. N. Williams, F. T. Quinlan, and T. A. Boden, United States Historical Climatology Network
(HCN) Serial Temperature and Precipitation Data, 391 pp., U.S. Department of Energy, Carbon
Dioxide Information Anslysis Center, Oak Ridge, TN, 1990.
Keeling, C. D., Industrial production of carbon dioxide from fossil fuels and limestone, Tellus, 25, 174-
198,1973.
Keeling, C. D. and R. B. Bacastow, Impact of industrial gasses on climate, in Energy and Climate, pp. 72-
95, National Academy of Sciences, Washington, D.C., 1977.
Kondo, J., Observations on wind and temperature profiles near the ground, Science Reprints from Tohoku
University, 14,91-124,1962.
Korzoun, V. I., Atlas of World Water Balance, USSR National Committee for the International Decade,
UNESCO Press, Paris, 1977.
Korzun, V. I., World water balance and water resources of the earth, USSR National Committee for the
International Decade, 663 pp., UNESCO Press, Paris, 1978.
Kramer, P. J., Water Relations of Plants, 489 pp., Academic Press, San Diego, 1983.
Larcher, W., Physiological Plant Ecology, 303 pp., Springer-Verlag, New York, 1980.
Legates, D. R. and C. J. Willmott, Mean seasonal and spatial variability in global surface air temperature,
Theoretical and Applied Climatology, 41,11-21,1990.
Lettau, H., K. Lettau, and L. C. B. Molion, Amazonia's hydrologic cycle and the role of atmospheric recy-
cling in assessing deforestation effects, Monthly Weather Review, 107,227-237,1979.
Lettenmaier, D. P. and S. J. Surges, Climate change: Detection and its impact on hydrologic design, Water
Resources Research, 14,679-687,1978.
Lettenmaier, D. P. and T. Y. Can, Hydrologic sensitivities of the Sacramento-San Joaquin river basin, Cali-
fornia, to global warming, Water Resources Research, 26,69-87,1990.
111-41
-------
Lettenmaier, D. P. and D. P. Sheer, Assessing the effects of global warming on water resources systems:
An application to the Sacramento-San Joaquin River Basin, California, Journal of Water Resources
Planning and Management (in press), (February, 1991).
Manabe, S. and R. T. Wetherald, The effects of doubling the CO 2 concentration on the climate of a general
circulation model, Journal of Atmospheric Sciences, 32,3-15, 1975.
Manabe, S. and R. T. Wetherald, On the distribution of climate change resulting from an increase in CO2
content of the atmosphere, Journal of Atmospheric Sciences, 37,99-118,1980.
Marks, D., Climate, energy exchange, and snowmelt in Emerald Lake watershed. Sierra Nevada, Ph.D.
Dissertation, 158 pp., Departments of Geography and Mechanical Engineering, University of Cali-
fornia, Santa Barbara, CA, 1988.
Marks, D. and J. Dozier, Climate and energy exchange at the snow surface in the alpine region of the
Sierra Nevada, (submitted to Water Resources Research), (1990).
Marks, D. and J. Dozier, A clear-sky longwave radiation model for remote alpine areas, Archiv fur
Meteorologie, Geophysik undBioklimatologie, Series B, 27,159-187,1979.
Martin, P., N. J. Rosenberg, and M. S. McKenney, Sensitivity of evapotranspiration in a wheat field, a
forest, and a grassland to changes in climate and direct effects of carbon dioxide, Climatic Change,
74,111-117,1990.
Mather, J. R. and G. A. Yoshioka, The role of climate in the distribution of vegetation, Annals of the Amer-
ican Association of Geographers, 58,29-41,1968.
Palmer, W. C., Meteorological Drought, Research Paper, U.S. Weather Bureau, Washington, D.C., 1965.
Parry, M. L. and T. R. Carter, An assessment of the effects of climatic change on agriculture, Climatic
Change,!'5,95-116,1989.
Parry, M. T., T. R. Carter, N. Konijin, and J. Lockwood, The impact of climatic variations on agriculture, in
Introduction to the IIASAIVNEP Case Studies in Semi-Arid Regions, International Institute for
Applied Systems Analysis, Laxenburg, Austria, 1987.
Perrier, A., Land surface processes: vegetation, in Land Surface Processes in Atmospheric General Circu-
lation Models, Proceedings of Greenbelt Study Conference, Cambridge University Press, New York,
1982.
Quinlan, F. T., T. R. Karl, and C. N. Williams, United States Historical Climatology Network (HCN) Serial
Temperature and Precipitation Data, 160 pp., U.S. Department of Energy, Carbon Dioxide Informa-
tion Anslysis Center, Oak Ridge, TN, 1987.
Ramanathan, V. R., R. J. Cicerone, H. B. Singh, and J. T. Kiehl, Trace gas trends and their potential role in
climate change, Journal of Geophysical Research, 90,5547-5566,1985.
Rind, D., The influence of vegetation on the hydrologic cycle in a global climate model, in Climate
Processes and Climate Sensitivity, edited by J. E. Hansen and T. Takahashi, Geophysical
IIM2
-------
Monograph, vol. 5, pp. 73-92, American Geophysical Union, Washington, D.C., 1984.
Rind, D., R. Goldberg, J. Hansen, C. Rosenzweig, and R. Ruedy, Potential evapotranspiration and the
likelihood of future drought, Journal of Geophysical Research, 95,9983-10005, 1990.
Rosenzweig, C. and R. Dickinson, Climate-Vegetation Interactions, Workshop Proceedings , 156 pp.,
NASA/Goddard Space Flight Center, 1986.
Salati, E. and P. Vose, The Amazon basin: a system in equilibrium, Science, 225,129-138,1984.
Schlesinger, M. E., Quantitative analysis of feedbacks in climate model simulations of C02-induced
wanning, in Physically-Based Modelling and Simulation of Climate and Climatic Change, edited by
M. E. Schlesinger, C: Mathematical and Physical Sciences; NATO ASI Series, vol. 243, pp. 653-
737, Kluwer, Boston, 1988.
Shugart, H. H., A Theory of Forest Dynamics, 278 pp., Springer-Verlag, New York, 1984.
Shugart, H. H. and D. C. West, Development of an appalachian deciduous forest succession model and its
application to assessment of the impact of the chestnut blight, Journal of Environmental Manage-
ment, 5,161-179,1977.
Shukla, J. and Y. Mintz, Influence of land-surface evapotranspiration on the Earth's climate, Science, 215,
1498-1501,1982.
Smith, J. B. and D. A. Tirpak, The Potential Effects of Global Climate Change on the United States, Report
to Congress, vol. PM-221, 412 pp., United States Environmental Protection Agency, Washington,
D.C., 1989.
Snyder, J. P. and P. M. Voxland, An Album of Map Projections, Professional Paper, 249 pp., U.S. Geologi-
cal Survey, 1989.
Solomon, A. M., Transient response of forests to CC^-induced climate change: Simulation modeling
experiments in eastern North America, Oecologia, 68,567-579,1986.
Spangler, W. M. L. and R. L. Jenne, World Monthly Surface Station Climatology (and Associated
Datasets), National Center for Atmospheric Research, Boulder, CO, 1989.
Stephenson, N. L., Climatic control of vegetation distribution: the role of the water balance, submitted to
American Naturalist, (1989).
Stewart, B. J., Sensitivity and significance of turbulent energy exchange over an alpine snow surface, M.A.
Thesis, 41 pp., Department of Geography, University of California, Santa Barbara, CA, 1982.
Strain, B. R., Physiological and ecological controls on carbon sequestering in terrestrial ecosystems,
Biogeochemistry, 7,219-232,1985.
Sutton, O. G., Micrometeorology, 333 pp., McGraw-Hill, New York, 1953.
Thompson, S. L. and S. H. Schneider, Carbon dioxide and climate: The importance of realistic geography
in estimating the transient temperature response, Science, 217,1031-1033,1982.
IIM3
-------
Thornthwaite, C. W., An approach toward a rational classification of climate, Geographic Review, 38, 55-
89,1948.
Thornthwaite, C. W. and J. R. Mather, The Water Balance, Publications in Climatology, vol. 8, 104 pp.,
Laboratory of Climatology, 1955.
U.S. Army Corps of Engineers, GRASS Users and Programmers Manual, U.S. Army Corps of
Engineers, Construction Engineering Research Laboratory, Champagne, IL, 1988.
WeatherDisc Associates, Inc., World WeatherDisc: Climate Data for the Planet Earth, WeatherDisc
Associates, Inc., Seattle, WA, 1990.
Whittaker, R. H., Communities and Ecosystems, MacMillian, New York, 1975.
Woodward, F. I., Climate and Plant Distribution, Cambridge, New York, 1987.
111-44
-------
Characterizing the Distribution of Precipitation
Over the Continental United States
Using Historical Data
JAYNE DOLPH
DANNY MARKS
NSI Technology Services, Inc.
U.S. Environmental Protection Agency
Environmental Research Laboratory
200 S.W. 35th Street
Corvallis, OR 97333
(503) 757-4657
Abstract
Characterizing the distribution of precipitation at regional scales is a requirement
for the development of regional scale, spatially distributed hydrologic water
balance models. This study performs a preliminary assessment of the utility and
limitations of historical hydro-meteorological data for providing spatially
distributed precipitation estimates at large scales. The historical data are used in
a spatial analysis to characterize regional patterns of precipitation and runoff
across the continental United States. Precipitation and runoff "surfaces"
generated from interpolation of point measurements capture broad regional
patterns. A distributed water balance is calculated over the continent using long-
term (1948 -1988) and seven-year (1982 -1988) annual average values to check
the reliability of precipitation estimates. The resulting "input-output surfaces"
illustrate the deficiency (low elevation bias) of historical precipitation
measurements in the mountainous western United States where snowmelt is an
important component of the annual runoff. The incorporation of high elevation
snow measurements into the precipitation record for the seven-year average
significantly improves the water budget in these regions and enhances the utility
of historical data for providing distributed precipitation estimates at regional and
continental scales. The analysis presented in this study is exploratory in nature,
and will ultimately have implications for the development of large scale
hydrological and ecological modeling and climate change research.
IV - 1
-------
1. Introduction
Spatially distributed hydrological and ecological analyses are required to assess
the effects of climate change on water resources and vegetation (Eagleson, 1986;
Dooge, 1986; Neilson et al., 1989). The development of spatially distributed
models requires spatially distributed precipitation estimates at regionali to
continental scales. What is not well understood is whether historical data can be
used effectively to characterize large scale precipitation patterns and thereby
provide the necessary input to spatially distributed models. This study provides a
preliminary assessment of the utility of historical data for providing spatially
distributed, large scale precipitation estimates, and is the first step in
characterizing surface hydrology at scales compatible with climate change
research.
The global hydrologic cycle is an important component of the coupled ocean-
atmosphere-land surface system. It contributes to atmospheric circulation and
exerts a strong influence on weather and climate. Global climate change,
predicted to occur due to increasing concentrations of green-house gases in the
atmosphere, will cause changes in the earth's hydrologic cycle (Smith and Tirpak,
1989). These changes may significantly alter physical processes of the terrestrial
biosphere, because changes in hydrologic regimes directly affect land surface
properties such as soil moisture, vegetation health and distribution, and surface
albedo (Rasmusson et al., 1990; Hall et al., 1988; Hansen et al., 1984; Rind, 1984;
Shugart and West, 1977).
1 For this study, "regional" refers to areas on the order of roughly 50,000 to 500,000 km2, or approximately the size
of one water resources region as defined by the US Water Resources Council (1978).
IV - 2
-------
A major uncertainty at this time is the effect that climate change will have on
regional hydrology and the distribution of biomes (e.g. Dooge, 1990; Neilson et
al., 1989). In particular, it is not well understood how the spatial patterns and
seasonality of precipitation may change. While much research has been directed
toward understanding historical temperature records .(eg. Hanson and Lebedeff,
1987; Jones et al., 1986; Wigley et al., 1985), relatively little attention has been
given to large-scale precipitation patterns.
Presently, the only tool for predicting the effects of climate change on physical
processes at large scales is the General Circulation Model (GCM). GCMs are
global scale numerical models of atmospheric circulation that simulate the
fundamental physical relationships of the ocean-atmosphere-land surface system
(Manabe and Wetherald, 1975; Hansen et al., 1988; Dickinson and Cicerone,
1986). GCMs predict that global precipitation is likely to increase (e.g. Smith
and Tirpak, 1989), but it is not clear how regional precipitation patterns will be
affected. Furthermore, the geographical distribution of predicted change differs
from model to model (Schlesinger and Mitchell, 1987; Rind 1988). These
inconsistencies are due, in part, to the unrealistic parameterization of land surface
processes in the models which is a function of their coarse resolution (Eagleson,
1986; Rind et al., 1990). The GCM distributes precipitation uniformly over one
grid cell, tens to hundreds of thousands of km2 in size. However, precipitation is
a highly variable process, particularly at large spatial scales, and in
topographically diverse regions. Given the simplified treatment of precipitation
distribution in these models, their predictions of hydrologic sensitivity to climate
change are unreliable. A more realistic representation of the global hydrologic
cycle in the models would improve the reliability of precipitation estimates. This
improvement requires a better understanding of basic hydrologic phenomena at
regional scales (e.g. Eagleson, 1986).
IV - 3
-------
An important step towards improving precipitation estimates for large scale
modeling and climate change research is the characterization of past and present
spatial distributions of precipitation at regional scales using historical data. This
will better link surface hydrology to large scale climatic processes by providing
spatially distributed precipitation estimates at resolutions which are ecologically
and hydrologically meaningful.
2. Objectives
The objectives of this study are:
1) the development of a comprehensive database of historical precipitation and
runoff data for analyses at large scales;
2) a preliminary assessment of the utility of historical precipitation data for
characterizing the spatial distribution of precipitation at regional and
continental scales using water balance techniques;
3) the identification of those regions where the use of historical precipitation
data may be most limited;
4) the recommendation of improvements to the historical data, and
alternatives to its use for large scale precipitation estimation.
3. Related Research and Databases
Historical hydro-meteorological data are an important source of information on
past and present hydrologic conditions (e.g. Karl and Riebsame, 1989). In the
United States, extensive networks of precipitation and runoff measurement sites
have been established, and historical data are becoming widely available in digital
IV - 4
-------
form. The development of comprehensive national information bases and data
bases has substantially increased the amount of observational information
available on hydro-climatic dynamics. The goal of these efforts is primarily
inventory-oriented, aimed at data collection and the identification and projection
of water resource problems. Thus far, a high-quality geographic database,
combining the historical hydro-meteorological data into a single usable form for
spatial analysis and modeling at the continental scale has not been developed.
Several studies in recent years have investigated the utility of historical data for
describing hydrologic parameters at large scales. Rasmusson (1985) and
Ropelewski et al. (1986,1987) have analyzed global precipitation patterns
associated with anomalous climate events such as the El Nino Southern Oscillation
(ENSO) cycle. Trends in precipitation fluctuations over the northern and
southern hemisphere have been studied with the objective of assessing the
significance of projected climate change from GCMs as compared with the
variability evident in precipitation and temperature records (e.g. Diaz, et al.,
1989, Bradley, et al., 1987). Additionally, multi-year fluctuations in
temperature and precipitation at the continental scale have been investigated in
sensitivity studies to assess the utility of historical data for detecting secular
climate change (Karl 1987; Karl and Riebsame, 1989).
These studies have emphasized the need for improved estimates of precipitation at
regional scales, and a better understanding of the spatial and temporal variability
of land surface processes at resolutions which are hydrologically and ecologically
meaningful. As Bradley (1987) points out, the "magnitude of observed trends,
their geographic distribution, and differences between seasons need to be
examined in more detail.". Initially, this will involve determining the utility and
limitations of the historical data for adequately characterizing physical processes
IV - 5
-------
at a variety of spatial scales. Past correlations and empirical relationships
describing the spatial distribution and variability of precipitation may be
inadequate with climate change conditions because changes are predicted to be
more severe and rapid than the historical record indicates (e.g. Hansen et al.
1988). By using historical data to characterize precipitation and runoff patterns
at regional to continental scales, we can begin to identify those regions and
conditions where historical data are most limited, and where better data networks
and more focused research efforts are required.
4. Methods
The approach taken in this study was to utilize Geographic Information System
(GIS) technology and historical data to generate "surfaces" that represent the
spatial distribution of precipitation and runoff across the continental United
States. Distributed water balance techniques were developed using these surfaces
to make a preliminary assessment of the utility of historical data to account for
the spatial variability of precipitation at the continental scale. The output is
spatially distributed, long-term annual average precipitation estimates for the
United States.
4.1 Data
Synthesizing the immense volume of historical data at the continental scale is a
formidable task, due both to the quantity of data available and data quality
uncertainties. Analyzing patterns of precipitation and runoff at this scale requires
a substantial amount of data aggregation, manipulation, and quality control. The
United States was chosen as the study area for this research because the historical
data are abundant, readily available in digital form, and are generally of higher
quality than many other parts of the world. The analysis techniques developed
here may later be used to characterize precipitation patterns in other regions of
IV - 6
-------
the world, where data may be less reliable.
Historical data were aggregated from existing digital databases and synthesized
into a single, geographically-referenced database. Such an effort has only
become possible recently with the advent of mass storage devices such as CD-
ROM for data collection and storage, improvements in computing efficiency, and
the growing availability of GIS technology. These advances have provided the
mechanisms necessary to synthesize and analyze historical data in a spatial
context.
Runoff Data
Monthly time-series runoff data were obtained from a newly-created hydro-
meteorological database for the United States (Wallace et al., in review). The
database consists of 1014 unregulated or minimally regulated gauging sites with
streamflow data that have been corrected for station moves and missing values.
The database was compiled from the Earthlnfo (Earthlnfo, 1990) database of
daily runoff values on CD-ROM which is a digital compilation of the USGS daily
and peak values files. Runoff measurements from this database are for the
period 1948 to 1988. Figure 1. is a map of gauging station locations and
illustrates the spatial distribution of measurement sites across the country.
The historical runoff data used in this analysis had to meet several criteria. The
first criterion was a representative range of drainage area sizes so that runoff
contributed from large drainage basins and tributary flow from smaller basins
would be represented. The drainage area sizes from the database range from
IV - 7
-------
UNREGULATED RUNOFF GAUGING SITES
Source: Wallace eI .a I . , 1990
oo
* * V ** "\F^ %
*f ^*V****J
V*j j> t **V^
* y > ifvv* 1
V * ^fcf&
umber of sites - 1014
Figure 1. Runoff gauging station location map
-------
4.14 km2 to 35,224 km2. Figure 2. is a histogram showing the distribution of
drainage area sizes.
The second criterion was that each runoff gauge be operational for a substantial
number of years and have a continuous measurement record so that average
conditions could be represented. Where discontinuities in runoff records did
arise, data filling techniques were used and are described by Wallace, et al. (in
review).
Precipitation Data
Precipitation measurements were obtained from the Carbon Dioxide Research
Division of the United States Department of Energy and the National Climatic
Data Center of NOAA's Historical Climatology Network (HCN) (Quinlan et al.,
1987). The HCN data represent a high-quality dataset of monthly precipitation
measurements for 1211 stations which are free from anthropogenic and localized
effects on precipitation. Each station has a continuous eighty year record and is
quality-controlled to account for missing data. The dataset was specifically
designed for analyses of climate change at the regional scale (Quinlan et al.,
1987).
Figure 3. is a map of station locations showing the spatial distribution of HCN
measurement sites across the country. The spatial distribution of HCN stations is
adequate across much of the United States, except for the southwest. A greater
density of measurement sites in mountainous regions is desirable for
characterizing precipitation, because in topographically diverse areas
precipitation is highly variable due to orographic influences. Unfortunately most
precipitation gauges in the United States are located in low-lying areas.
Therefore, mountainous regions are not well represented by the historical data
IV - 9
-------
10
20
30
40
50
60
70
SO
90
100
200
300
400
500
600
700
800
900
1000
2000
30QO
4000
5000
6000
7000
8000
9000
10000
20000
30000
fc-vi
" Xij
vV
0 10 20 30 40 SO 60 70 80 90 100 110 120 130 140
FREQUENCY
Figure 2. Histogram of runoff gauging station drainage areas
IV - 10
-------
PRECIPITATION MEASUREMENT SITES
Source: NOAA , 1988; SCS, 1990
• HCN
Number of sites = 121
+ SNOTEL
Number of sites - 405
Figure 3. HCN and Snotel station location map
-------
and the density of measurement sites is actually lowest in high elevation areas.
HCN data from 1948 - 1987 were used for near-compatibility with the runoff
data, as 1988 data was not yet available at the time of this analysis.
Monthly precipitation measurements were also obtained from the Soil
Conservation Service's Snotel database (USDA-SCS, 1988) to increase the
elevational range of precipitation measurement sites. Snotel measurements are
available for twelve states in the western United States and represent snow water
equivalent measured at snow course locations. These stations are located in
mountainous regions and measure precipitation (rainfall and snowfall) at high
elevations on a year-round daily basis. A limitation of these data are their
relatively short period of record. Measurements are only available in digital
form for about a ten year period, on average. In several states (ie. California),
the data are unavailable before 1989 and could not be used for this analysis.
Monthly values from 1982 through 1987 for 405 stations were used for this
study. Figure 3. shows the locations of the Snotel measurement sites.
Figure 4. illustrates the low elevational range of HCN stations as compared with
the Snotel sites. The majority of HCN sites are located below 1,300 meters with
several sites located at 2,700 meters. The inclusion of 405 Snotel sites increases
the elevational range to 3,500 meters.
4.2 Continental Data Surfaces
Continental surfaces were generated using these data to characterize general
spatial patterns of precipitation and runoff across the country and to illustrate
precipitation-runoff relationships at the continental scale. Monthly runoff values
for the 1014 gauging stations were normalized by converting values of cfs-days
to unit runoff for comparability with precipitation data. This resulted in an
IV - 12
-------
450 -,
400 -
o -"
100 300 500 700 900 1100 1300 1500 1700 1900 2100 2300 2500 2700 2900 3100 3300 3500
ELEVATION (meters)
TYPE
HCN
SNO
Figure 4. Histogram of HCN and Snotel station elevations
IV - 13
-------
average volume of flow, or depth, at each gauge relative to that gauge's drainage
area. Average annual values of runoff were computed for each gauge for a 41-
year time period, from 1948 through 1988. Monthly precipitation measurements
from the 1211 HCN stations were aggregated to annual precipitation averages for
the period 1948 through 1987.
A raster-based GIS (USAGE, 1988) was used to distribute the point values of
precipitation and runoff to a 5 minute digital elevation grid (approximately
10km X 10km resolution) (NOAA, 1989) for the continental United States. An
inverse distance-squared algorithm (Isaaks and Srivastava, 1989) was used to
interpolate the point values of precipitation and runoff to annual precipitation and
runoff surfaces. The interpolation algorithm fills the grid cell matrix with
interpolated values generated from the input data points (eg. precipitation or
runoff measurement sites), keeping the original data values. The twelve nearest
data points are used to determine the interpolated value of each cell in the
surface. The boundary of the continental United States was used as a mask so that
only those cells falling within the study area were assigned interpolated values.
Numerical approximation techniques such as weighted-averaging interpolation
provide a mechanism for representing complex, irregular surfaces by employing
restrictions to the spatial influence of measurement errors that may otherwise
bias results (Isaaks and Srivastava, 1989). Figure 5. shows the 41-year average
annual runoff and precipitation surfaces.
The continental surfaces illustrate the general spatial extent of runoff and
precipitation across the United States. The use of long-term average values tends
to smooth out the influence of extreme hydrologic events. What is readily
apparent from viewing the surfaces is the magnitude of spatial variability of
runoff and precipitation at the continental scale, and the steep gradients which
IV - 14
-------
en
Figure 5. Average annual runoff (1948-1988) and precipitation (HCN) (1948-
1987) surfaces
-------
exist between major hydrologic regions.
Figures 4. and 5. suggest that precipitation in the western United States may be
under-represented by the historical data, due to the poor spatial distribution of
measurement sites in mountainous regions. With limited precipitation
measurements at high elevations, the interpolation of precipitation can not capture
orographic influences on precipitation variability. Furthermore, the elevational
range of measurement sites in the west can not account for the substantial portion
of precipitation deposited as snow.
4.3 Distributed Water Balance Calculation
A distributed water balance was calculated over the continent using the long-term
average annual runoff and precipitation surfaces to provide a general idea of
areas where obvious inconsistencies exist and where better precipitation
estimation techniques beyond simple linear interpolation algorithms, better data,
or a combination of both are needed.
The water balance equation (Linsley et al., 1982) is:
where
P = Precipitation
Q = Runoff
ET = Evapotranspiration
AS = Change in storage
R = Data measurement error and model error
IV - 16
-------
The water balance calculation used for this analysis is a variation on this
equation. When a long-term average is used, changes in storage can be assumed
negligible (Linsley et al., 1982) and therefor are unaccounted for here.
Evapotranspiration (ET) represents a significant portion of water movement
through the hydrologic cycle and is an important component of the annual water
balance. ET has been estimated to be 50% or more of the annual precipitation at
continental to global scales (Budyko, 1974; Brutsaert, 1986). Rind et al. (1990)
have reported that transpiration alone may account for as much as 70% of
precipitation in the United States. The remaining percentage appears as
streamflow, as indicated in the water balance equation. The difference, then,
between measured precipitation and measured runoff over sufficiently long
periods of time is a measure of losses primarily by ET (Chow, 1964). Given
these assumptions, the continental water balance equation simplifies to:
P = Q + R
where R is now a residual term which includes ET, data measurement error, and
model measurement error.
The terms (P and Q) of the water balance equation were represented by the
average annual precipitation and runoff surfaces. Precipitation (P) should
exceed or equal runoff (Q) across the continent if the historical data adequately
account for the spatial distribution and total amount of precipitation input.
Theoretically, the residual (R) should be positive or zero. For this analysis
however, we did not expect that the water budgets would actually 'balance1
because: 1) measurement errors are inherent in the historical data; 2) the
interpolation of point values to a surface has limitations; and 3) precipitation
measurement sites in mountainous regions are poorly distributed, spatially and
IV - 17
-------
topographically. We did expect the water balance to indicate broad regions
where measured runoff is substantially greater than measured precipitation, and
thus point out those areas where the historical data may poorly characterize the
spatial distribution of precipitation.
5. Results
Using the 1948-1988 long-term average precipitation and runoff surfaces, a
distributed water balance was calculated over the continent by subtracting the
annual runoff surface from the annual precipitation surface. Figure 6. illustrates
the resulting annual "input-output surface". The blue areas represent regions
where the water balance calculation produced residuals which are positive
(measured precipitation exceeds runoff), with light blue representing those
regions having residual values falling within 20 cm of balance and dark blue
representing regions with larger positive residuals. These regions occur
consistently throughout the eastern and midwestern states and to a much lesser
extent in the mountainous west. Red areas depict regions of the country where
the water balance calculation produced residuals which are negative (measure
runoff exceeds measured precipitation), with lighter red representing regions
with residuals falling within 20 cm of balance (negative), and dark red
representing those regions with large negative residuals. These dark red regions
are areas where inconsistencies in the measured data are most severe. This
occurs almost exclusively in the mountainous west, most notably in the Pacific
Northwest, the Great Basin, and to a lesser extent in parts California. The input-
output surface illustrates the inadequacy of historical data to account for the
amount and spatial distribution of precipitation in the mountainous western
United States.
Snotel data were incorporated into the precipitation record to increase the
IV - 18
-------
CO
Figure 6. Annual input-output surface 1948-1988
-------
elevational range and spatial density of precipitation observations and to improve
the distributed water balance calculations. Average annual precipitation values
were calculated from monthly HCN and Snotel site data for the period 1982 to
1988. Monthly runoff values for 1982 to 1988 were aggregated to annual
averages. These values were interpolated using the procedure described earlier
to create new annual precipitation-snow and runoff surfaces. Additionally,
precipitation surfaces were generated using only the HCN data for the period
1982 to 1987 to compare the spatial distribution of precipitation with and without
the incorporation of snow data in the precipitation record (Figure 7.).
The spatial patterns of the short-term surfaces were visually compared to the 41-
year surfaces and found to be highly similar. While average annual precipitation
values are greater during the 1948 to 1988 period, particularly in parts of the
west, the general patterns of precipitation and runoff across the country do not
vary significantly when short-term average values, are used rather than the 41-
year average values. However, comparing the short-term precipitation-snow
surfaces to the short-term precipitation surfaces calculated using only the HCN
data, illustrates a significant increase in measured precipitation input in the
mountainous west with the inclusion of the Snotel data, most notably in the
Pacific Northwest, Great Basin, and Upper Colorado regions.
The short-term precipitation-snow and runoff surfaces were used to calculate a
new distributed water balance across the continent. An annual water balance was
also calculated using the short-term precipitation surfaces generated using only
the HCN data to compare and quantify the improvement when including snow in
the calculations (Figure 8). The input-output surfaces calculated with the
incorporation of Snotel data illustrate a substantial improvement in the distributed
water balance. As compared with the input-output surfaces calculated using only
IV - 20
-------
; Runoff
1' r e ci pi t a t ion (HCN on Iy )
i
N>
Figure 7. Average annual runoff (1982-1988), precipitation (HCN) (1982-
1987), and precipitation-snow (1982-1988) surfaces
-------
PS3
Figure 8. Annual input-output surfaces with and without snow 1982-1988
-------
the HCN data, a greater portion of the mountainous west, particularly large
regions of the Pacific Northwest, Great Basin, and Upper Colorado have positive
rather than negative residuals. This analysis illustrates that care must be taken in
the use of historical precipitation estimates to avoid under-estimation in
mountainous regions, and that the inclusion of snow into the precipitation record
enhances the utility of the historical data for characterizing the spatial distribution
of precipitation by increasing the elevational range and spatial density of
measurement sites. It also points out the need to consider orographic effects on
the spatial distribution of precipitation in mountainous regions.
Annual average runoff, precipitation, and precipitation-snow values from the
short-term surfaces were calculated for four major hydrologic regions in the
western United States to quantify the improvement in measured precipitation
volume with the inclusion of snow deposition data into the precipitation record.
The size of this region is 173.34 km2 and includes the Columbia Basin, the Great
Basin, the Upper Colorado River Basin and the Lower Colorado River Basin.
These values (in depths of water) are shown in Figure 9. The average annual
runoff value for this region is 50.6 cm. The average annual precipitation value
using only HCN data, at 46.5 cm, does not even equal the measured runoff. With
the inclusion of Snotel data, this value increases to 68 cm, accounting for the total
measured runoff plus a positive residual of 17.4 cm.
6. Limitations
The analysis presented is exploratory in nature, and part of a larger effort aimed
at the development of large scale ecological, hydrological, and biogeochemical
models for climate change research. It is important to note the limitations
associated with the data and the techniques which were used.
IV - 23
-------
E
u
Q.
V
Q
Volumes for the Western US
(values represent average annual depth)
runoff prscip precip-snow
Figure 9.
Bar graph showing the improvement in the annual water balance
calculation for the western US with the inclusion of snow data
IV - 24
-------
The use of unregulated gauges rather than the entire gauging network limits the
spatial density of point measurements, particularly in areas of the western United
States. Additionally, the extrapolation of runoff from one drainage to another
possibly quite distant drainage, has obvious limitations. However, at the
continental scale, detailed basin information is difficult to consider and the use of
a time-consistent, quality-controlled data set of unregulated flow measurements
was the natural place to begin. In some areas, we will need to incorporate more
measurement sites for future work, and undoubtedly include flow records from
regulated basins (for example, in the southwest data measurements are sparse and
regulated flow comprises a substantial portion of the measured runoff). The use
of regulated flow exclusively will become particularly critical for regional scale,
seasonal analyses using monthly time series data rather than long-term averages.
The point here is that while augmentation of the precipitation-runoff database
with additional data is required, we must also develop better techniques for
distributing these data both spatially and temporally in topographically diverse
regions.
'Our use of unregulated gauging stations exclusively for this study, introduced a
small drainage area bias. This bias may actually tend to over-estimate runoff, as
smaller basins yield greater depth of flow per unit area. Conversely, unregulated
gaifging stations having large drainage areas are typically "drier". In any event,
we do not feel that by using unregulated flow exclusively, our runoff estimates
are off by an order of magnitude, or any substantial amount for the purposes of
this analysis. The emphasis in this study was not to attempt to simulate runoff
distribution or predict runoff at a point, but rather to use the runoff surfaces as a
check on our precipitation estimates in a general sense, to identify those regions
where severe inconsistencies exist in the measured values of precipitation and
runoff.
IV - 25
-------
The procedure used to generate the precipitation and runoff surfaces is a simple,
linear interpolation algorithm. As such, it may introduce estimate bias in areas of
high topographic relief. However, while other interpolation algorithms may
yield precipitation and runoff estimates which are different, we do not feel that
they would contradict the conclusions from this analysis - that precipitation is
underestimated in regions of the western United States by the historical data, and
that the inclusion of snow data into the historical record improves the
precipitation estimates in mountainous regions. This analysis is a first-cut effort
at spatially distributing precipitation at the continental scale for the identification
of regions where more detailed analyses should be directed. Other models and
interpolation algorithms which can include elevation and other variables as
factors to better simulate the spatial distribution of precipitation-elevation
relationships should be considered in future work at a variety of spatial scales.
7. Conclusions
The development of large scale ecological and hydrological models requires
spatially distributed estimates of precipitation. While historical hydro-
meteorological data may be used to characterize the spatial distribution of
precipitation across large regions, geographic scale and topographic variability
must be considered. In the United States, historical data characterize
precipitation best in regions with minimal topographic variability, namely the
eastern and mid-western regions. However, the historical data may be severely
limiting in mountainous regions of the west, in part, because the data do not
account for precipitation inputs at high elevations.
This study has illustrated that the integration of snow measurements into the
precipitation record is essential for improving precipitation estimates. The
incorporation of snow data into the precipitation record for four major
IV - 26
-------
hydrologic regions of the western United States improved the precipitation input
by roughly 40%. While the addition of snow data significantly improves the
spatial distribution and elevational range of measurement sites, these data have a
limited geographical extent and period of record. Additionally, the inclusion of
these data will not, in itself, resolve the temporal variability of precipitation-
runoff relationships. Resolving the seasonal cycle of winter snow deposition
followed by spring snowmelt runoff is a more complex issue which is critical for
predicting water availability under both current and predicted 2XCO2 climate
conditions in water-stressed areas of the western United States. Ultimately,
techniques must be developed to use these and other sources of data to account for
snow depletion and runoff timing in mountainous regions.
This research has established a database for precipitation and runoff analyses at
regional and continental scales and has produced spatially distributed precipitation
estimates for the United States at a resolution which is ecologically meaningful.
Most importantly, it has demonstrated that it is insufficient to simply use
historical data to characterize precipitation at large scales, particularly in
mountainous regions where dramatic precipitation gradients exist. Spatially
distributed precipitation estimates that account for the spatial arrangement of
measurement sites, precipitation-elevation relationships, and topographic
constraints on precipitation distribution are needed.
IV - 27
-------
REFERENCES
Bradley, R.S., H.F. Diaz, J.K. Eischeid, P.O. Jones, P.M. Kelly and C.M.
Goodess, Precipitation fluctuations over northern hemisphere land areas
since the mid-19th century, Science, 1987.
Brutsaert, W., Catchment-scale evaporation and the atmospheric boundary layer,
Water Resources Research, 22, 39S-45S, 1986.
Budyko, M.I., Climate and Life, pp. 508, Academic Press, New York, 1974.
Chow, Ven Te, Handbook of Applied Hydrology, McGraw-Hill, New York,
1964.
Diaz, H.F., R.S. Bradley, and J.K. Eischeid, Precipitation fluctuations over global
land areas since the late 1800's, Journal of Geophysical Research, vol. 94,
1989.
Dickinson, R.E. and R.J. Cicerone, Future global warming from atmospheric
trace gases, Nature, 319, 109-115, 1986.
Dooge, James, C.I., Looking for hydrologic laws, in Trends and Directions in
Hydrology, ed. Stephen J. Burges, pp. 46S-58S, American Geophysical
Union, Washington, D.C., 1986.
Dooge, James, C.I., Hydrologic models and climate change, keynote address
Chapman Conference on Hydrologic Aspects of Global Climate Change,
1990.
Eagleson, P.S., Land Surface Processes in Atmospheric General Circulation
Models, in Proceedings of Greenbelt Study Conference, p. 560, Cambridge
University Press, New York, 1982.
Eagleson, P.S., The emergence of global-scale hydrology, Water Resources
Research, 22,6S-14S, 1986.
Earthlnfo, Hydrodata User's Manual - Daily and Peak Values, Denver, CO, 1988.
Hall, F.G., D.E. Strebel, and P.J. Sellers, Linking knowledge among spatial and
temporal scales: vegetation, atmosphere, climate, and remote sensing,
Landscape Ecology, vol. 2, 1988.
IV - 28
-------
Hansen, J., I. Fung, A. Lacis, D. Rind, S. Lebedeff, R. Ruedy, and G. Russell,
Global climate changes as forecast by Goddard Institute for Space Studies
three dimensional model, Journal of Geophysical Research, 93D, 9341-
9364, 1988.
Hansen, J., and S.Lebedeff, Global trends of measured surface air temperature,
Journal of Geophysical Research, 92, 13345-13372, 1987.
Hansen, J., A. Lacis, D. Rind, G. Russell, P. Stone, I. Fung, R. Ruedy, and J.
Lemer, Climate sensitivity, analysis of feedback mechanisms, in Climate
Processes and Climate Sensitivity, edited by I.E. Hansen and T. Takahashi,
Geophysical Monograph, vol. 5, pp. 130-163, American Geophysical
Union, Washington, D.C., 1984.
Isaaks, E.H. and R.M. Srivastava, Applied Geostatistics, pp. 561, Oxford
University Press, New York, 1989.
Jones, P.D., T.M.L. Wigley, and P.B. Wright, Global temperature variations
between 1861 and 1984, Nature, 322, 430-434, 1986.
Karl, T.R., Multi-year fluctuations of temperature and precipitation: the gray
area of climate change, Climatic Change, vol. 12, 1987.
Karl, T.R. and W.E. Riebsame, The impact of decadal fluctuations in mean
precipitation and temperature on runoff: a sensitivity study over the
United States, Climatic Change, vol. 15, 1989.
Linsley, R.K., M.A. Kohler, and L.H. Paulhus, Hydrology for Engineers,
McGraw Hill, 1982.
Manabe, S. and R.T. Wetherald, On the distribution of climate change resulting
from an increase in COi concentration on the climate of a general
circulation model, Journal of Atmospheric Sciences, 32, 3-15, 1975.
National Oceanographic and Atmospheric Administration, National Geophysical
Data Center, Geophysics of North America: Users Guide, U.S.
Department of Commerce, Boulder, Colorado, 1989.
IV - 29
-------
Neilson, R.P., G.A. King, R.L. DeVelice, J. Lenihan, D. Marks, J. Dolph, B.
Campbell, and G. Glick. Sensitivity of ecological landscapes and regions to
global climate change. U.S. Environmental Protection Agency,
Environmental Research Laboratory, Corvallis, Oregon, 1989.
Quinlan, F.T., T.R. Karl, and C.N. Williams, United States Historical
Climatology Network (HCN) Serial Temperature and Precipitation Data,
U.S. Department of Energy, Carbon Dioxide Information Analysis Center,
Oak Ridge, TN, 1987.
Rasmusson, E.M., El Nino and variations in climate, American Scientist, 73, 168-
177, 1985.
Rasmusson, E.M. and P.A. Arkin, Measuring and understanding global
precipitation variability, in Symposium on Global Change Systems - Special
Sessions on Climate Variations and Hydrology, Anaheim, CA, 1990.
Rind, D., The influence of vegetation on the hydrologic cycle in a global climate
model, in Climate Processes and Climate Sensitivity, edited by J.E. Hansen
and T. Takahashi, Geophysical Monograph, vol. 5, pp. 73-92, American
Geophysical Union, Washington, D.C., 1984
Rind, C., The doubled C02 climate and the sensitivity of the modeled hydrologic
cycle, Journal of Geophysical Research, 93, 5385-5412, 1988.
Rind, D., R. Goldberg, J. Hansen, C. Rosenzweig, and R. Ruedy, Potential
evapotranspiration and the likelihood of future drought, Journal of
Geophysical Research, 95, 9983-10005,1990.
Ropelewski, D.F.and M.S. Halpert, North American precipitation and
temperature patterns associated with the El Nino Southern Oscillation
Cycle, Monthly Weather Review, 1986.
Ropelewski, D.F. and M.S. Halpert, Global and regional scale precipitation
patterns associated with the El Nino Souther Oscillation Cycle, Monthly
Weather Review, 1987.
Schlesinger, M.E. and J.F.B. Mitchell, Climate model simulations of the
equilibrium climatic response to increasing carbon dioxide, Reviews of
Geophysics, 25, 760-798,1987.
IV - 30
-------
Shugart, H.H. and D.C. West, Development of an appalachian deciduous forest
succession model and its application to assessment of the impact of the
chestnut blight, Journal of Environmental Management, 5, 161-179, 1977.
Smith, J.B. and D.A. Tirpak, The Potential Effects of Global Climate Change on
the United States, Report to Congress, vol. PM-221, United States
Environmental Protection Agency, Washington, D.C. 1989.
U.S. Army Corps of Engineers, GRASS Users and Programmers Manual, U.S.
Army Corps of Engineers, Construction Engineering Research Laboratory,
Champagne, IL, 1988.
U.S. Water Resources Council, The Nation's Water Resources 1975-2000,
VolumeS Analytical Data Summary, Washington, D.C. 1978
U.S.D.A. Soil Conservation Service, Snow Survey and Water Supply Products
Reference, West National Technical Center, Portland, OR, 1988.
Wallace, J.R., D.P. Lettenmaier, and E.F. Wood, A daily hydroclimatological
data set for the continental United States, Water Resources Research, (in
review), 1990.
Wigley, T.M.L., J.Angell, and P.D. Jones, Analysis of the temperature record, in
Detecting the Climatic Effects of Increasing Carbon Dioxide, DOE/ER-
0235, pp. 198, Dept. of Energy, Washington, D.C., 1985.
IV - 31
-------
A COMPARISON OF GEOSTATISTICAL PROCEDURES FOR
SPATIAL ANALYSIS OF PRECIPITATION IN MOUNTAINOUS TERRAIN
Donald L. Phillips
U.S. Environmental Protection Agency
Jayne Dolph
NSI Technology Services, Inc.
Danny Marks
NSI Technology Services, Inc.
U.S. EPA Environmental Research Laboratory
Corvallis, OR 97333
The information in this document has been funded wholly or in part by the U.S.
Environmental Protection Agency. It has been subject to the agency's peer and
administrative review, and it has been approved for publication as an EPA
document. Mention of trade names or commercial products does not constitute
endorsement or recommendation for use.
V - 1
-------
ABSTRACT
Application of simulation models to assessment of global climate change
effects often requires spatially distributed estimates of precipitation, both
under current and future climate scenarios. Simple interpolation methods fail
to consider the effects of topography on precipitation and may be in
considerable error in mountainous regions. The performance of three
geostatistical methods for making mean annual precipitation estimates on a
regular grid of points in mountainous terrain was evaluated. The methods were
(1) kriging, (2) kriging elevation-detrended data, and (3) cokriging with
elevation as an auxiliary variable. The study area was the Willamette River
Basin, a 2.9 million hectare region spanning the area between the Coast Range
and the Cascade Range in western Oregon. Compared to kriging, detrended
kriging and cokriging .both exhibited better precision (coefficients of
variation of 16% and 17% vs. 21%; average absolute errors of 19 cm and 20 cm
vs. 26 cm) and accuracy (average errors of -1.4 cm and -2.0 cm vs. -5.2 cm) in
the estimation of mean annual precipitation. Contour diagrams for kriging and
detrended kriging exhibited smooth zonation following general elevation
trends, while cokriging showed a patchier pattern more closely corresponding
to local topographic features. Detrended kriging and cokriging offer improved
spatially distributed precipitation estimates in mountainous terrain on the
scale of a few million hectares. Application of these methods for a larger
region, the Columbia River drainage in the U.S. (57 million hectares), was
unsuccessful due to the lack of a consistent precipitation-elevation
relationship at this scale. Precipitation estimation incorporating the
effects of topography at larger scales will require either piecewise
estimation using the methods described here or development of a physically-
based orographic model.
V - 2
-------
INTRODUCTION
A substantial portion of the assessment of the potential effects of climate
change on natural resources and ecosystems has utilized simulation models in
conjunction with climate scenarios. For example, the EPA Congressional Report
on climate change effects in the United States (Smith and Tirpak, 1989) gave
results from models simulating hydrologic processes, crop growth, and forest
dynamics. These models used climate scenarios derived from general
circulation models (GCMs) of the atmosphere simulating conditions of doubled
atmospheric C02 concentrations. Typically, changes in modeled temperature and
precipitation between current and doubled C02 concentrations have been used to
adjust the historical weather record from weather stations in the area of
interest. However, U.S. Weather Service stations are located in an irregular,
coarse grid. Many areas, particularly in the western United States, have only
sparse coverage. In addition, topographic relief has large effects on
precipitation. Precipitation generally increases with elevation, and mountain
ranges also create "rain shadows" on the leeward side. Weather stations tend
to be sited at low elevations and may thus underestimate the regional
precipitation (Dolph, in review). Precipitation at higher (or lower)
elevations near a weather station may not be accurately reflected by the
weather station measurements.
For these reasons, especially in areas of high topographic relief, it will
often be insufficient to use data from the nearest weather station as the
climate input for hydrology or vegetation models to assess climate change
impacts. This is especially true if the models are being applied to a spatial
distribution of points to represent different facets of the landscape (Marks,
in review). Spatially distributed precipitation data, which take into account
the spatial arrangement of weather station data, precipitation-elevation
relationships, and topographic relief are needed.
Geostatistical interpolation methods, such as kriging, were originally
developed for spatial analysis of ore reserves in mining (Matheron, 1971) , but
have since been applied to a number of other problems, including spatial
interpolation of precipitation (Dingman et al., 1988; Istok et al., 1990a,b;
V - 3
-------
Tabios and Salas, 1985). Tabios and Salas (1985) found kriging to be superior
to other commonly used interpolation techniques such as Thiessen polygons,
polynomial trend surfaces, inverse distance, and inverse square distance
methods for precipitation estimation in a 52,000 km2 region in Nebraska and
Kansas.
Kriging does not explicitly account for the influence of elevation on
precipitation in mountainous areas except as reflected in the precipitation of
surrounding weather stations. If, however, the neighboring stations are at
different elevations than the point being estimated, then the estimate is
likely to be in error. To address this problem, Chua and Bras (1982) and
Dingman et al. (1988) performed linear regressions on precipitation vs.
elevation, subtracted the regressed elevation effect, and performed kriging on
the elevation-adjusted data. Cokriging is another geostatistical method which
uses a second correlated auxiliary variable to aid in estimation of the
primary variable. Cokriging may be especially advantageous in situations
where the auxiliary variable is highly correlated with the primary variable (r
> 0.5) and is oversampled compared to the primary variable (Vauclin et al.,
1983; Yates and Warrick, 1987). Only recently has this been applied to
precipitation estimation using elevation as the auxiliary variable (Istok et
al., 1990a,b). Both of these methods seem to hold promise for improved
spatially distributed estimation of precipitation. The purpose of this paper
is to evaluate the effectiveness of geostatistical procedures which utilize
topographic information (cokriging and elevation-detrended kriging) and those
which do not (kriging), for estimating precipitation across mountainous
terrain, in order to provide spatially distributed climate data for models
assessing climate change effects.
OVERVIEW OF GEOSTATISTICAL THEORY
The basic goal of geostatistical methods such as kriging and cokriging is to
interpolate values for points or areas which have not been sampled, using data
from surrounding sampled points. Any interpolation scheme must assign a
series of weights to the neighboring points to be used in order to compute an
interpolated value of the variable of interest (e.g., precipitation). Simple
V - 4
-------
interpolation schemes may assign equal weights to each of n nearest neighbors,
for example, or they may assign weights inversely proportional to the distance
to the estimation point. The first of these methods does not take into
account the spatial proximity of the neighboring points in interpolating a
value for the estimation point. The second does, but assumes a particular
relationship (inverse) between the weight and the distance. Thus, the weights
are not necessarily optimal. Kriging and cokriging are interpolation methods
which attempt to optimize the weights assigned to the neighboring data points
in computing the interpolated value.
Kriging and cokriging consist of three steps: (1) an examination of the
covariation of data values depending on their distances apart; (2) fitting
theoretical models to these relationships; and (3) using these models to
calculate the weights for a particular set of neighboring points and computing
the interpolated value. The first step is referred to as constructing a
sample semivariogram. All possible pairs of data points are examined, the
pairs are grouped by distance classes, and one half the variance of the
difference in values (the semivariance) is graphed vs. the distance. Second,
a theoretical curve (model semivariogram) is fit to these points either by eye
or using least squares regression. Lastly, this model determines the weights
to be used for each neighboring point to compute the interpolated values. For
kriging, interpolations are made using only the semivariogram for the variable
to be interpolated. For cokriging, additional semivariograms for other
correlated variables are also used to help make the estimate.
The theoretical semivariogram model is defined by the nugget variance (the
variance at zero distance), the sill (the variance to which the semivariogram
asymptotically rises), and the range (the distance at which the sill or some
predetermined fraction of the sill is reached). See Figure 3 for an example
semivariogram.
In their simplest form, kriging and cokriging assume that there is no "trend"
or "drift" in the data, that is, no consistent, directional gradient in the
variable(s). This is often not the case, but the assumption may be made if
the neighborhood of points used in interpolation is smaller than the distance
V - 5
-------
over which the trend is apparent (Journel and Rossi, 1989). A related concept
is isotropy, the condition under which the semivariance is the same for a
given distance, regardless of direction. Again, this is often not strictly
true, but small differences may be ignored and isotropy is often assumed for
procedural simplicity.
METHODS
Study Area
The study area is the Willamette River basin in Oregon. The Willamette River
begins in southwestern Oregon and flows north into the Columbia River at
Portland, Oregon. The basin is approximately 170 km wide and 270 km long and
covers 2.9 million hectares. It is bordered by two major topographic
features, the Coast Range to the west and the Cascade Range to the east. The
elevation ranges from near 0 m above sea level at the confluence with the
Columbia River to 3200 m at the peak of Mt. Jefferson in the Cascade Range
(Fig. 1). Annual precipitation ranges from approximately 100 cm to 300 cm.
The area has a Meditteranean climate with wet winters and dry summers.
Cyclonic storms are carried in on the polar front jet stream which dips
southward into the region in the winter and retreats northward in the summer.
The high mountains have minor convective activity in the summer.
Precipitation and Elevation Data
Precipitation data were obtained from 37 U.S. Weather Service stations
(Earthlnfo, 1990) and 15 Soil Conservation Service SNOTEL (snow telemetry)
stations in the basin (Soil Conservation Service, 1988) [Fig. 2]. The SNOTEL
data were added to increase the elevation range for precipitation data and to
give a better regional representation of precipitation patterns (Dolph, in
review). The U.S. Weather Service stations ranged in elevation from 46 m to
1095 m. The SNOTEL stations ranged in elevation from 610 m to 1494 m.
Precipitation from both sources includes rainfall and snow water equivalents.
Because of the recency of the SNOTEL network and our desire for a uniform
period of record for the two data sources, we used data from water years 1982-
V - 6
-------
45.831
45.23
o>
0)
"O
.g 44.63
jg
S
OJ
44.02
43.42
-123.67
-123.17 -122.67
Longitude (deg)
-122.17
-121.67
Elevation (m)
250
1000
:;«» 1750
500
1250
s; 2000
Figure 1. Elevation contour diagram of the study area.
V - 7
-------
Willome tte River Basin
Weather Service and SNOTEL Station Locations
+ Weather Service
O SNOTEL
Figure 2. Locations of the 52 U.S. Weather Service and SNOTEL
weather stations.
V - 8
-------
1988 (10/81-9/88). The results reported here will focus on mean annual
precipitation, but additional analyses are being done on a seasonal and
monthly basis. The 52 stations had a grand mean annual precipitation of 162
cm, with a range of 99 cm to 293 cm. The distribution of mean annual
precipitation for the 52 sites was skewed right (p<.01 for Kolmogorov-Smirnov
normality test). The data were decimal log transformed to approximate a
normal distribution (p>.05 for Kolmogorov-Smirnov normality test).
The elevation data set included the elevations for the 52 precipitation
stations, plus 478 additional points on a 5-minute latitude/longitude grid
covering the basin. Elevation values for this grid were derived from a
Digital Elevation Model with 5-minute resolution (National Geophysical Data
Center, 1989; Campbell and Marks, 1990). The mean elevation for the 530 data
points was 629 m with a range of 2 m to 2250 m. The distribution of elevation
values was also right skewed (p<.01 for Kolmogorov-Smirnov normality test), so
the data were decimal log transformed to approximate a normal distribution
(p>.05 for Kolmogorov-Smirnov normality test).
Semivariograms
For all semivariograms and cross -semivariograms, both isotropic and
anisotropic variograms were constructed. To examine possible anisotropy,
pairs of perpendicular directional semivariograms with tolerances of +45
degrees were calculated for angles of 0 to 90 degrees (relative to Albers
projection; 0 degrees is approximately north) in increments of 15 degrees.
For ordinary kriging of log annual precipitation (LAP), the sample
semivariograms were constructed using the 52 precipitation data points. The
same sample LAP semivariogram was used for cokriging. Also for cokriging, log
elevation (LEL) sample semivariograms were constructed using the 478 5-minute
grid points plus the 52 weather station elevations. The cokriging sample
cross-semivariograms used the 52 stations which had both LAP and LEL values.
For elevation-detrending of precipitation data, a linear regression was
performed with LAP as the dependent variable and LEL as the independent
variable (n=52). The residuals (RESID) were then used to construct
semivariograms for use in detrended kriging. SAS (SAS Institute, 1985) was
V - 9
-------
used for the linear regression. The GEOPACK geostatistical software package
(Yates and Yates, 1990) was used for all semivariogram, cross-semivariogram,
kriging, and cokriging procedures. The GEO-EAS geostatistical software
package (Englund and Sparks, 1988) and programs written in SAS were also used
for cross-validation procedures described below.
Kriging and Cokriging Procedures
The kriging and cokriging systems of equations are as outlined in Vauclin et
al. (1983). For detrended kriging, the elevation-detrended residuals (RESID)
were used instead of LAP. This detrending procedure is essentially equivalent
to one iteration of the universal kriging procedure for kriging in the
presence of trends (Isaaks and Srivastava, 1989).
Cross-Validation Procedures
Semivariograms for LAP and elevation-detrended residuals (RESID), and cross-
semivariograms for LAP-LEL were initially fit with a least squares procedure.
The model form was selected by visual examination of the sample semivariograms
and cross -semivariograms. Several different models were tried for each. A
jackknife procedure of cross-validation was used. Data points were deleted
one at a time, values were kriged or cokriged for the missing points, and the
estimated and actual values were compared. The Reduced Mean and Reduced
Variance cross-validation statistics were computed (Vauclin et al., 1983;
Yates and Warrick, 1987) . For a set of n points at locations XA and values
z(Xi), there are corresponding estimates z*(Xi) with estimation variances
ak2(xi) • The values z(Xj_) have sample variance s2. The Reduced Mean (RM) is
defined by:
RM = — _
n I^i
and should be close to 0 if the estimates are unbiased. The Reduced Variance
(RV) is defined by:
V - 10
-------
n
RV = —£
and should be close to 1 if the estimation variances are consistent. Nugget,
sill, and range coefficients were iteratively changed to improve the RM and RV
performance of the semivariograms and cross -semivariograms. To insure that
the cross-semivariogram was positive definite, an indicator of positive
definite condition (PDC) was overlaid on the model cross-semivariogram. This
indicator is defined by:
PDC=
where 7pp(h) and 7ee(h) are the semivariances at distance h for LAP and LEL,
respectively. The cross-semivariogram model was checked to be sure that the
semivariance 7pe(h) was less than PDC for all distances h.
RESULTS AND DISCUSSION
Precipitation Semivariograms
For LAP, the directional semivariograms oriented at 0 and 90 degrees showed
the greatest variability in the semivariances. For spherical models with
similar sills (0.020 and 0.018), the ranges were not greatly different (67 and
82). Because of the small size of the precipitation data set, the anisotropy
ratio near unity (82/67 = 1.2), and for simplicity of modeling, isotropy was
assumed. The semivariogram for LAP is shown in Figure 3, along with the
fitted model. The rise of the semivariance above the sample variance is
evidence of a trend for distances > 90 km. Model fitting was done for the
points in the semivariogram which fell below this distance. The nugget, sill,
range, reduced mean, and reduced variance values for the selected model are
shown in Table 1.
Elevation-detrended Precipitation Residual Semivariograms
Figure 4 shows the regression line for LAP as a function of LEL. The
V - 11
-------
0.022
0.021
0.020
0.019
0.018
0.017
0.016 -
0.015 -
* 0.014 :
g 0.013-
| 0.012 :
> 0.011-
I 0.010 :
05 0.009 -
0.008 :
0.007 :
0.006 :
0.005 :
0.004 :
0.003 :
0.002 :
0.001 -
0 10 20 30 40 50 60 70 80 90 100 110 120 130 140 150
Distance (km)
* * * actual
model
Figure 3. Sample and model semivariograms for LAP (loglO annual precipitation in cm). The
model was a spherical model with a nugget of 0.0016, sill minus nugget of 0.0154, and range
77 km. The horizontal dashed line shows the sample variance (n=52).
-------
2.5 H
2.4
i
§ ^
£3
3
J5.
I 2.2
"to
I
S 2.1:
2.0:
1.9 :
I l | i |— i f T [ r | i | i |- i | l p i | l p
1.0 1.2 1.4 1.6 1.8 2.0 2.2 2.4 2.6 2.8 3.0 3.2
LoglO Elevation (m)
Figure A. Regression of LAP (loglO annual precipitation in cm) vs. LEL (loglO elevation in m),
The regression model is LAP = 1.675 + 0.2108 LEL (F = 104.8; df = 1; p< 0.0001), with a
correlation coefficient r = 0.82.
-------
TABLE 1. Summary of semivariogram and cross-semivariogram models.
Variable (s)
LAP
LEL
RESID
LAP/LEL
n
52
530
52
52
Nugget
Model (C0)
Spherical 0.0016
Gaussian 0.029
Spherical 0.00167
Exponential 0
Sill
minus
Nugget
(Ci)
0.0154
0.346
0.00372
0.0794
Range
(a)
77
86
67
71
Reduced
Mean
0.037
0.003
0.000
0.040
Reduced
Variance
0.99
0.90
1.00
0.96
Models:
Spherical
(h) =
CQ + cjl.5 ha - 0.5 /—13| if h <. a
C + C - "' '
<-0 -r Oj_
otherwise
Exponential:
= C
cl -
Gaussian:
y (h) = 1 - exp -3
h2
V - 14
-------
regression equation was:
LAP = 1.675 + 0.2108 x LEL
where LAP and LEL are the decimal logs of annual precipitation (cm) and
elevation (m), respectively. For the Willamette River basin, the two
variables were highly correlated with r = 0.82 (n - 52). Regressions were
performed for both log-transformed and untransformed annual precipitation and
elevation data. The LAP vs. LEL regression had the highest correlation, the
best spacing of data along both axes, used the variables already employed in
kriging and cokriging, and thus was chosen for the elevation detrending. The
sample semivariogram and model semivariograms for the residuals from this
regression (RESID) are shown in Figure 5. The directional semivariograms did
not show a great deal of anisotropy so an isotropic model was fit. The rise
of the semivariance above the sample variance shows evidence of a trend for
distances > 115 km. The model was fit for the 12 points below this distance
(see Table 1).
Elevation Semivariograms
Anisotropy was examined by same method as for LAP. The maximum variability of
semivariances was evident in directional variograms oriented at 45 and 135
degrees, rather than the 0 and 90 degree orientation for precipitation.
Variability among directional semivariances was fairly small for 0 and 90
degrees. Even though anisotropy was evident for distances above 30 km at
angles of 45 and 135 degrees, isotropy was assumed for the following reasons:
o This did not coincide with the direction of maximum semivariance
variability for LAP, the primary variable.
o With a regular 5 minute grid of elevation values, the number of
neighbors (8) used in cokriging would all occur within about 10-20
km of the point being estimated. Anisotropy at these distances
was not strongly evident regardless of the angle examined.
o Cokriging estimates of precipitation are to be made at points
coinciding with the location of elevation values. Therefore, the
V - 15
-------
<;
i
0.0070 -
0.0065 -
0.0060 -
0.0055 -
0.0050 -
-------
influence of surrounding elevation values (within 10-20 km) will
be slight, and anisotropy will make little difference.
o Ease of modeling and cokriging.
The sample and model semivariogram are shown in Figure 6. As was the case
with LAP, there was evidence of a trend (rise above sample variance) for
distances > 90 km. The model was fit for sample semivariogram points where h
< 90 km (see Table 1).
Precipitation/Elevation Cross -Semivariograms
The directional cross-semivariograms for LAP/LEL showed patterns very similar
to those for the semivariograms for LAP. Ln the isotropic cross-
semivariogram, as with both LAP and LEL semivariograms, there was evidence of
a trend (a rise above the sample covariance) for distances > 90 km (Fig. 7).
The model was fit for points below this distance (see Table 1).
Precipitation Estimation
Figures 8, 9, and 10 show the precipitation estimates determined by ordinary
kriging, elevation detrended kriging, and cokriging, respectively for the 478
5-minute grid points using a search radius of 60 km. The median and range of
the precipitation estimates by each method are summarized in Table 2. The
associated estimation coefficients of variation (CV) are shown in Figures 11,
12, and 13. The mean CV for precipitation estimates was 21% for kriging, 16%
for detrended kriging, and 17% for cokriging. All three methods showed the
largest CVs on the edges of the study area. Along the boundary, the size of
the neighborhoods incorporating the nearest 8 points were larger, and in some
cases less than 8 neighbors occurred within the maximum ellipse radius of 60
km.
The precipitation estimates by ordinary kriging show fairly smooth zonal
patterns, with minimum precipitation at low elevations in a band running
roughly north-south, and increasing precipitation to the west and east as
elevation generally increased up the Coast Range and Cascade mountains. Of
V. - 17
-------
00
0.36-
0.34-
0.32
0.30-
0.28 :
0.26 :
0.24
| 0.22-
I 0.20
-------
i
>->
vO
o,
o
-------
45.83
43.424-
-123.67
Precip. (cm)
-123.17
90
210
-122.67
Longitude (deg)
-122.17
-121.67
m 240 *mmm 270
Figure 8. Estimates of annual precipitation (in cm, back-transformed
from LAP estimates) from ordinary kriging.
V - 20
-------
45.83
45.23
0)
•
44.63
44.02
43.42
-123.67
Precip. (cm)
-123.17
90
210
-122.67
Longitude (deg)
120
240
-122.17
-121.67
150
270
180
Figure 9. Estimates of annual precipitation (in cm, back-transformed from
LAP residual estimates) Erorn detrended kriging.
V - 21
-------
45.83
45.23
O)
0)
T3
.g 44.63
us
CO
44.02
43.42
-123.67
Precip. (cm)
-123.17
90
210
-122.67
Longitude (deg)
120
240
-122.17
150 180
270
-121.67
Figure 10. Estimates of annual precipitation (in cm, back-transformed
from LA? estimates) from cokrir.ing.
V - 22
-------
45.83
45.23 -
O)
0)
_g
"ro
44.63
44.02
43.424
-123.67
-123.17
-122.67
Longitude (deg)
-122.17
-121.67
X. CV of estimate
10 BSBSftiUlSBiSBB lo %SSSS3$$$$$8& 20
p c ..,-...-,....,.,.,....,,, qn ,-,,-,,,,,,.-,,,,.. q c;
£.\J i';.:-:-:-:-:':-;-!-:-:-:-:-:-:-: Ow :-:•:•:•:.:-:•:•:•:-:•:•:-:-:•:•: OO
Figure 11. Coefficients of variation (CV, in %) for annual precipitation
estimates from ordinary kriging.
-------
45.83 T-—
43.42
-123.67
-123.17 -122.67
Longitude (deg)
-122.17
&S8S888S888S88& IO.U
-121.67
/. CV of estimate tm^m 10.0
8KS858S58 17.5 'tmmm 20.0
Figure 12. Coefficients of variation (CV, in %) for annual precipitation
estimates from detrended kriging.
V - 2k
-------
45.83
45.23
44.63
03
44.02
43.42
-123.67
-123.17
-122.67
Longitude (cleg)
-122.17
X. CV of estimate
10
25
30 35
-121.67
Figure 13. Coefficients of variation (CV, in %) for annual precipitation
estimates Ifrom cokriging.
V - 25
-------
TABLE 2. Summary of precipitation estimates by kriging, detrended kriging,
and cokriging for 478 points in a 5 minute grid. The estimate statistics are
presented in the original (back-transformed) units of cm annual precipitation.
The variance ratios are for loglO precipitation values on which the kriging
and cokriging were done, and compare the estimation variances of each method
with those of ordinary kriging.
Min Max Median Mean % CV mean %
estimate estimate estimate for ppt. variance
Method (cm) (cm) (cm) estimate reduction
Kriging 104 265 151 21
Detrended 97 259 171 16 38
Kriging
Cokriging 77 319 162 17 28
V - 26
-------
course, elevation at the kriging estimation points was not taken into account
with this method, so these patterns result from the patterns of the weather
station data. Interpolation between stations within the search radius of 60
km produced the smoothed contours seen in Fig. 8.
A similar smooth pattern was seen for the detrended kriging method (Fig. 9).
The minimum and maximum precipitation estimates were similar to those from
ordinary kriging (Table 2), but the contour levels within the basin were
shifted westward compared to the kriging contours. Thus, overall the
detrended kriging estimates were somewhat higher than the kriging estimates,
with a median of 171 cm compared to 151 cm for kriging.
The cokriging estimates exhibited a greater range, with a lower minimum and a
higher maximum than either of the other two methods (Table 2). The median
estimate (162 cm) was intermediate. The pattern of precipitation contours is
much more broken and less smooth (Fig. 10). This same phenomenon was observed
by Istok et al. (1990b) and Martinez-Cob (1990) when comparing cokriging (with
elevation) estimates with kriging estimates of precipitation and
evapotranspiration, respectively. They attributed this to a closer
correspondence between the precipitation at a point and local orographic
factors which are considered in the cokriging method. While detrended kriging
does take elevation into account, it does so by incorporating a regional
average effect of elevation into a kriging interpolation of the precipitation
data within the search neighborhood. Cokriging, on the other hand, considers
not only the local variation of precipitation within the search neighborhood,
but the local variation of elevation as well, and thus the estimates are more
closely tied to localized orographic features.
Both detrended kriging and cokriging showed considerable average reductions in
estimation variance compared to ordinary kriging (Table 2) of 38% and 28%,
respectively. Dingman et al. (1988) similarly found reductions in estimation
variance for detrended kriging of precipitation, as did Istok (1990b) for
cokriging of precipitation, and Martinez-Cob (1990) for cokriging of
evapotranspiration. Cross-validation on the 52 weather station points showed
average absolute errors of 26 cm for kriging, 19 cm for detrended kriging, and
V - 27
-------
20 cm for cokriging. Thus, both of the latter two methods seem capable of
producing estimates of greater precision than ordinary kriging. Accounting
for elevation effects, either by detrending for a regional average effect
(detrended kriging) or by incorporation of elevation semivariograms and cross-
semivariograms (cokriging), also gave increased accuracy of precipitation
estimates. In cross-validation, the average errors for the 52 weather station
locations were -5.2 cm for kriging, -1.4 cm for detrended kriging, and -2.0 cm
for cokriging.
Issues of Scale
While detrended kriging and cokriging using precipitation-elevation
relationships seem to be promising methods for producing spatially distributed
precipitation scenarios for global climate change effects research, the scale
of the geographic area to which they are applied must be considered. An
initial examination was made of the precipitation-elevation relationship in a
larger regio'n, the Columbia River drainage area in the United States. This
drainage area covers 57 million hectares and includes 8 other major hydrologic
subregions besides the Willamette River basin. Because of the dramatic
precipitation gradients across this region and the inclusion of major
orographic features, the precipitation-elevation relationship was not
consistent across the region. The correlation coefficient r for LAP and LEL
was 0.00 (n = 491). The Willamette River Basin study area covered sizable
ranges of both elevation and precipitation, and included several very
important orographic features. However, the advantage of using a
hydrologically defined area such as this is that the major orographic features
occurred on the edge of the region, and all the points in the region fell on
the same side of a given orographic feature. Expansion to too large a
geographic area includes orographic effects in the interior and weakens the
relationship between elevation and precipitation, which is the basis of the
improved precipitation estimates by either detrended kriging or cokriging.
There are several alternatives for spatially distributed estimation of
precipitation over a large region. The estimates could be done piecewise by
using cokriging or detrended kriging on a number of smaller subregions with
V - 28
-------
consistent precipitation/elevation relationships. Ordinary kriging could be
performed over the entire region. However, ordinary kriging was shown to be
inferior to the other two methods which explicitly consider elevation when
precipitation was strongly correlated with elevation. Also, there would
undoubtedly be directional trends and anisotropic effects of considerable
magnitude over such a large area. Other geostatistical techniques to be
explored include fitting polynomial trend surfaces and using universal
kriging, and median polish kriging (Cressie, 1986). Simpler interpolation
methods such as inverse distance weighting could be used, but were found to be
inferior to geostatistical estimation of precipitation by Tabios and Salas
(1985).
Kriging and cokriging are exact interpolation methods, i.e., the surfaces
generated always pass through the known data points. Noise in the
precipitation data may cause some points to be outliers, especially with
fairly short periods of record. Other non-exact interpolation techniques less
sensitive to outliers, such as nonparametric trend surface analysis methods,
should be explored as well.
A final alternative is the development and use of a physical orographic model
incorporating effects of topography, storm tracks, atmospheric thermodynamics,
etc. (Peck and Schaake, 1990). However, such a model may be considerably more
data-intensive than the statistical methods above.
Future Work
For assessment of impacts on natural resources, global climate change research
requires regional scale hydrologic data, including spatially distributed
estimates of precipitation (Dooge, 1986; Eagleson, 1986). This paper has
demonstrated the utility of several geostatistical methods in providing
improved precipitation estimates for hydrologically defined regions. One
application for these methods will be estimation of precipitation and
temperature for input into forest "gap models" to simulate forest dynamics
across a Pacific Northwest landscape under climate change scenarios.
V - 29
-------
Further analyses will be performed to examine the variation of precipitation
patterns in extreme wet and dry years to examine regional patterns of climatic
variability as historic analogs to climate change. Seasonal and monthly
analyses will be done to increase the temporal resolution of these methods.
In addition, methods for spatially distributed estimation of precipitation in
larger regions will be examined and tested.
CONCLUSIONS
Detrended kriging and cokriging both provide spatially distributed estimates
of precipitation which take into account precipitation-elevation
relationships. The inclusion of these relationships leads to precipitation
estimates of improved accuracy and precision in mountainous terrain on the
scale of a few million hectares. Such estimates are needed for the use of
spatially distributed models of hydrology and vegetation to assess the effects
of climate change scenarios. Precipitation-elevation relationships weaken at
larger scales, necessitating a different approach for providing spatially
distributed climate data.
V - 30
-------
LITERATURE CITED
Campbell, W.G. and D. Marks (1990). A geographic database for modeling
the role of the biosphere in climate change. Biospheric Feedbacks to
Climate Change: The Sensitivity of Regional Trace Gas Emissions,
Evapotranspiration, and Energy Balance to Vegetation Redistribution,
Chapter II. EPA report, ERL-Corvallis.
Chua, S. and R.L. Bras (1982). Optimal estimators of mean areal
precipitation in regions of orographic influence. Journal of Hydrology
57: 23-48.
Cressie, N. (1986). Kriging non-stationary data. Journal of the American
Statistical Association 81: 625-634.
Dingman, S.L., D.M. Seely-Reynolds, andR.C. Reynolds III (1988).
Application of kriging to estimating mean annual precipitation in a
region of orographic influence. Water Resources Bulletin 24: 329-339.
Dolph, J. (1990). Characterizing the distribution of precipitation and
runoff over the continental United States using historical data.
Biospheric Feedbacks to Climate Change: The Sensitivity of Regional
Trace Gas Emissions, Evapotranspiration, and Energy Balance to
Vegetation Redistribution, Chapter IV. EPA report, ERL-Corvallis.
Dooge, J.C.I. (1986). Looking for hydrologic laws. Water Resources
Research 22: 465-585.
Eagleson, P.S. (1986). The emergence of global-scale hydrology. Water
Resources Research 22: 65-145.
Earthlnfo, Inc. (1990). ClimateData User's Manual, TD3200 Summary of the
Day - Cooperative Observer Network. Earthlnfo, Inc., Denver, Colorado.
V - 31
-------
England, E. and A. Sparks (1988). GEO-EAS (Geostatistical Environmental
Assessment Software) User's Guide. U.S. Environmental Protection
Agency, Las Vegas, NV.
Isaaks, E.H. and R.M. Srivastava (1989). An introduction to applied
geostatistics. Oxford University Press, New York. 561 pp.
Istok, J.D., J.A. Hevesi, and A.L. Flint (1990a). Precipitation estimation
in mountainous terrain using multivariate geostatistics: 1. Structural
analysis. Unpublished manuscript.
Istok, J.D., J.A. Hevesi, and A.L. Flint (1990b). Precipitation estimation
in mountainous terrain using multivariate geostatistics: 2. Isohyetal
maps. Unpublished manuscript.
Journel, A.G. and M.E. Rossi (1989). When do we need a trend model in
kriging? Mathematical Geology 21: 715-739.
Marks, D. (1990). The sensitivity of potential evapotranspiration to
climate change over the continental United States. Biospheric Feedbacks
to Climate Change: The Sensitivity of Regional Trace Gas Emissions,
Evapotranspiration, and Energy Balance to Vegetation Redistribution,
Chapter III. EPA report, ERL-Corvallis.
Martinez-Cob, A. (1990). Multivariate geostatistical analysis of
evapotranspiration and elevation for various climatic regimes in Oregon.
Ph.D. dissertation, Agricultural Engineering Department, Oregon State
University, Corvallis. 228 pp.
Matheron, G. (1971). The theory of regionalized variables and its
applications. Cahiers du Centre de Morphologie Mathematique, Ecole des
Mines, Fountainebleau, France, 211 pp.
V - 32
-------
National Geophysical Data Center, National Oceanic and Atmospheric
Administration (1989). Geophysics of North America: User's Guide. U.S.
Department of Commerce, Boulder, Colorado.
Peck, E.L. and J.C. Schaake (1990). Network design for water supply
forecasting in the west. Water Resources Bulletin (in press).
SAS Institute (1985). SAS User's Guide: Statistics, Version 5 Edition.
SAS Institute, Inc., Gary, NC.
Smith, J.B. and D.A. Tirpak (1989). The Potential Effects of Global Climate
Change on the United States. Report to Congress, U.S. Environmental
Protection Agency, Washington, D.C.
Soil Conservation Service, U.S. Department of Agriculture (1988). Snow
Survey and Water Supply Products Reference, West National Technical
Center, Snow Survey Program, Portland, Oregon.
Tabios, G.Q. and J.D. Salas (1985). A comparative analysis of techniques
for spatial interpolation of precipitation. Water Resources Bulletin
21: 365-380.
Yates, S.R. and M.V. Yates (1990). Geostatistics for Waste Management:
A User's Manual for the GEOPACK (Version 1.0) Geostatistical Software
System. U.S. Environmental Protection Agency, Ada, OK.
V - 33
-------
EFFECTS OF GLOBAL CLIMATE CHANGE ON GLOBAL VEGETATION
George A. King
NSI Technology Services Corporation
US EPA Environmental Research Agency
200 SW 35th St
Corvallis, OR 97333
Rik Leemans
Global Change Department
National Institute of Public Health and
Environmental Protection, RIVM
P.O. Box 1
3720 BA Bilthoven
the Netherlands
September 1990
VI - 1
-------
ABSTRACT
EFFECTS OF GLOBAL CLIMATE CHANGE ON GLOBAL VEGETATION
The potential redistribution of vegetation types in response to
climate change has been estimated in several analyses using
climate/vegetation correlation systems such as that of Holdridge
combined with GCM climate scenarios. A Review of results using
five different GCMs revealed a general agreement in terms of the
sign of the predicted change in the areal extent of specific
vegetation types, i.e., deserts, boreal forest, and tundra biomes
decrease, while grasslands and temperate and tropical forests
generally increase in area. These changes reflect the increases
in precipitation expected in temperate and tropical areas, and
the effect of rising temperatures in high latitudes that displace
or eliminate tundra and boreal biomes. The proportion of land
surface area changing from one vegetation type to another ranged
from 16 to 56 % depending on the GCM used.
-------
Introduction
Changes in fluxes of carbon to the atmosphere from the terrestrial
biosphere can significantly affect atmospheric CO2 concentrations
(e.g. Tans et al. 1990). Projected changes in global climate (see
Appendix A) could result in a significant redistribution of global
vegetation, resulting in large changes in carbon pools and fluxes.
These changes could provide significant positive or negative
feedbacks to climate change. To begin an evaluation of the
potential significance of this feedback mechanism, the current best
estimates of changes in global vegetation caused by climate change
are summarized here. The data presented in this paper forms the
basis for calculating changes in terrestrial carbon storage in
Turner et al. (1990).
Global Vegetation Models
Several models describing the global distribution of vegetation (or
at least classification schemes relating climate to vegetation) are
available with which to estimate the redistribution of vegetation
in response to climate change. These include models of Holdridge
(1947, 1964), Koppen (1900, 1918, 1936), Box (1981), and Lashof
(1987). Tuhkanen (1980) and Prentice (1990) have reviewed these
models. The original Holdridge (1947, 1964), modified Holdridge
(Prentice 1990), Box (1981), and Lashof (1987) approaches have been
VI - 2
-------
used to predict the redistribution of vegetation in response to
climate change.
The Holdridge life zone classification uses three climate
parameters to define the occurrence of major plant formations:
biotemperature, mean annual precipitation, and a potential
evapotranspiration (PET) ratio. Biotemperature is a temperature
sum over the course of a year, with unit values (e.g., daily
values) that are less than 0°C set to 0°C. PET, as Holdridge (1964)
defined it, is a linear function of biotemperature, and thus it
does not add degrees of freedom to the model. Thus, the Holdridge
classification system is really only based on two climate
variables, biotemperature and total annual precipitation. The PET
ratio is defined as PET/mean annual precipitation. Holdridge
created an axis system using these three climate parameters to
classify life zones (Figure 1). Mean annual precipitation forms
two axes of an equilateral triangle, with the third being the PET
ratio. Annual biotemperature forms a separate axis perpendicular
to the base of the triangle. The Holdridge classification thus
creates hexagons that define specific life zones.
VI - 3
-------
V"
Aliitudinal
lielts
Ci
6
a>
O
'3
=Warm Temperate n
^uyuu.MV.^ «X. -AN.... >A^«.r,>^'««_
Tropical'^/ V ™ 'V.^A "" ^ "" 1 ."* t ,"". 'I' T
•/ ,K '" ,k 4 A ' A ' /K ,\ '"••" /'s '
7—*~T—TTr*i ^77TVV//VT'/7VTV"™"VTn^7/77777W7;
^ 2-1.0 5
-1 I-
3.U
G.U
12.0
^rViSv ^ 2-1.0
Critical
tenifierature
line
Figure 1. Holdridge life zone classification system
(Holdridge 1967).
VI - 4
-------
A significant factor in evaluating simulations of future vegetation
patterns is understanding how well the vegetation model simulates
current vegetation patterns. The Holdridge life zone
classification reproduces broad-scale global vegetation patterns
(Leemans 1990), but it is inaccurate for many regions of.the world
(only vegetation in 40% of 1° x 1° gridboxes are correctly simulated
by the Holdridge system (Prentice 1990)). Prentice (1990)
evaluated arid refined the Holdridge system to improve its accuracy.
In the initial simulation, in which climate space was divided on a
finer scale than in the original Holdridge classification,
vegetation in 58% of the land gridboxes was correctly simulated.
An additional refinement was to aggregate observed vegetation units
based on similarity of vegetation and climate. For the final
analysis, 18 primary vegetation types and 11 transitional zones
were defined. Using this scheme, vegetation in 77% of the
gridboxes was correctly predicted.
A more complicated but more biologically realistic global
vegetation model was developed by Box (1981) . Instead of analyzing
climate-biome relationships, Box analyzed the distributions of 77
plant life forms (e.g., summergreen broadleaved trees) throughout
the world. For each life form, a set of eight different climate
values were correlated with the range limits of the life form. In
essence, Box created a set of 77 different climate envelopes within
which the life form occurs and outside which the life form is
absent. The Box model can predict combinations of growth forms at
VI - 5
-------
any given location, so it is capable of predicting canopy
structure.
Box validated his model by simulating life forms present at 74
sites around the world. The actual dominant growth forms were
predicted for 92% of the sites, but all dominants and codominants,
and no others, were correctly predicted for only 50% of the sites.
Lashoff (1987) developed a statistical model of climate-vegetation
relationships using Olson et al.'s (1983) vegetation database.
Since this database reflects actual rather than potential
vegetation, the Lashoff model is the only currently available
approach that incorporates land use.
Limitations of Global Vegetation Models
The four global vegetation models just summarized have significant
limitations; for predicting future global vegetation patterns.
First and foremost, all are steady-state models and non-dynamic.
They give no information on how long it would take the vegetation
to return to equilibrium with climate. It may take 200-500 years
for forest and shrub species to respond to a large climate change
(Davis et al. 1986, Webb 1986). Thus, these models cannot be used
directly to estimate vegetation patterns that could exist in the
next century. Second, the Holdridge and Lashoff approaches, and to
a lesser degree the Box approach, are based on empirical
VI - 6
-------
relationships between climate and vegetation. Since future climate
regimes in some regions may be different than any present in the
world today, vegetation models based on the mechanistic response of
species or growth forms to climate must be developed. Third, the
models are imperfect predictors of present vegetation, which
introduces bias into the simulations of future vegetation patterns.
Fourth, none of the models take into account the effect of soils
(e.g., nutrient availability, texture) or land use (with the
exception of the Lashoff model) on vegetation distribution. Thus,
simulating future vegetation using these models assumes that there
are appropriate changes in soils to support the predicted
vegetation type. In regions where soils have a major effect on
vegetating cover, predictions by the vegetation models will likely
be incorrect. Fifth, the direct effects of CO2 on plants are not
incorporated into the models.
Equilibrium Simulations of Future Vegetation Patterns
Despite the limitations discussed above, simulations of the
potential response of vegetation to future climate change are
useful for understanding both the magnitude of possible vegetation
change and biospheric feedbacks to climate change. Several global
simulations have been recently completed that form the basis for
estimating changes in carbon pools and fluxes. Emanuel et al.
(1985) made the first global projection of future vegetation
VI - 7
-------
patterns using the Holdridge approach, although only changes in
temperature were used to calculate future vegetation patterns. A
more recent and complete simulation using the Holdridge approach
and five cli:nate scenarios has been completed by Leemans (1990) and
Smith et al.(1990). Prentice and Fung (1990) used the refined
Holdridge system (Prentice 1990) to project future vegetation
patterns and carbon storage. Smith et al. (1990) also used the Box
model to project changes in vegetation in the tropical and boreal
regions of the world.
Major changes in the distribution of the world's vegetation are
projected under all the double-C02 climate scenarios and the
different vegetation models (Emanuel et al. 1985a,b, Smith et al.
1990, Prentice and Fung 1990). The results of each analysis are
briefly summarized here. It should be emphasized that these
analyses are. for potential natural vegetation, and do not take into
account human land-use (e.g., agriculture).
Emanuel et al. (1985a,b) created a future climate scenario using
temperature data from Manabe and Stouffer's (1980) simulation of
global climate under quadruple-C02 concentrations. Temperature
changes were divided by two to create a double-CO2 climate
scenario. Under this scenario, tropical forests, grasslands,
subtropical deserts, and boreal deserts showed significant
expansion, while subtropical forests, boreal forests, warm
temperate forests, tundra, and ice contracted significantly (Table
VI - 8
-------
Table l. Changes in Global Vegetation Distribution in a
Double-CO2 Atmosphere using the Holdridge Classification System as
Applied by Emanuel et al. 1985a,b.
Vegetation
Type
Ice
Tundra
Boreal Deserts
Boreal Forests
Cool Temperate Deserts
Grasslands
Cool Temperate Forests
Warm Temperate Deserts
Warm Temperate Forests
Subtropical Deserts
Tropical Deserts
Subtropica.1 Forests
Tropical Forests
Total
Current
Area
(106-km2)
2.22
4.47
1.31
17.26
4.84
22.78
11.29
6.75
15.81
2.91
10.74
11.96
19.03
131.37
2 X
Are
(106-
0.
3.
2.
10.
4.
28.
11.
5.
14.
3.
12.
9.
24.
131.
co2
a
km25
57
03
59
88
04
52
63
41
69
66
63
38
33
36
Percent
Change
-74
-32
98
-37
- 16
25
3
20
- 7
26
18
-22
28
1) . Smith et al. (1990) used five different climate scenarios
generated from output from the four GCMs described in Appendix A.
The fifth scenario is a second run of the GFDL model using a more
realistic heat flux, and is referred to here as GFDL-QFlux. Under
all climate scenarios, the tundra, cold parklands, forest tundra,
boreal forest, cool desert, warm temperate forest, and tropical
VI - 9
-------
seasonal forest biomes generally decreased significantly in areal
extent (Table 2). The temperate forest, tropical semi-arid,
tropical dry forest, and tropical rain forest biomes increased in
areal extent. Tropical rain forests doubled in area under the GISS
and OSU climate scenarios. The GFDL-Qflux scenario results
represent the least change from the present, with only the cold
parklands and tropical rain forests changing in size by more than
20%.
Table 2. Changes in Global Potential Vegetation Distribution in a
Double-CO2 Atmosphere using the Holdridge Classification System as applied by
Smith et al.(1990).
Biome
Tundra
Cold Parklands
Forest Tundra
Boreal Forest
Cool Desert
Steppe
Temperate Forest
Hot Desert
Chaparral
Warm Temperate Forest
Tropical Semi-Arid
Tropical Dry Forest
Tropical Seasonal Forest
Tropical Rain Forest
Total
Current
Area
(106-km2)
9.30
2.79
8.90
15.03
4.01
7.39
9.94
20.85
5.58
3.17
9.56
14.86
15.13
8.46
134.97
GFDL
-66
1
-56
-36
-24
57
19
- 1
33
-38
46
32
-34
82
Percent
GFDL-QFlx
- 9
-28
- 1
3
-13
-10
17
- 4
- 3
7
-10
- 5
- 5
47
:age Chai
GISS
-54
-15
-34
-10
-42
- 6
35
-15
- 2
-40
75
30
-48
105
ige
OSU UKMO
-49
- 4
-33
- 6
-21
18
16
- 7
-12
-23
27
0
-33
137
-69
-39
-62
-32
-48
- 3
31
- 4
54
- 9
74
75
-49
52
Prentice and Fung (1990), using a GISS climate scenario, also
reported significant increases in the areal extent of tropical rain
forests, as well as savanna, cold deciduous broadleaved forest and
woodland, and drought deciduous forest (Table 3). Drought
deciduous woodland, arid grassland, desert, evergreen needle-leaved
forest, tundra, and temperate evergreen seasonal forest all showed
significant decreases in area.
VI - 10
-------
Table 3. Changes in Global Vegetation Distribution in a
Double-CO2 Atmosphere using the Holdridge Classification System as
modified by Prentice (1990) and applied by Prentice and Fung (1990)
Area
Vegetation Type (106'km2)
Polar Desert and Ice
Tundra
Cold-Deciduous Needleleaved
Forest and Woodland
Evergreen Needleleaved Forest
and Woodland
Mesic Grassland
Drought-Deciduous Woodland
Arid Grassland and Shrubland
Cold-Deciduous Broadleaved
Forest and Woodland
Temperate Evergreen Seasonal
Broadleaved Forest
Mediterranean Forest and Woodland
Desert
Savanna
Drought-Dec iduous , Drought-
Seasonal Broadleaved Forest
Tropical Rain Forest
3.00
7.00
16.00
6.00
2.00
6.00
30.00
12.00
2.00
2.00
14.00
5.00
9.00
19.00
Perc.
Change
0
- 63
5
- 66
- 17
- 42
- 22
40
- 31
4
- 62
35
21
75
Total Area 133.00
2 X C02
Area
3.00
2.59
16.80
2.04
1.66
3.48
23.40
16.80
1.38
2.08
5.32
6.75
10.89
33.25
129.44
In contrast to the results using the Holdridge system presented,
preliminary estimates using the Box approach suggest that tropical
forests will decrease in area rather than increase (Smith personal
communication) .
To compare the results of Emanuel et al. (1985)y Smith et al.
(1990), and Prentice and Fung (1990), the vegetation classes they
used have been grouped into six broad vegetation types (Table 4,
Figure 2). Generally, there is agreement in terms of the sign of
the predicted vegetation change. Deserts, boreal forests, and
tundra typ«s decreased in areal extent, while grasslands, temperate
forests, and tropical forests generally increased in area under
double-CO2 climates. Across the seven different simulations and
VI - 11
-------
six vegetation categories, there are only four differences in the
direction of the predicted change. The GFDL-Qflux climate scenario
caused the smallest changes in global vegetation. Despite the
general agreement in the sign of the changes, there are significant
differences in the estimated magnitude of the vegetation changes.
For instance, estimates of the future distribution of boreal forest
differ by a factor of two.
VI - 12
-------
Table 4. Comparison of Global Vegetation Distribution in a
Double-CO2 Atmosphere using the Holdridge Classification System, as
applied by three different teams of investigators.
A. SMITH ET AL. 1990
Vegetation Type
Deserts
Grasslands/Shrubland
Temperate Forests
Boreal Forests
Tropical Forests
Tundra
TOTAL
B. PRENTICE AND FUNG
Deserts
Grasslands/Shrubland
Temperate Forests
Boreal Forests
Tropical Forests
Tundra
TOTAL
C. E MANUEL ET AL. 1985b
Desert
Grasslands
Temperate Forests
Boreal Forests
Tropical Forests
Tundra
Predictions
Current GFDL GFDL-QFlx GISS OSU
Area
20.80
26.60
13.10
26.70
38.50
9.30
135.00
T^ *-*. -»• J-« _r» »"» !• rx. _"J -r» ^^V*. rx
- 1 - 4 -15 - 7
35 - 9 18 9
5 14 17 7
-39 - 2 -19 -15
17 6 16 17
-66 - 9 -54 -49
UKMO
- 4
30
21
-43
21
-69
1990 Predictions
Current
Area
14.00
38.00
16.00
22.00
33.00
10.00
133.00
Predictions
Current
Area
20.40
27.62
27.10
18.57
30.99
6.69
GISS
Percentage
Change
- 62
- 25
27
- 14
54
- 44
Percentage
Change
6
18
- 3
-27
9
-46
TOTAL
131.37
VI - 13
-------
60
CE DIFFERENCE
K> *.
O O O
5 -20
u
x —40
a.
-60
1 1
1 '
-
-
1
r-^H
' '
GFOL GISS GFOL OFLX GISS
Cmanual Pf *ntlc«
•l ol. ana
1985 '«•«.
1990
OSU
-
n "
-
_
-
UKMO
Smith •( ot. '.990
-
-
— ., i 1
^n r-,n
-
_ _
-
GFOL GISS GFDL QFLX GISS OSU UKMO
Cmanuat Pr.ntlco Smith .t al. 1 990
al al. <"">
1985 Pu«9
1990
60
40
20
0
-20
-40
-60
BOREAL FOREST
GFOL GISS GFDL QFLX GISS OSU UKMO
ca Smith at al. 1990
•t ai.
1980
GRASSLAND/SHRUBLAND
CFDL GiSS GFDL OFLX GISS OSU UKMO
Cmanual Prantica
•t al. ana
1986 f""9
1990
Smith «t al. 1990
-------
The results presented above in effect are summaries of global
vegetation before and after a climate change caused by a doubling
of C02 concentrations. What is not indicated by these numbers is
the amount of land that changed from one vegetation type to
another. This could have a significant impact on the global carbon
cycle because of the transient release of carbon to the atmosphere
from vegetation dieback (see section 3.2 in King et al. 1990). In
the Prentice and Fung (1990) scenario, 60% of the vegetated
landscape on the globe changed vegetation type. In the Smith et
al. (1990) scenarios, 16% (GFDL-QFlux) to 56% (UKMO) of the earth's
land surface changed vegetation type (Table 5, Figure 3). Such
predicted changes would have significant impacts on biodiversity
(discussed next), water resources, forest resources, agriculture,
and land management.
Table 5. Areal extent of land on the globe changing
vegetation cover under double-CO2 conditions as estimated
using the Eoldridge system (Smith et al. 1990) .
Climate Amount of Land Changing Perc. of Land
Vegetation Cover Surface Changing
(10*6 km2)
GFDL 65.24 48.34
GFDL-QFlux 21.51 15.94
GISS 60.36 44.72
OSU 53.73 39.81
UKMO 76.21 56.46
Total Land Area 134.97
VI - 15
-------
Discussion of Vegetation Scenario Results
The fact that the vegetation scenarios are similar (in terms of the
sign of the change) using the Holdridge approach is not surprising,
given that the GCM scenarios on a global basis predict warmer and
wetter conditions than present (see Appendix A) . In general, a
point in climate space on the Holdridge diagram (Figure 1) can be
expected to.move down and to the right under double-CO2 climate
conditions. More detailed results from Smith et al. (1990) not
presented here in fact show that this is often the case. For
instance, areas of current tundra vegetation are estimated to
change in the future to forest tundra or boreal forest. Tropical
seasonal forest regions that change type tend to change to tropical
rain forest.
Estimates of current area of the six vegetation types can differ
substantially between investigators. That the Prentice and Fung
(1990) areas are different is not surprising, as they used a
different vegetation classification system from the original
Holdridge system. However, Emanuel et al. (1985b) and Smith et al.
(1990) used the same basic vegetation units in their analyses, and
they aggregated the units in the same way to obtain the areas for
the vegetation types presented in Table 4. For instance, estimates
of boreal forest area differ by over 8 x 106 km2. A possible reason
for this is; that Emanuel et al. (1985a,b) used monthly temperature
values to calculate annual biotemperature, whereas Smith et al.
VI - 16
-------
(1990) used daily values. Consequently, differences in the percent
changes listed in Table 4 are in part due to differences in how the
current vegetation areas were calculated, and how the individual
vegetation units were aggregated to form the vegetation units used
in the table;.
The other reason the estimated changes in vegetation differ is that
there are Icirge differences in the climate scenarios used to drive
the vegetation models (see Appendix A).
Conclusions and Research Needs
Global climate change on the order currently projected by GCMs will
cause large changes in the distribution of global vegetation.
However, because of differences in GCMs, climate scenarios and
vegetation models, and the simplifying assumptions on which the
vegetation models are constructed, the specifics of future, steady-
state, vegetation patterns are highly uncertain. One of the
weaknesses of current global vegetation models is that they are
based on correlations between climate and vegetation, correlations
which may not persist under altered climate conditions.
Furthermore, current models do not incorporate soils or the
potential effects of CO2 fertilization on vegetation. Land-use is
not treated by many of the models.
In addition, current global vegetation models can not be used to
VI - 17
-------
estimate transient vegetation dynamics. The ability to predict the
transient response of global vegetation to climate change is
essential for estimating biospheric feedbacks and regional
sensitivity to climate change in the next century, and for
evaluating the effectiveness of various mitigation strategies
(e.g., reforestation).
There is much work to be done in producing a dynamic and realistic
global vegetation model. The following areas of research need
attention in order to develop such a model:
o Development of a plant life-form classification system that
can form the basis of a global vegetation model. Life-forms
grouped together should have common autoecological characters,
such as physiognomy, seed dispersal and response to
macroclimate and soils.
o Determination of the ecological mechanisms through which
climate controls the distributions of these life-forms.
o Incorporation of disturbance and migration into global
vegetation models.
o Production of global vegetation maps based on remotely sensed
data to calibrate and validate vegetation models.
o Generation of digitized databases of soil texture and nutrient
availability at a resolution appropriate for global modeling.
o Calculation of the effects of CO2 enrichment on ecosystem
water-use efficiency.
VI - 18
-------
V.
Figure 3. Areas (black) in which predicted future vegetation is
different from current vegetation using the UKMO
double-C02 climate scenario (Smith et al.1990).
a. Global changes in vegetation cover.
b. North American changes in vegetation cover.
VI
-------
LITERATURE CITED
Box, E.G. 1981. Macroclimate and Plant Forms: An Introduction to
Predictive Modeling in Phytogeography. Dr. W. Junk Publishers, The
Hague.
Davis, M.B. 1986. Climatic instability, time lags, and community
disequilibrium. In: J. Diamond and T.J. Case, eds., Community
Ecology, Harper and Row, N.Y.
Emanuel, W.R., H.H. Shugart, and M.P. Stevenson. 1985a. Climatic
change and the broad-scale distribution of terrestrial ecosystem
complexes. Climate Change 7:29-43.
Emanuel, W.:R. , H.H. Shugart, and M.P. Stevenson. 1985b. Response
to comment: Climatic change and the broad-scale distribution of
terrestrial ecosystem complexes. Climate Change 7:457-460.
Holdridge, L.R. 1947. Determination of world formations from
simple climatic data. Science 105:267-268.
King, G.A. , J.K. Winjurn, R.K. Dixon, and L.Y. Arnaut, eds.
Submitted, 1990. Response and feedbacks of forest systems to
global climate change. Report prepared for Environmental Protection
Agency, Washington, DC. 204 pp.
Koppen, W. 1936. Das Geographische System der Klimate. In:
Handbuch der Klimatologie, by W. Koppen, and R. Geiger, 1C, Gebr
Borntraeger, Berlin, 46 pp.
Koppen, W. 1900. Versuch einer Klassification der Klimate,
vorzugsweise nach ihren Beziehungen zur pflanzenwelt. Geographische
Zeitschrift. 6:593-611, 657-679.
Koppen, W. 1918. Klassification der Klimate nach Temperatur,
Neidenschlag und Jahreslauf. Petermanns Geogr. Mitt. 64:193-203,
243-248.
Lashof, D. 1987. The role of the biosphere in the global carbon
cycle: eveiluation through biospheric modeling and atmospheric
measurement. Ph.D. Dissertation, Energy and Resources Group,
University of California, Berkeley, CA.
Leemans, R. 1990. Possible changes in natural vegetation patterns
due to a global warming. Publication Number 108 of the Biospheric
Dynamics Project, International Institute for Applied Systems
Analysis.
Leemans R. and Prentice, I.e.-1990 (In prep.) Possible changes in
natural vegetation patterns using GCM climate scenarios.
Prentice, K.C., and I.Y. Fung. Bioclimatic simulations test the
VI - 20
-------
sensitivity of terrestrial carbon storage to perturbed climates.
Nature. In press.
Prentice, K.C. Bioclimatic distribution of vegetation for GCM
studies. J., Geo.Res. In press.
Smith, T.M., R. Leemans, W. Cramer, I.e. Prentice, A.M. Solomon,
and H.H. Shugart. Submitted. Simulating changes in the
distribution of global terrestrial vegetation under C02-induced
climate change: Comparison among scenarios based on general
circulation models. Climate Change.
Tans, P.P., I.Y. Fung, and T. Takahashi. 1990. Observational
constraints on the global atmospheric CO2 budget. Science
247:1431-143.8.
Webb, T. 1986. Is vegetation in equilibrium with climate? How to
interpret late-quaternary pollen data. Vegetatio 67:75-91.
VI - 21
-------
APPENDIX A: CLIMATE SCENARIOS
In order to determine the possible impacts of global climate change
on vegetation, quantitative estimates are needed of the magnitude
of trace gas induced climatic change. General Circulation Models
(GCMs) of the earth's atmosphere are the only tool available for
making quantitative estimates of climate variables on a global
scale. GCMs are complex computer models based on fundamental
principles of physics and thermodynamics (e.g. Schlesinger 1988).
In the past decade, several GCMs have been used to simulate the
possible effect on global climate of a doubling of CO2
concentrations. Results of simulations of four different GCMs
(Table Al) are widely available for use by researchers and were
used by Smith et al. (1990) in generating their Holdridge
vegetation scenarios. Some of the overall conclusions on future
climate change and uncertainties in the GCMs are discussed in this
appendix, as well as the general methodology for creating the
future climate scenarios.
TABLE Al GENERAL CIRCULATION MODELS USED TO GENERATE DOUBLE C02
CLIMATE SCENARIOS
Model Name Reference
Geophysical Fluid Dynamics Model (GFDL) Manabe and Wetherald
1987
Goddard Institute for Space Studies (GISS) Hansen et al. 1988
Oregon State University (OSU) Schlesinger and Zhao
1989
United Kingdom Meteorological Office (UKMO) Mitchell et al.
1989
VI - 22
-------
The mosjt recent results suggest that double CO2 conditions
will increase global temperatures from 1.9 to 4.4°C (Schlesinger
and Zhao 1989, Washington and Meehl 1989, Mitchell and Warilow
1987, Mitchell et al. 1989, Wetherald and Manabe, In prep.).
Global precipitation could increase from 7.8% to 11% (Smith and
Tirpak 1989) . Regional and seasonal predictions are more variable
between modeils, sometimes differing even in the predicted direction
of the change (e.g. whether precipitation increases or decreases in
a region). The speed of these changes depends upon the rate of
increase of CO2 concentrations in the atmosphere. Current
estimates suggest that CO2 concentrations will double their pre-
industrial level between 2055 and 2080 without any effort at
stabilizing emissions (Lashoff and Tirpak 1989). Considering both
the magnitude and rate of change, trace gas induced climate change
could be greater than any climate change during the last 15,000
years.
Although GCMs are expansive, complicated and computationally
intensive computer programs, they are far from perfect in
simulating present climate (Crotch 1988, Dickenson 1989, Neilson et
al. 1990). For instance, the GFDL model predicts winter
temperatures that are l.8°C cooler than those observed for North
America. The GISS model simulates summer temperatures 3.1°C cooler
than those observed for this same region. Precipitation estimates
are even more problematic. Four major problems exist with how the
models simulate climate: 1) poor simulation of cloud processes; 2)
VI - 23
-------
inadequate atmosphere-ocean coupling; 3) overly simplistic
biosphere; cind 4) poor spatial resolution. Consequently, the
accuracy of future climate simulations is considered highly
uncertain. Major research efforts are underway to improve GCMs
(the Department of Energy's ARM and CHAAMP programs) and to
determine how well they simulate past climates as a validation
technique (COHMAP 1989).
Despite the limitations of GCM output, especially their poor
spatial resolution, they remain the only tool that can provide
spatially distributed estimates of future climates based on the
principles of atmospheric physics. The challenge for researchers
estimating the effects of climate change on some component of the
biosphere is to generate reasonable scenarios of future climate for
input into their models based on GCM output. The most common
technique is to calculate the difference between (or for
precipitation, the ratio of) the double CO2 estimate for a
particular gridpoint and the control estimate (current conditions)
for that same gridpoint (Parry et al. 1987, Smith and Tirpak 1989,
ICF 1989). These differences are then applied to the corresponding
historical weather data (often the mean value for the 1951-1980
time period) . For instance, if a GCM estimates that July
temperatures will be 2°C warmer for a particular gridpoint, then
the mean July temperature for a weather station in the grid box is
increased by 2° to estimate future July temperatures for that
particular location.
VI - 24
-------
LITERATURE CITED
Cooperative Holocene Mapping Project (COHMAP). 1988. Climatic
changes of the last 18,000 years: Observations and model
simulations. Science 241:1043-1052.
ICF, Inc. 1989. Summary Report: Scenarios Advisory Meeting
Report, 31 Aug-1 Sept, Boulder, CO. Report for U.S. Environmental
Protection Agency, Office of Policy, Planning and Evaluation,
Washington, DC, 14 pp.
Neilson, R.P., G.A. King, R.L. DeVelice, J. Lenihan, D. Marks, J.
Dolph, B. Campbell, and G. Click. 1989. Sensitivity of ecological
landscapes and regions to global climate change. U.S.
Environmental Protection Agency, Environmental Research Laboratory,
Corvallis, OR.
Parry, M. , T. Carter, N. Konijin, and J. Lockwood. 1987. The
impact of climatic variations on agriculture: Introduction to the
IIASA/UNEP Case Studies in Semi-Arid Regions. Laxenburg, Austria.
Schlesinger, M.E. and Z.C. Zhao. 1989. Seasonal climatic change
introduced by doubled CO2 as simulated by the OSU atmospheric
GCM/mixed-layer ocean model. J.Climate 2:429-495.
Schlesinger, M.E., ed. 1986. Physically-Based Modelling and
Simulation of Climate and Climatic Change (Part 1). Proceedings of
the NATO Advanced Study Institute, Erice, Italy.
Smith, J.B., and D. Tirpak, eds. 1989. The Potential Effects of
Global Climate Change on the United States. Report to Congress.
EPA-230-05-89-050, U.S. Environmental Protection Agency,
Washington, DC 413 pp.
Washington, W.M. and G.A. Meehl. 1989. Climate sensitivity due to
increased CO2: experiments with a coupled atmosphere and ocean
general circulation model. Climate Dynamics. 4:1-38
Wetherald, R.T., and S. Manabe. 1988. Cloud feedback processes in
a general circulation model. J. Atmos. Sci. 45:1397-1415.
VI - 25
-------
TOWARD A RULE-BASED BIONE MODEL
Prepared by:
Ronald P. Neilson
Oregon State University
U.S. EPA Environmental Research Lab
200 S.W. 35th Street
Corvallis, Oregon 97333
George A. King
NSI Technology Services Corporation
U.S. EPA Environmental Research Lab
200 S.W. 35th Street
Corvallis, Oregon 97333
Greg Koerper
Oregon State University
U.S. EPA Environmental Research Lab
200 S.W. 35th Street
Corvallis, Oregon 97333
VII - i
-------
Toward A Rule-Based Biome Model
Current projections of the response of the biosphere to global climatic change
indicate as much as 50% to 90% spatial displacement of extratropical biomes.
The mechanism of spatial shift could be dominated by either 1) competitive
displacement of northern biomes by southern biomes, or 2) drought-induced
dieback of areas susceptible to change. The current suite of global biosphere
models cannot distinguish between these two processes, thus determining the
need for a mechanistically based biome model. The first steps have been taken
toward the development of a rule-based, mechanistic model of regional biomes
at a continental scale. The computer model is based on a suite of empirically
generated conceptual models of biome distribution. With a few exceptions the
conceptual models are based on the regional water balance and the potential
supply of water to vegetation from two different soil layers, surface for
grasses and deep for woody vegetation. The seasonality of precipitation
largely determines the amount and timing of recharge of each of these soil
layers and thus, the potential mixture of vegetative life-forms that could be
supported under a specific climate. The current configuration of rules
accounts for the potential natural vegetation at about 94% of 1211 climate
stations over the conterminous U.S. Increased temperatures, due to global
warming, would 1) reduce.the supply of soil moisture over much of the U.S. by
reducing the volume of snow and increasing winter runoff, and 2) increase the
potential evapotranspiration (PET). These processes combined would likely
produce widespread drought-induced dieback in the nation's biomes. The model
is in an early stage of development and will require several enhancements,
including explicit simulation of PET, extension to boreal and tropical biomes,
a shift from steady-state to transient dynamics, and validation on other
continents.
-------
TOWARD A RULE-BASED BIOME MODEL
Ronald P. Neilson1
George A. King2
Greg Koerper1
Introduction
Trace-gas induced global climatic change has been projected to
potentially shift the world's major biotic regions by hundreds of kilometers
(Smith and Tirpak 1989). Under one scenario as much as 55% of the world's
terrestrial vegetation could change to a different type (Smith et al. in
prep., King et al. in prep.)- Such changes could produce complicated
feedbacks to the global climate, possibly exacerbating the 'greenhouse
effect', and would certainly produce significant ecological and economic
upheavals. These projections are based, in part, on the Holdridge life-zone
approach that correlates the distribution of major vegetation types to annual
precipitation, potential evapotranspiration and biotemperature (Holdridge
1967). Another global biome model (Box 1981), more detailed in climate
parameters, is; also correlational. The two approaches generally agree on the
direction of raid- to high-latitude vegetation changes under 2xC02, but
disagree on the sign of the change in the tropics. The tropical forests
expand under the Holdridge model, but contract under the Box model (Smith et
al. in prep., King et al. in prep.). Both approaches assume a steady state,
that is, they cannot simulate the transition of biomes from one type to
another. The accuracy with which these approaches predict current vegetation
ranges from <50% to about 77% (Prentice in press, Stephenson 1990). These
models also do not simulate biosphere-atmoshere feedbacks. These and other
limitations suggest the need for a more mechanistic approach to biome
modeling, one that can be incrementally developed to incorporate transient
'Oregon State University, U.S. EPA Environmental Research Lab, 200 S.W.
35th St., Corvallis, Oregon 97333
2NSI Technology Services Corporation, U.S. EPA Environmental Research
Lab, 200 S.W. 35th St., Corvallis, Oregon 97333
VII - 1
-------
behavior, ecosystem productivity, trace gas emissions, and disturbance regimes
(e.g. wildfire).
A new approach to biome modeling is being developed that is more closely
related to physiological plant processes. The intent is to simulate distinct
life-forms or physiognomies (Beard 1978), mixtures of which produce different
biomes. The model was constructed from the premise that the climate is the
principal determinant of global vegetation distribution and that variations in
topography, soils, disturbance regimes and biotic interactions modify these
distributions (Allen and Starr 1982, Neilson et al. 1989, O'Neill et al. 1986,
Stephenson 1990, Vankat 1979, Woodward 1987). The construction of this
preliminary version of the model represents a partial test of this premise.
Successful classification of vegetation based exclusively on climatic
information willl be viewed as supportive of the premise. The present approach
is limited to steady-state, potential natural vegetation. Future enhancements
should include land-use considerations. Model development has been limited to
the conterminous U.S. at this stage, but will be extended to global vegetation
in future versions.
Conceptual Development
New developments in biogeography are providing a mechanistic
conceptualization of the biosphere (Bryson 1966, Neilson and Wullstein 1983,
Neilson 1986, Neilson 1987, Neilson et al. 1989, Stephenson 1990, Woodward
1987). The model described here is based on mechanistic, conceptual models
described by Neilson et al. (1989). These resulted from transect analyses of
over 1200 weather stations in the conterminous U.S. (Figs. 1, 2) and over
7,000 USGS gaging stations (Quinlan et al. 1987, US West 1988a, US West
1988b). The approach relates the seasonality of temperature, precipitation
and runoff patterns to the physiological requirements of plants during
different parts of their life-cycles and seasonal cycles. Details of the
approach and results are published elsewhere (Neilson et al. 1989).
The continental transects of climate and runoff revealed
regionalpatterns of climate and runoff seasonality that coincide with the
boundaries of the major biomes of the conterminous U.S. (Fig. 1) (Neilson et
al. 1989). These generalizations form the basis for the model development
described he:re and can be cast as rules for prediction of the occurrence woody
VII - 2
-------
•f" '95" '00° 95
Northern
£ Transect
Biotic Regions
'"•'~5-'4ii California Vegetation
PNW Forests
Great Basin Sage-Steppe
Great Basin Desert Scrub
Northern Rocky Mountain Forests
Southern Rocky Mountain Forests
Southwestern Deserts
Short Grass Prairie
Tall Grass Prairie
South Texas Savanna
Northern Hardwoods
Eastern Deciduous Forests
Southeastern Forests
Subtropical Forests
Figure 1. Principle biomes of the conterminous U.S. (simplified from Dice
1943, Kuchler 1964). The locations of two transects of the HCN network of
weather stations (Fig. 2) are shown (details of these are described elsewhere,
Neilson et a.l. 1989) .
VII - 3
-------
UHASS UKAI'lllCS WINUUW
HISTORICAL CLIMATOLOGY NETWORK SITES
Figure 2. Locations of weather stations in the HCN network (Quinlan et al.
1987). Conceptual models of biomes were developed from a sub-sample of these
stations using transect analyses (Neilson et al. 1989). The full set of
stations was used to calibrate the quantitative model presented here.
-------
vegetation, grasslands and deserts. Woody vegetation in the U.S. appears to
receive sufficient winter precipitation to recharge a deep soil reservoir of
water. Deep soil water is apparently required to balance the potential
evapotranspiration (PET) during the growing season. If the amount of winter
precipitation is large, a region will support a closed forest. Intermediate
amounts should support a lower stature or open forest (e.g., savanna or
pinyon-juniper woodland), while lesser amounts should support a shrubland
(Neilson et al. 1989). The amount of winter precipitation (i.e., deep soil
water) required is, a function of growing season PET and rainfall less runoff.
Regional runoff patterns suggest that the deep soil reservoir is virtually
depleted each year by transpiration and runoff (Ibid.).
These observations are consistent with current theory and plant
physiology (Neilson et al. 1989, Stephenson 1990, Woodward 1987). If excess
water were available in deep soil layers at the end of a growing season and
ecosystems were not energy or nutrient limited, they should increase biomass
and leaf area in subsequent seasons, thus increasing the rate of withdrawal of
deep soil water. If leaf area over a landscape were so high that the rate of .
water withdrawal completely depleted deep soil water, then plants would die or
leaves and branches would be sloughed (Woodward 1987). Leaf area would be
reduced, as would the rate of withdrawal of deep soil water. Thus, in theory,
ecosystems should be in dynamic equilibrium (steady state) with the regional
water balance by virtue of these intrinsic feedback processes.
While maintenance of woody vegetation appears to require deep soil
water, maintenance of grasses appears to require sufficient surface soil
moisture during their growth and reproduction, usually the spring to summer
months. Mixtures of the two life-forms can occur if there is sufficient
moisture in the appropriate soil layers during the growing season and if
interference by one life-form does not preclude the existence of the other.
Grasslands can occur if the woody canopy is sufficiently open to allow high
light penetration to the soil surface and if there is a sufficient supply of
surface soil moisture during the active growing season of grasses, generally
the spring (Neilson 1987, Neilson et al. 1989). This implies a sufficiently
low level of deep soil water (i.e., winter precipitation), such that a closed
forest would be precluded from the site. If spring rains are accompanied by
high mid-summer rains, surface soil will remain moist and a large stature
VII - 5
-------
grassland should be supported, e.g., the Tall Grass Prairie. If both winters
and springs are quite dry, maintenance of either woody vegetation or grasses
is hindered and a desert may be expected. High mid-summer rains in such a
desert area ca:n potentially support a low-stature grassland, as in the
Chihuahuan and Sonoran Deserts of the Southwest (Neilson 1986, Neilson 1987).
A few biome boundaries in the U.S. appear to be directly controlled by
cold temperature rather than water balance. The most prominent of these in
the U.S. are 1) the boundary of the temperate forests with the boreal forest
(Burke et al. 1976), and 2) possibly the boundary between the Southeast pines
and hardwoods with the oak-hickory forest to the north (Neilson et al. 1989).
The current configuration of our biome model does not address these
boundaries.
The model is being developed in stages that will progress from empirical
and correlational to simulation of water balance and thermal constraints. The
empirical rules described here are generally adequate to define most dominant
vegetation types in the conterminous U.S. In our model, threshold values of
precipitation amounts during different seasons determine transitions between
different life-forms, as described above. This configuration addresses the
supply side of the water balance. The demand function, PET, is assumed to be
fixed and is not directly factored into the rules. Given the assumption of
fixed PET, the current configuration of the rules cannot be used to assess
future climate effects. This initial version of the model was constrained by
the unavailability of data for the physically based calculation of PET at
large spatial scales. PET will be incorporated in later versions.
The rules allow mixtures of life-forms, such as trees and grasses, if
the canopy i£i open. These mixtures can produce complicated biotic
interactions and disturbance regimes. For example, the eastern deciduous
forest is characterized by both wind and fire disturbances with wind perhaps
being the mo.-st important (Runkle 1985). However, the adjacent prairie
supports a natural, high-frequency fire regime (Abrams et al. 1986, Vankat
1979). If the tree canopy is open, such that a well-developed grass layer
forms, woody vegetation is at risk from fire. The presence of high grass
biomass also places woody vegetation at risk from herbivores and from
competition between shrub seedlings and rapidly growing grasses for both light
and water. Under these conditions, strictly climatic rules for prediction of
VII - 6
-------
biome physiognomy will not suffice. Secondary influences from disturbance and
biotic interactions are important and can remove one or more climatically
favored life-forms (Abrams et al. 1986, Vankat 1979). Therefore, under
certain climatic conditions where specific mixtures are favored, secondary
-------
Cloverdole, California
JFMAMJJASOND
o
0)
L_
O
OJ
Q.
E
Figure 3. Example of rule-based partitioning of the seasons. Precipitation,
open circles; Temperature, open squares; Potential Evapotranspiration, closed
circles. Vertical, dotted lines indicate the locations of temperature
thresholds usied in the model to delineate seasonal transitions. Temperature
thresholds are used to define the beginning and ending of seasons for each
weather station. Within each season precipitation amounts are accumulated for
seasonal totals or examined in terms of precipitation per month. Different
rainfall amounts within the different seasons are used by the rule-base to
differentiate between different life-forms that would potentially be supported
by that climate.
VII - 8
-------
precipitation during any one season would not represent precipitation amounts
over the same period of time at all map locations.
The beginning and ending of winter are defined from both hydrologic and
biotic perspectives (Fig. 3). In temperate zone summers PET is generally high
enough that incident precipitation does not infiltrate to deep soil layers,
but either runs off directly or is evapotranspired from vegetation surfaces
and surface soil layers (Major 1963, Thornthwaite 1948, Thornthwaite and
Mather 1955). In cooler months PET is low enough that incident precipitation
can infiltrate to deep soil layers. Accumulated snow will also infiltrate, or
run off, upon snowmelt. Thus, we define a temperature threshold above which
precipitation cannot reach deep soil layers and below which it can. The
infiltration p
-------
(ElOMd formtt \
ABr )
(Cloud For««t ^ (Clattt POP«
Bromli.f I I (sufttroBlei
J V aroMlMO
f ShruUliml J I
( Pygiy For«t J f (SSw-t CPUI* y
Figure 4. Flow chart of rule-base. Monthly mean temperature and monthly
total precipitation are input for each station. Seasons are defined as in
Fig. 3 and precipitation amounts are linearly interpolated for each season
from the monthly data. The rule-base is a dichotbmous key that classifies
each station as to the potential natural vegetation based on climate. The
enclosed rules (11 and 12, dashed lines) indicate implied, secondary rules
(see discussion in text), whereby disturbance regimes and biotic interactions
are inferred to constrain one or more life-forms from a dominant expression in
the mixture of life-forms.
VII - 10
-------
to infiltrate to deep soil layers at some time during the winter with the
remainder being returned to the atmosphere (evaporation, sublimation) or
routed to surface waters (runoff). However, during the spring and summer
months the supply of water (precipitation) and demand for water
(evapotranspiration) are essentially simultaneous when viewed from a monthly
time-step. Therefore, precipitation thresholds during spring and summer are
defined in units of mm/mo.
The rules (Fig. 4) are equivalent to a dichotomous key with simple
yes/no decisions determining the trajectory through the rules and the ultimate
declaration of biorae type for a station (or cell if applied to a full grid).
The flowchart is arranged such that all winter decisions are in the first
column of diamond shapes, followed by a column of spring decisions and a
column of summer decisions. The decisions enclosed by dashed lines (rules 11,
13) are implied rather than explicitly coded rules in the current
configuration. The implied rules arise from assuming the application of
secondary biome rules for biotic interaction and disturbance regimes. The
bubbles to the left of the flowchart (Fig. 4) are the simplified
classifications applied at the termination of the adjacent rule. They
indicate the influence of the secondary, implied rules accounting for fire and
biotic inters.ctions. For example, a mixture of open-canopy, deciduous trees
with tall grass prairie is directly classified as tall grass prairie.
Similarly, all Pygmy forests (pinyon-juniper woodland) with or without grass
are classified as Pygmy forest. Future versions of the model will retain all
such mixture:; and distinctions.
The calibrated threshold separating closed forest from non-closed forest
is 375 mm of winter precipitation (rule 2, Fig. 4). The amount of summer
precipitation per month (rule 3) separates broadleaf from conifer forests.
The physiological inference is that broadleaf trees cannot withstand the
degree of mid-summer vapor-pressure stress across the leaf surface that
conifers can withstand (Marshall and Waring 1984, Neilson et al. 1989, Running
1975, Waring and Franklin 1979).
The current configuration of the model becomes inadequate as the
stations approach sub-tropical climates. Rule 4 (Fig. 4) is a temporary proxy
for a set of rules that will deal with sub-tropical climates. As one
approaches the Florida peninsula, winters become ever shorter and eventually
VII - 11
-------
non-existent as defined by the temperature thresholds. Thus, the total amount
of precipitation during the winter at these stations is not sufficient to
support a forest, a clear artifact of the rules. As the growing season
becomes longer, the infiltration period of winter becomes shorter. Under
these situations, infiltration can still occur .if the amounts of rainfall more
than compensate for a high PET. This does occur in forested areas within the
subtropical climates of the Southeast U.S. and elsewhere (Neilson et al.
1989). However, implementation of such a rule must await a future version of
the model that: incorporates physically-based PET calculations. A workable
place-holder is a threshold for annual precipitation of about 1100 mm, given
that winter precipitation is below a closed-forest threshold (rule 4). This
correctly classifies closed-forest stations in the far southeastern U.S., yet
does not interfere with appropriate winter, rule-based decisions in more
northerly latitudes.
After partitioning the high winter precipitation, i.e., closed-forest,
stations, rule 5 selects sites with winter precipitation less than 80 mm.
These stations are ultimately classified as desert or grassland based on
spring precipitation (rule 6, 20 mm/mo) and tall or short grass based on
summer precipitation (rule,?, 70 mm/mo).
The remaining stations will be classified as either shrub1ands, open
forest (savanna), or grasslands potentially supporting woody species.
Shrubland and open forest are distinguished depending on whether winter rains
are above or below 305 mm (rules 9, 11, 13, 14). Each type (shrub or tree)
can be broadleaf or microphyllous (small leaf), depending on the level of
summer rains (rules 10, 12), reflecting again the inferred vapor pressure
gradient across the leaf surface. If summer rains are sufficient to support a
broadleaf type (rule 10), the result is either a broadleaf, tall-grass savanna
or tall-grass; shrub savanna. In either case, secondary rules are invoked to
classify thes;e sites as tall grass prairie, with the principle mechanism for
grass dominance assumed to be fire.
Given sufficient winter rains to support shrubs or open forest (rule 5),
but less than 70 mm/mo of summer rainfall (rule 10), only a xeromorphic leaf
o
structure can be supported. The tree form will be conifer and could be a
xeric conifer savanna, such as ponderosa pine or a smaller stature, but more
dense pygmy forest of pinyon pine and juniper. In either case the conifer
VII - 12
-------
leaf area is comparatively low. If only the shrub form can be supported (rule
14), the shrubs would be microphyllous. If summer rains are above 28 mm/mo
and spring rains are above 20 mm/mo (rule 8), then a well-developed short
grass community can be supported. Under these conditions, secondary rules are
again invoked to remove the woody component (rule 13) and the communities are
classified as short grass. Below 28 mm/mo of summer rains, the grassland is
of lower bioma.ss, weakening the secondary rules, resulting in either a shrub
or pygmy forest savanna depending on winter rains (rule 14).
Model Calibration
The modeil parameters were adjusted to provide an optimal fit to the
specified biome boundaries. This was accomplished iteratively by visual
examination of the residuals and calculations of percent correct
classification. Two classes of parameters require calibration: 1) thermal
constraints controlling the beginning and ending of seasons, and 2)
precipitation thresholds within seasons for different life-forms. These must
initially be calibrated in tandem with sensitivity of both classes being
examined. However, once the thermal thresholds are set, primary calibration
is with the precipitation thresholds. Future versions of the model that
incorporate PET will use the PET less precipitation estimates in conjunction
with thermal information to control seasonal timing and length.
Most rules exhibit one or more 'balance' points. That is, increasing or
decreasing the value of a parameter may continue to increase the correct
classification of a particular biome, but there will be a point above (or
below) which further adjustment occurs at the expense of correct
classification in other biomes. The parameters were adjusted close to these
'balance' points. Some parameters were clearly more sensitive than others,
however, we have not attempted a formal sensitivity analysis.
The site-specific accuracy of the calibration was 79% with the highest
success being 94% in the eastern U.S. forests and the lowest being 0% for the
11 grassland stations in the central valley of California (Table 1, Fig. 5).
The current model does not subdivide the eastern forest into the three
different sub-sections. Undoubtedly, when these rules are implemented, errors
in classification will occur near the ecotones. A few sites near the border
VII - 13
-------
Table 1. Summary of classifications, n is the number of weather stations, C.F. -
conifer forest;, Srb. - shrubland; S.D. - southwest deserts; P.J. - pinyon-juniper;
S.G. - short grass; T.G. - tall grass; B.F. - broadleaf forest. Entries in the
table are percentages. The bold numbers correspond to the groupings of similar
biomes. The italics highlight the correct classification category.
Conifer Forests
Northwest
Southwest
n
91
58
33
C.F. Srb. S.D. P.J. S.G. T.G. B.F.
82
91
67
9
2
21
8
5
12
1
2
California Grassland 11 45 36 18 0
Shrubland 78 12 81 1 5 1
Northern Great Basin 63 14 79 6
Southern Great Basin 15 87 7 7
Southwest Deserts 34 12 88
Mixed Forest 114 15 53 2
Northern Rockies 36 28 39
Southern Rockies 78 9 59 5
Short Grass Prairie 81(122)3 2 2
Central 67(108) 3 1
South Texas savanna 14 7
Tall Grass Prairie
Broadleaf Forests
Northern Hardwood
Deciduous Forest
Southeast Forest
Sub-tropical Forest
Total
Total (drop Rocky Mts.)
Total (accept: Rockies,
California,, and mixed-
grass as correct)
242(201)
560 1
129 6
234
189
8
1211 79(82)
1097 85(88)
1211 94
<1
12
11 16 2
11 22
12 13 3
84(89) 11
93(95) 3
43 50
17(0) 70(84) 13(16)
4
8
6
94
86
94
100
88
'Numbers in parentheses are the sample sizes (n) and percentages if the
northern mixed grass prairie of Kvichler (1964) is considered to be short grass
prairie as determined by the rule-base. The current configuration of rules
does not explicitly attempt to distinquish mixed-grass as intermediate between
short and tall grass.
VII - 14
-------
Figure 5. Rule-based classification of the HCN network sites (Fig. 2),
overlain on the pre-classification of U.S. biomes (Fig. 1).
-------
between the northern hardwoods and the boreal forest were classified as
conifer, a reasonable occurrence. The mixed forests of the Rocky Mountains
were only predicted at 17% accuracy (28% northern Rockies, 12% southern
Rockies (Table 1). The next lowest success was 70% for the tall grass prairie
with the remaining biomes being predicted at better than 80% accuracy.
Closer examination of the residuals (Table 1, Fig. 6) reveals that the
model is more accurate than these indications, since pre-classification error
is incorporated in the results. The California grasslands were predicted as
conifer forest: (45%), shrubland (36%) and pinyon-juniper (P-J) woodland (18%) .
Given the wet winters and very dry summers, these predictions may be quite
reasonable. The conifer forest and P-J woodland stations cluster in the north
end of the valley and at the boundaries with the neighboring conifer forests.
These could be: accurate classifications given the potential inaccuracies of
the pre-classification biome map. A shrubland classification for the
remaining sites may also be acceptable. Recall that the rule-
basedclassifications have been simplified to indicate only the dominant life-
form, even if a mixture is actually being selected (Fig. 4). The shrubland
classification in the current configuration of rules is actually a mixture of
shrub and short grass life-forms (rule 14, Fig. 4). Climatically, the mixed
shrub-grass classification for the central valley is quite reasonable;
however, the relative mix of shrubs and grass is likely inaccurate in that the
model does not yet account for the unusually long spring period over which
grass biomass would build. Given sufficient grass biomass, the natural and
anthropogenic influence of fire could have rendered the valley more of a
monotypic grassland (Vankat 1979). Other possibilities for a lack of shrubs
in the central valley include 1) the presence of fine surface soils that might
hinder percolation of deep soil water, and 2) the lack of sufficient summer
rains that are usually necessary for the establishment of shrubs (Neilson
1986).
Both the northern and southern Rocky Mountain regions were poorly
predicted (Table 1, Fig. 6). The over-simplistic pre-classification into one
category each, conifers in the north and pinyon-juniper in the south, does not
capture the vertical zonation of vegetation in these regions. Most of the
weather stations in this highly dissected terrain occur in valley bottoms that
are often dominated by shrubland or grassland (Kiichler 1964, Vankat 1979).
VII - 16
-------
j 5 s e r t
5; h c r t 3 :
nil
p foresl
..-•.- .- •-_.:• i- annp
Figure 6. Residuals from the rule-based classification (Fig. 5). Only those
sites in Fig. 5 that were not correctly classified according to the pre-
classification (Fig. 1) are displayed.
-------
The classifications for the northern Rocky Mountains include conifer (28%),
pinyon-juniper (11%), shrubland (39%), and short grass (22%). The southern
Rocky Mountains are wanner, more xeric in the winter and wetter in the summer
with conifer (9%), pinyon-juniper (12%), shrubland (59%), desert (5%), short
grass (13%), a.nd tall grass (3%). We examined most of these stations
individually a.nd concluded that these classifications are generally accurate.
If the Rocky Mountain, rule-based classifications are accepted as correct, the
model performance shifts from 79% to 90% overall. If the Rocky Mountains are
removed from the classification the performance is 85% (Table 1).
A third set of residual mis-classifications stems from 17% of the tall
grass prairie stations being classified as short grass prairie (Fig. 6, Table
1). These stations occur in a region in the north plains designated by
Kiichler (1964) as mixed grass, a life-form for which we have not written
arule. Thus, our pre-classification as either short or tall grass is
arbitrary. If these stations are accepted as a correct classification, the
performance of the model in the tall grass shifts from 70% to 84% and the
overall rating increases from 79% to 82% (Table 1). If the mixed-grass is
accepted, but both the Rockies and the Central Valley are excluded, the
accuracy of the model is 88%. If both the Rocky Mountains and the mixed grass
classifications are accepted, the overall accuracy shifts to 93%. If the
Central Valley classifications are also accepted, the overall accuracy is 94%
(Table 1).
The remaining cluster of residuals occurs where much of the prairie
peninsula is classified as broadleaf forest rather than grassland.
Classification of the prairie peninsula has been a particular challenge to
biogeographers and climatologists (Borchert 1950, Corcoran 1982, Kiichler 1972,
Manogaran 1983, Transeau 1935). There is considerable debate as to whether
the prairie peninsula is climatically determined, or whether man historically
induced the p>rairie through fire (Vankat 1979). Fires clearly play a major
role in reducing the dominance of woody vegetation. However, a climatic
classification of the prairie peninsula, based on a moisture index of
Thornthwaite, indicates that the region is climatically unique, (Manogaran
1983). The current configuration of the rule-base lacks the PET calculations
that will be necessary to reflect this unique climatology. The next version
of the model will incorporate the PET calculations. The prairie peninsula
VII - 18
-------
also occurs at: the convergence of two air mass gradients (Bryson 1966). As a
consequence, regional rainfall gradients are relatively shallow (Neilson et
al. 1989) and should exhibit considerable year-to-year variability. We tested
the sensitivity of the prairie-forest border to small changes in the winter
rainfall threshold. The prairie peninsula was most sensitive to these changes
while the border locations to the north and south were more stable. It is
possible that year-to-year variability of weather patterns produces extreme
years that preclude the long-term maintenance of forest in the prairie
peninsula. We expect that the model rendition of the prairie peninsula will
improve with future enhancements.
Future Model Developments
The primary limitation of the current configuration of the model is that
it simulates only the supply side of the water balance. The next enhancement
will be to include the demand side as driven by PET. The precipitation
thresholds will be re-evaluated as a function of seasonal PET and seasonal
rainfall supply. Within the water balance thresholds for each life-form water
balance will be treated as a continuum and calculated using energy balance
equations, allowing a prediction of leaf area as well as life-form (Woodward
1987) . Regional scale information on soil water holding capacity and texture
will be required to fine-tune the regional water balance estimates. The
continued use of thresholds in the model recognizes the use of thresholds in
defining physiognomic distinctions. It may be possible in the future to
loosen this constraint and treat life-form physiognomies as a continuum.
Grass physiog;nomy extends toward large and woody, as in bamboo; and, woody
plants extend toward small ephemerals, as in many semi-arid species.
With the incorporation of PET the model will be sufficiently complete
for a preliminary estimate of 2xC02 impacts on biome distribution over the
conterminous U.S. Qualitative estimates have been made based on the
conceptual pre-cursor to this model (Neilson et al. 1989). Biome boundaries
are expected to shift several degrees of latitude and hundreds of meters of
elevation.
Temperature constraints will also be added to the model to differentiate
the mixed forests of the southeast U.S. from the eastern deciduous forest
(Fig. 1) and the temperate forest from the boreal forest. The model will also
VII - 19
-------
be extended to both boreal and tropical regions. These extensions will
require very different kinds of rules than those developed for temperate zone
biomes. For example, the current, thermally defined summer shortens with
increasing latitude and disappears at about the latitude of the boreal forest.
Likewise, the thermally defined winter does not occur in the subtropics, i.e.,
the vicinity of Florida. Transect analyses, as applied in the U.S. (Neilson
et al. 1989), will be extended to other parts of the world to assist in the
formulation of additional rules.
The calibration of the model with regard to secondary rules, those
invoking fire and biotic interactions, will require continuing investigation.
Since the natural environment expresses the influence of these processes, it
is particularly difficult to assign thresholds that adequately constrain
biomes in the absence of secondary rules. For example, the winter
precipitation threshold below which shrubs cannot be supported is currently
set at 80 mm (rule 5, Fig. 4). This produces shrubland in the Great Basin and
desert in the Southwest. It also produces shrubland throughout the Great
Plains. The simultaneous presence of grassland in the plains invokes the
secondary rules to shift the biome classification to grassland. If the winter
threshold (rule 5) is set to 200 mm, fewer plains sites are attributed a shrub
component, but the Great Basin becomes a desert. The Great Basin is
characterized by a series of parallel mountain ranges separated by narrow
valleys, wherein the shrublands occur. The mountains collect winter
precipitation which can run off and infiltrate to deep soil layers in adjacent
alluvial fans (Schlesinger and Jones 1984) providing greater deep soil water
for support of shrubland than incident precipitation alone. Thus, we are left
with a choice for secondary rules: 1) invoke fire in the plains to remove
shrubs, or 2) invoke 'run on' in the Great Basin to augment deep soil water
and support shrubs. Currently, the first option is used. However, there is
little information as to the comparative degree to which these processes
control shrubland distribution.
The model could be described as a demographic rather than an ecosystem
model, since it simulates life-form (Beard 1978), but does not address
nutrient dyns.mics (O'Neill et al. 1986). Alternative approaches to simulation
of productivity and nutrient dynamics will be explored (e.g., Jarvis and
Leverenz 1983). Once the model incorporates an energy balance approach to
VII - 20
-------
determining the life-form thresholds, water flux through the ecosystems will
be directly calculated. This information, coupled with water-use-efficiency
and energy-uses-efficiency parameters (Landsberg and McMurtrie 1985, Lander
1985, Montieth 1977, Tucker et al. 1986) should allow estimates of potential
primary productivity. Potential productivity could be modulated by nutrient-
use-efficiency and soil nitrogen estimates (Prescott et al. 1989, Vitousek
1984). Decomposition rates could be parameterized as functions of temperature
and moisture (Meentemeyer 1978).
Increased levels of C02 are known to increase productivity and water-
use-efficiency in many plants (Strain and Cure 1985). The potential
importance of these effects at landscape and biome scales is under debate. If
these effects are significant at a regional scale, then the impacts of
climatic change could be considerably ameliorated, assuming that the gradually
increasing influence of direct effects of C02 is not preceded by adverse
climatic effects of increasing trace gases, such as widespread drought. Also,
even if direct effects prove important over a wide array of plants, it is not
clear how much influence plant scale water balance, modulated by the stomata,
has over landscape scale water balance. The latter appears to be largely
driven by radiation and modulated by root density and depth and jzfy leaf area/J)
(Federer 1982, Jarvis and McNaughton 1986). Clearly, stomatal processes will
control rates of drying; however, the accurate simulation of drying curves may
not be important at time steps of one month (Cowan 1965).
A particularly important enhancement to the model will be to move from
steady state to transient dynamics. The potential spatial redistribution of
biomes could be mediated by major diebacks or declines in some areas and
advances in others (Neilson et al. 1989, Solomon 1986, Overpeck et al. 1990).
Simulation oJ: these dynamics will require both demographic and ecosystem
processes. 'Che dispersal of important tree types is expected to lag
considerably behind the rapidity of climatic change and the potential movement
of boundaries (Davis 1986-, Smith and Tirpak 1989). It will be important to
incorporate functional categories of seed dispersal in the life-history
characterizations (Beard 1978). As extra-tropical forests shift toward the
poles, large areas of their current distributions could undergo severe,
drought induced decline and dieback (Neilson et al. 1989, Overpeck et al.
1990). These would be susceptible to fire. The release of C02 into the
VII - 21
-------
atmosphere from these fires and subsequent decay of the remaining dead biomass
could produce a positive feedback to the 'greenhouse effect' (Neilson et al.
in prep.)- The regrowth of burned landscapes and the potential expansion of
the tropics could ameliorate this C02 pulse to some extent (Ibid.)- We have
begun modeling these processes and will incorporate these dynamics into future
versions of the model described here.
The sinuilation of transient dynamics could also be necessary for
accurate delineation of some biome boundaries. Sensitivity analyses on the
winter precipitation parameter separating closed from open forest (375 mm),
indicates that the eastern extent of the prairie peninsula is quite sensitive
to this parameter. Year to year variability of rainfall could be shifting
this climatic transition quite considerably (Coupland 1958, Kiichler 1972).
Even though the average climate of the prairie peninsula may be suitable for
forest establishment and growth, the extremes due to natural climatic
variability CDuld preclude their long-term viability in the region.
As model development continues, calibration and validation will be
increasingly important. Although the focus has been on enhancements to the
model, it is apparent that the calibration is very dependant on the accuracy
of the pre-classification. In the Rocky Mountains, for example, correct model
calibration will require a much higher spatial resolution of biome type than
needed for less mountainous terrain. Literature and existing maps will be
useful in this endeavor. However, satellite technology has the potential to
provide this classification at a comparatively high resolution (Tucker et al.
1985) . Remote classification of vegetation should be a high priority for all
the world's vegetation.
Three approaches to model validation will be explored: 1) implementation
on another continent, 2) simulation of paleoclimates (Webb et al. 1990), and
3) rescaling the model to a smaller, but heterogeneous extent. There will be
an attempt to validate the model through extension to other continents, and
also attempts; to reconstruct past vegetation and to rescale the model. Since
it is physically driven and is purported to be mechanistically based, it
should apply at any scale of resolution. Thus, application to a heterogeneous
watershed, such as the Colorado drainage would be a very interesting test of
the model.
VII - 22
-------
Conclusions
The initial stage of development of a rule-based, mechanistic model of
vegetative life-form distribution has been described. Different life-form
mixtures deduced from the rules produce different biomes with demonstrated
prediction potential of up to 94% accuracy. The current configuration of
rules is based entirely on the seasonal patterns of regional water balance and
the relation of these patterns to different plant life-forms. The apparent
success of the; model makes two important points. First, regional water
balance does appear to be a critical climatic mechanism determining plant
distributions. Second, the success of the model, even when the demand side of
regional water balance (PET) was not directly considered, implies that on
regional scales water balance is currently in equilibrium with the vegetation.
It follows that the atmospheric demand for water (PET") is in balance with the <
supply of water through transpiration^ The regional rate of transpiration,
according to theory (Woodward 1987), should be related to the amount of leaf
area (biomass) over that region. The amount of leaf area over a region should
be a direct result of ecosystem feedback processes that select for a specific
leaf area as a function of water supply and atmospheric demand. The
\ __ ———~~~~
implication is that ecosystems are perched on a precarious balance with
respect to regional water balance and that a rapid change in either supply or
demand for water could produce a catastrophic response from regional
vegetation.
Being mechanistic in concept, future model enhancements to incorporate
ecosystem and. spatial demographic processes with temporal dynamics should be
relatively straightforward, albeit challenging to implement. The value of
these efforts, particularly when coupled with global atmosphere models and
land-use characteristics, should be realized in a much improved predictive
potential of future global change and biosphere-atmosphere feedbacks.
VII - 23
-------
References
Abrams, M.D., A.K. Knapp, and L.C. Hulbert. 1986. A ten-year record of
aboveground biomass in a Kansas tallgrass prairie: effect of fire and
topographic position. Amer. J. Bot. 73(10):1509-1510.
Beard, J.S. 1978. The physiognomic approach, pp. 35-64. In: Robert H.
Whittaker, ed. Classification of Plant Communities. The Hague,
Netherlands.
Borchert, J.R. 1950. The climate of the central North American grassland.
Annals of the Association of American Geographers 15(l):l-38.
Box, E.O. 1981. Macroclimate and Plant Forms: An Introduction to Predictive
Modeling in Phytogeography. Dr. W. Junk Publishers, The Hague.
Bryson, R.A. 1966. Air Masses, streamlines, and the boreal forest.
Geographical Bulletin 8(3):228-269.
Burke, M. J., L. V. Gusta, H. A. Quamme, C. J. Weiser, and P. H. Li. 1976.
Freezing injury in plants. Annual Review of Plant Physiology 27: 507-
528.
Corcoran, W.T. 1982. Moisture stress, mid-tropospheric pressure patterns,
and the forest/grassland transition in the south central states.
Physical Geography 3(2):148-159.
Coupland, R.T. 1958. The effects of fluctuations in weather upon the
grasslands of the Great Plains. The Botanical Review 23(5):273-317.
Cowan, I.R. 1965. Transport of water in the soil-plant-atmosphere system.
J. Applied Ecology 2:221-239.
Davis, M.B. 1986. Lags in the response of forest vegetation to climatic
change. pp. 70-76. In: C. Rosenzweig and R. Dickson, eds., Climate-
Vegetation Interactions. OIES-UCAR, Boulder, CO.
Dice, L.R. 1943. The biotic provinces of North America. University of Michigan
Press, Ann Arbor. 1-79.
Federer, C.A. 1982. Transpirational supply and demand: plant, soil, and
atmospheric effects evaluated by simulation. Water Resources Research
18(2)::155-362.
Jarvis, P.G.„ and J.W. Leverenz. 1983. Introduction: Temperate Forest. In
Productivity of temperate, deciduous, and evergreen forests. IN:
Encyclopedia of Plant Physiology, editors. Springer-Verlag, Berlin.
243-280.
Jarvis, P.G., and K.G. McNaughton. 1986. Stomatal control of transpiration:
scaling up from leaf to region. Advances in Ecological Research 15:1-
VII - 24
-------
49.
Kiichler, A.W. 1964. The potential natural vegetation of the conterminous
United States. American Geographical Society Special Publication no.
36.
Kiichler, A.W. 1972. The oscillations of the mixed prairie in Kansas.
Erdkunde, Archiv fur wissenschaftliche Geographic. 26:120-129.
Landsberg, J.J., and R. McMurtrie. 1985. Models based on physiology as tools
for research and forest management, pp. 214-228. In: J.J. Landsberg
and W. Parsons (eds.), Research for Forest Management. Proceedings of a
conference organized by the Division of Forest Research, CSIRO, 21-25
May 1984, Canberra, Australia.
Linder, S. 1985. Potential and actual production in Australian forest
stands, p. 11-34. In: J.J. Landsberg and W. Parsons (eds.), Research
for Forest Management. CSIRO, Australia.
Major, J. 19613. A climatic index to vascular plant activity. Ecology
44:485-498.
Manogaran, C. 1983. The prairie peninsula: a climatic perspective.
Physical Geography 4(2):153-166.
4
Marshall, J.D,. , and R.H. Waring. 1984. Conifers and broadleaf species:
stomatal sensitivity differs in western Oregon. Can. J. For. Res.
14:905-908.
Meentemeyer, V. 1978. Macroclimate and lignin control of litter decom-
position rates. Ecology 59(3):465-472.
Monteith, J.L. 1977. Climate and the efficiency of crop production in
Britain. Phil. Trans. R. Soc. Lond. B. 281:277-294.
Neilson, R.P. 1986. High-Resolution Climatic Analysis and Southwest
Biogeography. Science 232:27-34.
Neilson, R.P. 1987. Biotic regionalization and climatic controls in western
North America. Vegetatio 70:135-147.
Neilson, R.P., G.A. King, R.L. DeVelice, J. Lenihan, D. Marks, J. Dolph, B.
Campbell, and G. Click. 1989. Sensitivity of ecological landscapes and
regions to global climate change. U.S. Environmental Protection Agency,
Environmental Research Laboratory, Corvallis, OR.
Neilson, R.P., and L.H. Wullstein. 1983. Biogeography of two southwest
American oaks in relation to atmospheric dynamics. Journal of
Biogeography 10:275-297.
O'Neill, R.V., D.L. DeAngelis, J.B. Waide, andT.F.H. Allen. 1986. A
Hierarchical concept of ecosystems. In: Monographs in population
VII - 25
-------
biology. R.M. May, editor. Princeton University Press, Princeton, N.J.
Overpeck, J.T., D. Rind, and R. Goldberg. 1990. Climate-induced changes in
forest disturbance and vegetation. Nature 343:51-51.
Prentice, K.C. in press. Bioclimatic distribution of vegetation for GCM
studies. J. Geo.Res.
Prescott, C.E., J.P. Corbin, and D. Parkinson. 1989. Biomass, productivity,
and nutrient-use efficiency of aboveground vegetation in four Rocky
Mountain coniferous forests. Can. J. Res. 19:309-317.
Running, S.W. 1975. Environmental control of leaf water and conductance in
conifers. Can. J. Res. For. Res. 6:104-112.
Quinlan, F. T. , T. R. Karl, and C. N. Williams, Jr. 1987. United States
Historical Climatology Network (HCN) serial temperature and
precipitation data. NDP-019, Carbon Dioxide Information Analysis
Center, Oak Ridge National Laboratory, Oak Ridge, Tennessee.
Runkle, J.R. 1985 Disturbance Regimes in Temperate Forests. In: S.T.A.
Pickett and P.S. White, eds., The Ecology of Natural Disturbance and
Patch Dynamics, pp 17-33.
Schlesinger, W.H., and C.S. Jones. 1984. The comparative importance of
overland runoff and mean annual rainfall to shrub communities of the
Mojave Desert. Hot. Gaz. 145(1):116-124.
Smith, J.B., and D. Tirpak, eds. 1989. The Potential Effects of Global
Climate Change on the United States. Report to Congress. EPA-230-05-
89-050, U.S. Environmental Protection Agency, Washington, DC 413 pp.
Solomon, D. M. 1986. Transient response of forests to C02-induced climate
change: simulation modeling experiments in eastern North America.
Oecologia 68:567-579.
Stephenson, N. L. 1990. Climatic Control of Vegetation.Distribution: The Role
of the Water Balance. American Naturalist 135:649-670.
Strain, B.R. and J.D. Cure. 1985. Direct effects of increasing carbon
dioxide on vegetation. DOE/ER-0238, U.S. Department of Energy, Carbon
Dioxide Research Division, Washington, DC.
Thornthwaite, C.W. 1948. An approach toward a rational classification of
climate:. Geographical Review 38:55-94.
Thornthwaite, C.W., and J.R. Mather. 1955. The Water Balance. Publications
in Climatology, Volume VIII, Number 1. Drexel Institute of Technology.
Centerton, New Jersey.
Transeau, E.N. 1935. The prairie peninsula. Ecology 16(3):423-439.
VII - 26
-------
Tucker, C.J., I.Y. Fung, C.D. Keeling, and R.H. Gammon. 1986. Relationship
between atmospheric C02 variations and a satellite-derived vegetation
index. Nature 319:195-199.
Tucker, C.J., J.R.G. Townshend, T.E. Goff. 1985. African land-cover
classification using satellite data. Science 227:369-375.
US WEST. 1988a. Climatedata User's Manual: TD3200 Summary of the Day --
Cooperative Observer Network. US WEST Optical Publishing. Denver, CO.
US WEST. 1988b. Hydrodata User's Manual: USGS Daily and Peak Values, Version
' 2.0. US WEST Optical Publishing. Denver, CO.
Vankat, J.L. 1979. The Natural Vegetation of North America. John Wiley and
Sons, New York.
Vitousek, P.M. 1984. Litterfall, nutrient cycling, and nutrient limitation
in tropical forests. Ecology 65(1):285-298.
Waring, R.H., and J.F. Franklin. 1979. Evergreen coniferous forests of the
Pacific Northwest. Science 204:1380-1386.
Webb, T.III, ?.J. Bartlein, and J.E. Kutzbach. 1990. Climatic Change in
Eastern North America During the Past 18,000 Years: Comparisons of
Pollen 3ata with Model Results. In: North America and Adjacent Oceans
During the Last Deglaciation. Geology of North America. W.F. Ruddiman,
and H.E. Wright, eds. Geological Society of America, Boulder, Colorado,
in press.
Woodward, F.I. 1987. Climate and Plant Distribution, Cambridge University
Press, London, England. 174 pp.
VII - 27
-------
EFFECTS OF CLIMATE CHANGE ON CARBON STORAGE IN
TERRESTRIAL ECOSYSTEMS: EQUILIBRIUM ANALYSES AT THE
GLOBAL LEVEL
David P. Turner
NSI Technology Services Corporation
US EPA Environmental Research Laboratory
200 SW 35th Street
Corvallis, OR 97333
and
Rik Leemans
Global Change Department
National Institute of Public Health & Environmental Protection
Netherlands
VIII - 1
-------
Abstract
Storage of carbon in the terrestrial biosphere is dependent on the distribution of vegetation
types. This analysis evaluates potential changes in above- and belowground carbon pools
associated with redistribution of vegetation types as driven by climate change. The
approach makes use of the Holdridge life zone classification system to distribute
vegetatation types under the current climate, and doubled-CO2 climate scenarios from four
General Circulation Models (GCMs). An estimate of the global terrestrial carbon pool for
a given climate scenario was made by assigning a representative value for above- and
belowground carbon to each vegetation type and using the areal extent of the vegetation
types to sum stored carbon. All GCMs predicted a net flux of carbon from the atmosphere
to the biosphere under a doubled-CCh climate when considering just the aboveground
biomass. This flux reflected increases in the areal extent of carbon-rich vegetation types
such as tropical humid forests. Two of the GCMs predicted a net loss of belowground
carbon because of large decreases in the areal extent of tundra ecosystems with a high level
of belowground storage. In terms of ppm C02, these fluxes represent a range from adding
13 ppm to the amiosphere to removing 84 ppm. These values suggest the potential for a
moderate negative feedback to global warming over the long term via this mechanism.
VIII - 2
-------
Introduction
Since at least the time of the 18th century geographer Alexander von Humbolt, it has been
recognized that vegetation type or physiognomy is correlated with climate. The
constraints on plant form imposed by climate mean that climate also strongly regulates the
maximum aboveground carbon in terrestrial ecosystems. Correlations of belowground
carbon with climate have also been observed (Post et al. 1983). A change in the areal
extent of different biomes as a result of global climate change will thus be likely to result.
in fluxes of carbon to or from the biosphere, with corresponding changes in the
atmospheric CO;^ concentration. These changes are of interest in terms of positive or
negative feedbacks from the terrestrial biosphere to climate change.
The specific objective of this section is to investigate the potential for a net flux of carbon
between the terrestrial biosphere and the atmosphere due to climate change. This is an
equilibrium analysis (i.e. the assumption is made that climate and vegetation are always in
equilibrium) and a bookkeeping approach is used in which: 1) vegetation types are
distributed across the land surface based on the current climate and double-CO2 climate
scenarios using an existing vegetation-climate correlation system (Holdridge 1947); 2) the
vegetation types are assigned representative above- and belowground carbon pools; 3)
terrestrial carbon storage is summed by climate scenario; and 4) the differences between
current storage and future storage are determined.
VIII - 3
-------
Methods
The Vegetation/Climate Correlation
A number of systems have quantified the vegetation-climate correlation. Holdridge's is
perhaps the earliest and most widely cited (Holdridge 1947, Emanuel et al. 1985). In the
Holdridge system, potential vegetation distribution is a function of annual precipitation and
temperature. Vegetation is divided into 30 types, and the whole system can be displayed in
a triangular grid as in Figure 1. Applying global data sets from weather stations for the
climate variables and a modified version of the Holdridge scheme, the system was correct
for 77% of current observed vegetation in a recent study (Prentice 1990). Similar
vegetation/climate correlation systems (e.g., Box 1981) use additional climate variables
such as seasonally of temperature and precipitation. Research is under way at ERL-C to
develop a rule-based system that takes into account geographic patterns in the seasonality
of precipitation ;uid runoff. In the analysis presented here, the Holdridge scheme is used.
However, the original 30 vegetation classes have been aggregated into 14 biome-level
classes (Table 1) more appropriate to the level of resolution achievable in assigning carbon
pools to specific vegetation types.
Vegetation Distribution under Current and Double-CCh Climates
The basic approach to predicting how biomes or vegetation types may be redistributed due
to global climate change is to combine a spatially distributed climate scenario and a climate-
vegetation correlation system (King 1990). Several investigators have used the Holdridge
classification system to predict the future distribution of vegetation types in this manner
(Emanuel et al. 1985a; Prentice and Fung, in press).
General Circulation Models (GCMs) are used to produce the climate scenarios. Current
GCMs are configured with grid cells several hundred km on a side but predictions of
climate at a higher spatial resolution can be made by application of the difference between
1 x C(J2 and doubled-COi GCM runs to the hisotrical climate (Smith and Tirpak 1989).
VIII - 4
-------
u
O
ro
O)
Q.
E
o>
ro
r>
c
c
ro
C
ro
0)
Latitudinal
belts
Altitudinal
belts
Subalpine _
rv-' __ _ _ __ ,_
Cool
Temperate
Lower --^
montane E
T, Deioil \T "">'" x I' c ' \l \
• .III \l/ B,,ih Nl/ Sl.pp. \l/ Foie" I/ Fo.Ml
Temperate
1 /™«^J^^^^-/^'
Premontane E
/ / / / /1ii/riU't7iiT/"''i""/'"'/'"y*i
12.0 —
24.0
-, 1.5
- 3.0
6.0
12.0
-^ 24.0
Critical
temperature ***
line
Figure 1. Holdridge life zone classification system. (Holdridge 1967)
-------
As the spatial resolution of the GCMs and their success in recreating the current climate
improves, more direct coupling of GCMs to the vegetation will become possible.
In this study, doubled CO2 runs from the GCMs of four institutions were used: UKMO
(United Kingdom Meteorological Organization, OSU (Oregon State University), GISS
(Goddard Institute for Space Science, and GFDL (Geophysical Data Center). Current
climate was based on long-term weather station data from 13,000 global sites interpolated
to 0.5° of latitude: and longitude. The assignment of a vegetation type to each grid cell and
the determination of the total areal extent of each vegetation type was done by Leemans
(1990) using a geographic information systems approach.
Carbon Pools
Aboveground carbon ranges from 25 kg/m2 in dense forests to less than 0.5 kg/m2 in the
arctic (Olson et al. 1983). The summary of Olson et al. (1983) is used in this study as a
basis for assigning aboveground carbon pool values to the aggregated Holdridge vegetation
types (Table 2). The assignment of belpwground carbon pools to the Holdridge vegetation
types (Table 2) was taken from the study of Post et al. (1982) in which data from 2696
global sites were used to construct isolines for belowground carbon storage within the
Holdridge climate diagram (Figure 2). Based on the current climate, this analysis
estimated totals for aboveground carbon (852 Gt) and belowground carbon (1456) which
are consistent with other studies (Schlesinger 1977, Woodwell et al. 1978, Oades 1988).
Certainly there i.s large spatial heterogeneity in these pools within each vegetation type but
evaluation of that heterogeneity will require detailed survey information that is not
currently available. Mean values were using in this analysis as a first approximation to
accounting for the spatial heterogeneity.
The absolute value of the representative carbon pools is also in question. Recent analysis
of boreal forest ecosystems (Botkin and Simpson 1990) suggest that previous estimates of
aboveground carbon in these and other ecosystems have been uniformly high. Additional
VIII - 6
-------
Table 1. Aggregation Scheme for Combining Holdridge
Life Zones into Biomes.
Biomc
1 Tundra
2 Cold Parklands
3 Forest Tundra
4 Boreal Forest
5 Cool Desert
6 Steppe
7 Temperate Forest
8 Hot Desert
9 Chapparal
10 WaiTn Temperate Forest
11 Tropical Semi-Arid
12 Tropical Dry Forest
13 Tropical Seasonal Forest
14 Tropical Rain Forest
Holdridge Zone
Ice
Polar Desert
Subpolar Dry Tundra
Subpolar Moist Tundra
Subpolar Wet Tundra
Subpolar Rain Tundra
Boreal Desert
Boreal Dry Scrub ^——
Boreal Moist Forest
Boreal Wet Forest
Boreal Rain Forest
Cool Temperate Desert
Cool Temperate Desert Scrub
Cool Temperate Steppe
Cool Temperate Moist Forest
Cool Temperate Wet Forest
Cool Temperate Rain Forest
Warm Temperate Desert
Warm Temperate Desert Scrub
Subtropical Desert
Subtropical Desert Scrub
Tropical Desert
Tropical Desert Scrub
Warm Temperate Thorn Steppe
Warm Temperate Dry Forest
Warm Temperate Moist Forest
Warm Temperate Wet Forest
Warm Temperate Rain Forest
Subtropical Thorn Woodland
Tropical Thorn Woodland
Tropical Very Dry Forest
Subtropical Dry Forest
Tropical Dry Forest
Subtropical Moist Forest
Tropical Moist Forest
Subtropical Wet Forest
Subtropical Rain Forest
Tropical Wet Forest
Tropical Rain Forest
VIII - 7
-------
Polar
Subpolar
Boreal
Cool temperate
Warm temperate
Subtropical
Tropical
Nival
Alpine
\. Subalpine
_•£/
>V%
O Montane
^3 Lower montane
V9-r
'ho Premontane
1.5C
3° ^
rs
60 L.
i»
I
u
12-I
24°
2 468 10 1418 22
Carbon in mineral soil (kg m~2)
Figure 2. Relationship of belowground carbon pools to holdridge life zones. (Post et al.
1932)
studies, probably coordinated with remote sensing, are needed to evaluate this question. In
the present analysis, the carbon pools (Kg/m2) associated with each vegtation type are held
constant, and a difference between current and future storage is calculated. Thus, the
approach is not particularly sensitive to a consistent overestimate of above- or
belowground carbon across all vegetation types.
VIII - 8
-------
Table 2. Above- and Belowground
Biome
1 Tundra
2 Cold Parklands
3 Forest Tundra
4 Boreal Forest
5 Coot Desert
6 Steppe
7 Temperate Forest
8 Hot Desert
9 Chaparral
10 Wairm Temperate Forest
11 Tropical Semi- Arid
12 Tropical Dry Forest
13 Tropical Seasonal Forest
14 Tropical Rain Forest
Pools for World
Aboveground
Carbon Density
(kg/m2)
0.5
0.8
6.0
11.0
0.6
1.5
11.0
0.4
4.0
10.0
5.0
7.0
10.0
15.0
Biomes
Belowground
Carbon Density
(kg/m2)
22.0
10.0
11.0
15.0
9.0
13.0
18.0
1.0
8.0
10.0
4.0
7.0
11.0
19.0
Aboveground data after Olson et al. 1983
Belowground data after Post et al. 1982
VIII - 9
-------
Results and Discussions
Differences in the areal extent of the biomes, as predicted by four GCMs, are listed in
Table 3. Results for the four GCMs show some basic similarities. At middle to high
latitudes, tundra and boreal forests contract, and temperate forest (mostly coniferous)
expands. In the tropics, both the semi-arid woodland and the tropical rain forest have
large increases in areal extent, mostly at the expense of tropical seasonal forests.
Analysis of changes in carbon storage reveals that all climate scenarios predict an uptake of
carbon from the atmosphere for the aboveground component of the biosphere (Figure 3),
ranging from 37 to 116 Gt. There are great discrepancies between the model runs for the
change in belowground carbon storage. The UKMO model predicts a flux to the
atmosphere of 126 Gt, while the OSU model predicts a belowground accumulation of 37
Gt. The large fluxes to the atmosphere appear to be associated with loss of carbon from
reduction in the area of tundra and boreal forest, both having relatively high levels of
belowground carbon.
The net change of above- and belowground carbon ranged from a 169 Gt uptake predicted
by the OSU model to a 68 Gt release predicted by the UKMO model. A key question
appears to be how great the gains will be in tropical rain forest. Because both above- and
belowground components have high carbon storage in these forests, the changes from
tropical dry forest to tropical wet forest tend to drive the global trends. In terms of
climate, the question is whether precipitation will substantially increase in tropical and
subtropical latitudes.
A number of other investigators have made estimates of potential changes in carbon
storage based on the climate-vegetation correlation systems and other GCM doubled-COi
climate scenarios. For aboveground carbon, Sedjo and Solomon (1988) predicted a net
flux of 13.9 Gt to the atmosphere. Much of that change in aboveground carbon was
accounted for by loss of boreal forest and increases in savanna. Lashof (1987, 1989),
evaluating potential above- and belowground changes, reported a range from a 64 Gt
uptake from the atmosphere to a 26 Gt release, again depending on the GCM used. A
VIII - 10
-------
Table 3. Changes in Area Extent of Different Vegetation Types as Predicted
by 4 GCMs.
Smith & Leemam Aggregated
Holdridge Life Zones
1 Tundra
2 Cold Parklands
3 Forest Tundra
4 Boreal Forest;
5 Cool Desert
6 Steppe
7 Temperate Forest
8 Hot Desert
9 Chaparral
10 Warm Temperate Forest
11 Tropical Semi- Arid
12 Tropical Dry Forest
13 Tropical Seasonal Forest
14 Tropical Rain Forest
TOTAL
^
Current
J^P^P*
r£-eJP"'1
9.30
2.79
8.90
15.03
4.01
7.39
9.94
20.85
5.58'
3.17
9.56
14.86
15.13
8.46
134.97
GFDL
v
-6.11
0.03
-5.02
-5.45
-0.97
4.20
1.92
-0.20
1.83
-1.22
4.43
4.71
-5.11
6.95
GISS
-5.05
-0.41
-3.03
-1.54
-1.67
-0.46
3.49
-3.22
-0.13
-1.25
7.18
4.49
-7.24
8.85.
OSU
-4.56
-0.01
-2.90
-0.89
-0.82
1.30
1.63
-1.42
-0.69
-0.72
2.58
0.00
-4.98
11.57
UKMO
-6.43
-1.09
-5.50
-4.85
-1.93
-0.21
3.04
-0.92
2.99
-0.29
7.07
11.19
-7.48
4.40
VIII - 11
-------
GCM PREDICTED CHANGES IN C DISTRIBUTION
Above Ground
GFDL
L
GISS ^
OSU
UKMO •
Below Ground
GFDL~-
i
GISS ;-
L
OSU
UKMO
Total
GFDL -
GISS -
OSU
UKMO
-150 -100 -50 0
Flux to Atmosphere
50 100 150
Uptake by Biosphere
200
CARBON CHANGE (Gt)
Figure 3. Potential changes in terrestrial carbon storage based on redistribution of
vegetation types. Values are differences between stored carbon under the
current climate and under doubled-CCh climates as predicted by four different
general circulation models (GFDL, GISS, OSU, UKMO).
VIII - 12
-------
modified Holdridge classification system '(Prentice 1990), which better represented the
current vegetation, suggested a much greater potential flux of carbon (270 Gt) to the
surface (Prentice and Fung, in press). As in our analyses, the uptake was related primarily
to expansion of Ihe tropical rain forest biome which has relatively high levels of above-
and belowground carbon.
To begin evaluating the magnitude of feedbacks associated with these carbon fluxes, it is
necessary to know: 1) how much of the carbon would be retained in the atmosphere (in the
case of a positive feedback); 2) the radiative forcing of that retained component; and 3) the
sensitivity of the climate system to that forcing. Likewise, for a negative feedback, the
magnitude can most readily be evaluated via the impact on atmospheric CO2 concentration.
The proportion of carbon released to the atmosphere due to fossil fuel combustion and
deforestation that has accumulated in the atmosphere has been approximated at 40%
(International Panel for Climate Change, 1990). The remainder has been taken up by the
ocean, in part due to photosynthetic uptake of CO2 by plankton, and by the terrestrial
biosphere via a net imbalance between CO2 assimilation during photosynthesis and CO2
loss via plant, animal and microbial respiration. The atmospheric retention factor is
difficult to determine because of uncertainties about many terms in the global carbon cycle
(Keeling et al. 1989). Assuming a retention factor of 40%, the 68 Gt release of carbon in
the case of the maximum positive feedback discussed earlier would result in 27 Gt increase
in the atmospheric pool. Given a net change in the atmospheric carbon pool, a conversion
to a change in atmospheric CO2 concentration can be made using an approximation of 2 Gt
carbon to 1 pprn CO2- Thus the increase would be about 13 ppm CO2. In the case of a
negative feedback, Gt can be converted directly into ppm CO2 extracted from the
atmosphere by (dividing by two.
The estimation of a climate forcing associated with a change in atmospheric CO2,
independent of feedbacks in the climate system, is relatively straightforward (Hansen et al.
1988) and the GCMs are in general agreement (Cess et al. 1989). Following the approach
of Hansen, the climate forcing of 13 ppm added to a reference CO2 concentration of about
VIII - 13
-------
450 ppm for a doubled CO 2 climate (the rest of the warming being provided by other trace
gases), is about + 0.05°C. The effect of subtracting 84 ppm (169Gt/2) would be about -
0.28°C. These compare to the doubled COi climate forcing without feedbacks of +
1.25°C. The GCMs are in much less agreement in accounting for various feedbacks in the
climate system given an initial forcing (Hansen et al. 1984). The total climate forcing for
the given potential changes in COi is thus much less certain. However, they would
probably scale to the magnitude of the orginal forcing, so a negative feedback of perhaps
20% of the original forcing could be expected due to changing carbon pools.
Modifiers and Uncertainties
The most troublesome assumption using the equilibrium approach described here is that
the climate/vegetation correlation will remain constant. Physiological responses of plants
to CO2 suggests that water use efficiency increases with ambient CO2 concentration. If this
effect is large, aboveground carbon associated with a particular water regime might be
expected to increase and thus the magnitude of the predicted feedbacks associated with
carbon pool changes would be different.
A second concern is that the analysis described above assumes that climate and vegetation
are now, and will remain, in equilibrium. In fact, the predicted rates of climate change
(>0.1°C per decade) are an order of magnitude faster than what natural systems
experienced during the glacial-interglacial cycles of the Pleistocene period. Analyses of
tree seed dispersal distances and pollen records indicate that the rate of climate change may
exceed rates of vegetation redistribution (Davis 1981). The predicted discrepancy between
these rates may have significant implications for changes in aboveground carbon pools.
Large transient carbon fluxes could occur if there is a rapid burnoff of carbon as the
vegetation drifts out of equilibrium with the climate and vegetation recovery is slow
because of limited migration rates for appropriate species and slow regrowth.
There will also be lags in the equilibrium between belowground carbon pools and climate.
Where carbon gains are expected, as with the transition from tropical seasonal to tropical
VIII - 14
-------
rain forest, there is little data on how rapidly the predicted accumulation would occur.
Factors such as soil mineralology and texture may have strong local influences (Oades
1988). Land use considerations will likewise influence carbon storage because of the
tendency for cultivation to decrease stored carbon.
Where equilibrium analyses suggest a reduction in belowground carbon, an important
consideration will be the fractionation of the soil organic matter. Microbial respiration
increases with temperature, assuming optimal moisture, but belowground carbon is not all
readily oxidizable. A large proportion of grassland soil carbon pools has been classified as
recalcitrant to decomposition, with a turnover time of up to 1000 years (Parton et al.
1988). Forest soils are also characterized by a large recalcitrant fraction, and lags in the
reduction of soil organic matter due to warming may be on the order of hundreds of
years. Continued research on the mechanisms and modeling of soil organic matter
turnover is needed.
These analyses also ignore the human factor which will exert a tremendous influence on
rates of vegetation change. Currently anthropogenic factors use or co-opt approximately
40% of the earth's potential net primary production and over large areas the productive
capacity of the land is being reduced (Vitousek et al. 1986). Land used for human
purposes will probably not be readily transformed to the high-carbon-storage vegetation
types predicted in this analysis. However, humanity may be able to promote changes that
would favor carbon sequestration. Management options are available for reducing carbon
fluxes to the atmosphere and maximizing carbon sinks associated with the terrestrial
biosphere. Their implementation will require an understanding of current climate-
vegetation-soil relationships and of locations where the climate will favor specific
vegetation types in decades to come.
VIII - 15
-------
Conclusions
These equilibrium analyses suggest that the changes in terrestrial carbon pools induced by
climate warming will be a negative feedback to that warming. That is, as global warming
increases, carbon storage will increase and the rate of increase in atmospheric CO2
concentration will be moderated. The greater the increase in precipitation for the low
latitudes, thus creating tropical wet forest, the stronger that feedback will be. These
analyses do not consider 1) potential short term fluxes of carbon due to changing
disturbance regimes, 2) the relative slow response of belowground carbon pools to climate
change, and 3) potential enhancements or reductions of carbon storage due to management
practices.
Acknowledgements
Distribution of the Holdridge vegetation types under current and doubled-CC>2 climates
were generously provided by Rik Leemans and Tom Smith from the International Institute
for Applied Systems Analysis.
VIII - 16
-------
References
Botkin, D.B. and L.G. Simpson. 1990. Biomass of the North American boreal forest: A
step toward accurate global measures.
Biogeochemistry. 9:161-174.
Box, E.O. 1981,, Macroclimate and Plant Forms: An Introduction to Predictive Modeling
in Phytogeography. Dr. W. Junk Publishers, The Hague.
Cess, R.D., G.L. Potter, J.P. Blanchet, G.J. Boer, S.J. Ghan, J.T. Kiehl, H. Le Treut, Z.-X
Li, X.-Z Liang, J.F.B. Mitchell, J.-J Morcrette, D.A. Randall, M.R. Riches, E. Roeckner,
U. Schlese, A. Slingo, K.E. Taylor, W.M. Washington, R.T. Wetherald, and I. Yagai.
1989. Interpretation of cloud-climate feedback as produced by 14 atmospheric general
circulation models. Science. 245:513-516
Davis, M.B. 1981. Quaternary history and the stability of forest communities. In: D.C.
West, H.H. Shugart & D.B. Botkin, eds., Forest Succession: Concepts and Application.
Springer-Verlag, New York. 517 pp.
o
Emanuel, W.R., H.H. Shugart, and M.P. Stevenson. 1985. Climatic change and the broad-
scale distribution of terrestrial ecosystem complexes. Climatic Change 7:29-43
Hansen, J., A. Lacis, D. Rind, G. Russel, P. Stone, I. Fung, and J. Lemer. 1984. Climate
sensitivity: Analysis of feedback mechanism. In: Climate Processes and Climate
Sensitivity, Geophys. Monogr. Ser., AGU, Washington, DC. 29:130-163.
Hansen, J., I. F'ung, A. Lacis, D. Rind, S. Lebedeff, R. Ruedy, G. Russell, and P. Stone.
1988. Global climate changes as forecast by Goddard Institute for Space Studies Three-
Dimensional Model. J. Geophys. Res. 93:9341-9364.
Holdridge, L.R. 1947. Determination of world formulations from simple climatic data.
Science 105:367-368.
VIII - 17
-------
Intergovernmental Panel on Climate Change. 1990. Scientific Assessment of Climate
Change. Report for WGI Plenary Meeting.
Keeling, C.D., R.B. Bacastow, A.F. Carter, S.C. Piper, T.P. Whorf, M. Heimann, W.G.
Mook, and H. Roeloffzen. 1989. A three-dimensional model of atmospheric CC»2 transport
based on observed winds: analysis of observational data. Geophysical Monograph 55:165-
236.
King, G.A. 1990. Effects of global climate change on global vegetation. This document.
Lashof, D.A. 1989. The dynamic greenhouse: feedback processes that may influence
future concentrations of atmospheric trace gases and climatic change. Climatic Change
14:213-242.
Lashof, D.A. 1987. The role of the biosphere in the global carbon cycle: evaluation
through biospheric modeling and atmospheric measurement. PH.D. Dissertation, Energy
and Resources Group, Univ. of Ca, Berkeley.
Oades, J.M. 1938. The retention of organic matter in soils. Biogeochemistry 5:35-70.
Olson, J.S., J.A., Watts, and L.J. Allison. 1983. Carbon in live vegetation of major world
ecosystems. ORNL-5862. Oak Ridge National Laboratory, Oak Ridge, Tn. 180 pp.
Parton, W.J., J.W.B. Stewart, and C.V. Cole. 1988. Dynamics of C,N, P and S in
grassland soils: a model. Biogeo. 5:109-131.
Post, W.M., W,,R. Emanuel, PJ. Zinke & A.G. Stangenberger. 1982. Soil carbon pools
and world life zones. Nature. 298:156-159.
Prentice, K.C. 1990. Bioclimatic distribution of Vegetation for General Circulation
Model Studies. J.Geophy. Res. 95:11811-11830.
VIII - 18
-------
Prentice, K.C., and I.Y. Fung. Bioclimatic simulations test the sensitivity of terrestrial
carbon storage to perturbed climates. Nature. In press.
Schlesinger, M.E. and Z.C. Zhao. 1989. Seasonal climatic change introduced by doubled-
C02 as simulated by the OSU atmospheric GCM/mixed-layer ocean model. J. Climate
2:429-495.
Schlesinger, W.H. 1977. Carbon balance in terrestrial detritus. Ann. Rev. Ecol. Syst.
8:51-81.
Vitousek, P.M., P.R. Ehrlich, A.H. Ehrlich, and P.A. Matson. 1986. Human
appropriation of the products of photosynthesis. BioScience 36:368.
VIII - 19
-------
CLIMATE CHANGE
AND
ISOPRENE EMISSIONS FROM VEGETATION
D.P. Turneri, J.V. Baglioi, D. Pross3, A.G. Wonesi,
B.D. McVeety2, R. VongS, D.L. Phillips*
iNSI Technology Services, Corvallis, Oregon
2Pacific Northwest Laboratories, Richland, Washington
3Oregon State University, Corvallis, Oregon
4U.S. Environmental Protection Agency, Corvallis, Oregon
IX- 1
-------
ABSTRACT
A global model was developed for estimating spatial and temporal patterns in the emission
of isoprene from vegetation under the current climate and doubled CO2 climate scenarios.
Current emissions; were estimated on the basis of vegetation type, foliar biomass (derived
from the satellite-generated Global Vegetation Index), and global databases for air
temperature and photoperiod. The model had a monthly time step and the spatial
resolution was 0.5 degrees latitude and longitude. Doubled CO2 climate emissions were
estimated based on predicted changes in the areal extent of different vegetation types, each
having a specific rate of annual isoprene emissions. The global total for current emissions
was 560 Tg, which agrees reasonably well with estimates generated by other means. The
isoprene emissions under a doubled CO2 climate were about 25 % higher than current
emissions due mainly to the expansion of tropical humid forests which had the highest
vegetation-specific emission rates. An increase in isoprene emissions is expected to
increase atmospheric concentrations of ozone and methane which are important
greenhouse gases; however, detailed treatment of this question awaits incorporation of
these emission surfaces into 3-D atmospheric chemistry models.
IX-2
-------
INTRODUCTION
Measurements of chemical emissions from plants began over two decades ago (Rasmussen
and Went, 1965) and since then it has become evident that plants emit a great variety of
volatile compounds. The most commonly studied chemical species of plant origin are the
nonmethane hydrocarbons (NMHCs). In terms of the flux of NMHCs, isoprene (CsHg)
and the monoterpenes such as alpha-pinene (CioHie) are the most important. However,
hundreds of other biogenic nonmethane hydrocarbons (NMHCs) have been identified and
different plant species vary greatly in the type and quantities of hydrocarbons emitted
(Winer, 1989). Although plants may expend up to a few percent of their fixed carbon on
NMHC emissions (Zimmerman et al. 1988, Monson and Fall 1989), there is limited
understanding of their physiological and ecological significance (Harborne, 1988).
The goal of evaluating of NMHC emissions in the context of global climate change lies in
relating their emissions to the current and future tropospheric concentrations of
radiatively important trace gases (RITG) such as methane (CH4), ozone (Os), and
chloroflorocarbons (CFCs). These compounds are also know as greenhouse gases. A
schematic depiction of some of the important tropospheric chemical reactions involving
these gases is given in Figure 1.
Over the next 50 years, the effect on Earth's radiation balance of increasing concentrations
of radiatively important biogenic and anthropogenic trace gases is expected to equal that of
increasing COi (Ramanathan et al. 1985). Methane and many other important trace
species are removed from the atmosphere primarily by the hydroxyl radical, which is also
consumed in the oxidation of NMHCs. Because NMHCs consume the hydroxyl, NMHC
emission rates will influence the atmospheric lifetime, and hence the concentration, of
other RITG species. NMHCs are also important in relation to photochemical production
of ozone in the troposphere (Logan 1985).
This report describes an approach to estimating global patterns of plant emissions under
the current climate and under climate scenarios associated with a doubling of atmospheric
C02. We have initially focused on isoprene emissions to facilitate the development of
IX-3
-------
spatially distributed estimation methods. In the future, we anticipate the refinement of this
technique followed by the application of this approach to additional biogenic species.
Isoprene was chosen "as the test case because it has a relatively short atmospheric lifetime,
and therefore, knowledge of its spatial and temporal emission patterns is important to
evaluating its role in regional and global atmospheric chemistry. There are also several
other global emission estimates that have been derived using methods different than
described here, and thus can be used for intercomparison and validation (Zimmerman et
al. 1978, Crutzen and Gidel 1983, Logan et al. 1985, Rasmussen and Khalil 1988).
Figure 1. Principal chemical reactions relevant to tropospheric chemistry (from
NASA, 1988). RH indicates hydrocarbons.
IX-4
-------
The approach in iJiis study to estimating current emissions was to apply satellite based
remote sensing for characterization of global spatial and temporal patterns in foliar
biomass, and a global temperature database for driving temperature/emissions
relationships. The projected emissions for a doubled CO2 climate are based on an
equilibrium approach. Here, the average annual emission rate is determined for each
vegetation type under current conditions, and the predicted change in the areal extent of
the vegetation types due to climate change is used to estimate future emissions. This
approach makes the obviously unreal assumption that vegetation and climate always stay in
equilibrium. However, it is a start towards determining the sign and magnitude of possible
feedbacks to climate change mediated by biogenic NMHC emissions.
The primary objectives of this work were thus to 1) provide a basis for estimating
terrestrial biogenic emissions under current climate conditions and forecasting emissions
under future climate scenarios, 2) develop a spatial-temporal terrestrial biogenic emissions
model that could be incorporated into the three dimensional global atmospheric chemistry
model, and 3) establish an ecosystem model framework based on remote sensing that will
be able to take advantage of the increasingly refined satellite remote sensing capability
being developed by NASA under the EOS (Earth Observation System) program.
The background material which follows summarizes the relationship of NMHC emissions
to the concentrations of radiatively and photochemically important trace gas species in the
troposphere, and provides information on plant and environmental factors influencing
emission rates.
Atmospheric Chemistry
NMHCs are representative of the general pattern of biogenic release of reduced chemical
species which are oxidized in the atmosphere and returned to the surface via precipitation
scavenging, biological uptake or surface deposition. NMHCs have relatively short
atmospheric lifetimes (hours to days) due to rapid oxidation by hydroxyl radicals (OH)
IX-5
-------
and ozone (Warneck 1988). They may be only partially oxidized, and scavenged as
organic acids, or further oxidized to CO and eventually CO2. It is currently believed that
a significant proportion of the global CO budget is related to biogenic NMHC emissions
(Logan et al. 1981). Total biogenic NMHC emissions are approximately an order of
magnitude larger than anthropogenic NMHC (volatile organic compound) emissions
associated with fossil fuel combustion (Hanst et al. 1980).
Relationship to Tropospheric Ozone
The oxidation of NMHCs, and products such as CO, can result in the formation of
tropospheric ozone when atmospheric NOX concentration is greater than about 20 - 200
ppbv. It is unlikely that the oxidation of most biogenic NMHCs currently involves ozone
production because of the low NOX levels (Crutzen 1988). However, NOX levels are rising
over large geographical areas. On a global basis, anthropogenic sources of NOX via fossil
fuel burning are already greater than biogenic NOX emissions (Logan 1983). Fossil fuel
NOX emissions have decreased in the U.S. in the last decade but may be expected to
increase globally due to rapid fossil-fuel-based industrialization in less developed countries
(Kavenaugh 1987). In the northern hemisphere, tropospheric ozone is increasing at a rate
of about 1 % pei' year and there is some evidence that anthropogenic NO emissions are the
cause of the increase (Crutzen and Gidel 1983, Dignon and Hameed 1985, Liu et al. 1987).
NOX levels in niral Pennsylvania, in combination with both biogenic and anthropogenic
NMHCs, are high enough to promote ozone buildup to over 100 ppbv in the summer
(Trainer et al. 1987), i.e., more than twice the probable background level.
Where NOX is not high enough to promote ozone formation during hydrocarbon
oxidation, ozone can be consumed by interactions with NMHCs. Several authors have
suggested that the terrestrial land masses located in the tropics play a significant role in the
global ozone budget (e.g., Thompson et al. 1989), acting mainly as a sink. The ozone sink
strength is a function of both plant uptake and interactions with NMHCs. The exception to
IX-6
-------
this would be during periods of extensive biomass burning (Delany et al. 1985) when the
combination of NOX and hydrocarbons, produced during combustion, produces ozone.
Hydroxyl Radical
A second major impact of biogenic NMHC emissions on tropospheric chemistry concerns
the hydroxyl radical (see Figures 1 and 2). The NMHCs are an important sink for OH
radicals in the plaaetary boundary layer (Jacob and Wofsy 1988) but also to a great extent,
via oxidation of carbon monoxide (an intermediate in the NMHC oxidation pathway), in
the free troposphere. The hydroxyl radical is one of the most reactive species in the
troposphere, having an atmospheric lifetime on the order of seconds. Its concentration
largely controls the atmospheric lifetime and degree of escape into the stratosphere of
important species such as methane, a greenhouse gas, and NOX) a catalyst for ozone
formation. The variety of atmospheric chemical reactions that involve the hydroxyl
radical are shown in Figure 2.
M. 0.
3 XO. 0,
Figure 2.
A depiction of the central role OH plays in the oxidation of tropospheric.
trace gases (from NRC, 1984).
IX-7
-------
To the degree thai: current and future NMHC emissions reduce OH, they will contribute to
global warming because of an increased atmospheric lifetime for methane and HCFCs, the
less inert substitutes for stratospheric ozone depleting CFCs. Models of atmospheric
chemistry suggest a global decrease in OH (Thompson and Cicerone 1986) due to
increasing methane and carbon monoxide emissions over the last century, Concentrations
of CO and CH4 are increasing at rates of 1-2% per year (Khalil and Rasmussen 1990) and
as much as half of the methane increase may be due to OH decrease (Thompson and
Cicerone 1986). Some modeling efforts indicate instability in the OH concentration under
the influence of increasing NMHC, CO or CFLi emissions (Thompson and Cicerone 1986,
Thompson et al. 1990, Guthrie 1990), however, there may be unidentified or poorly
modeled feedbacks which would come into play as the chemical climate changes.
Peroxyacetyl Nitrate
Another important chemical reaction involving NMHCs is the formation of peroxyacetyl
nitrate (PAN), an intermediate in NOX oxidation. Gaseous hydrocarbon emissions favor
production of PAN (Singh 1987). In addition to being toxic to plants and humans, PAN
effectively prevents NO from acting as an agent in further ozone production. Later
thermal breakdown of PAN releases NO, and this mechanism provides a reservoir of NOX
and a means for long range transport of NOX to areas of relatively clean air. Such
transport will increase tropospheric ozone formation in rural areas.
Aerosols
NMHCs originating in forest habitats may also participate in formation of aerosols,
particularly in areas of high NMHC emission such as over wet tropical forests (Talbot et
al. 1988). The blue haze seen over the Great Smoky Mountains before heavy
industrialization was probably from this source. Aerosols participate in the climate system
as cloud condensation nuclei and absorbers of solar radiation, thus influencing the
IX-8
-------
precipitation regime and the distribution of heat in the atmosphere. Many aspects of these
phenomena are as yet poorly understood.
Isoprene Emissions Rates
A great variety of plant species have been examined for biogenic emissions using field and
laboratory enclosures (Zimmerman 1979, Evans et al. 1982, Winer et al. 1989).
Techniques differ considerably between investigators, but some comparisons among
species are possible by standardizing emissions to a common temperature using
experimentally determined algorithms (e.g., Tingey et al., 1979). There is a tendency for
either isoprene or monoterpenes to dominate emissions in a given species. For species
with mainly isoprene emissions, the rates tend to fall below 40 ug C gdw-i hr-i, where
gdw is the grams dry weight of plant tissue. In the study of Evans et al. (1982), for
example, emission of isoprene ranged from 0.01 to 38 ug C gdw-i hr-i for a group of 16
tree species of various physiognomic types. These surveys unfortunately have been limited
primarily to plant species of the U.S.
In isoprene emitting plants, temperature and light strongly regulate the rate of emissions
(Tingey et al. 1979). Emissions typically increase exponentially with temperature
(Monson and Fall 1989) and are directly coupled to the light regime, and hence most likely
to photosynthesis (Monson and Fall 1989). Unlike the monoterpenes which accumulate in
glands and are emiited a rates independent of light intensity, isoprene biosyntheses and thus
emissions are quite low in the absence of light.
IX-9
-------
METHODS
Current Global Isoprene Emissions
The current global isoprene flux was estimated at a spatial resolution of 0.5 latitude by
0.5 longitude and a monthly time step. The general approach was as follows: 1) assign a
monthly active foliar biomass (Kg/m2) to each grid cell based on vegetation type and the
satellite derived Global Vegetation Index , 2) reduce that biomass by a vegetation type-
specific proportion (%) of non-isoprene emitting biomass, 3) calculate an emission rate
based on mean monthly temperature (ug g-i hr-i m-2) and 4) multiply by daylight hours in
the month (hrs). Total annual global isoprene emissions (in Tg) was computed by
multiplying the flux per unit area times the cellular area for each grid cell and summing
over the entire terrestrial surface (59,049 grid cells) and over the year.
Annual Emissions = (foliar biomass per grid cell) * (% plant species emitting
isoprene per grid cell) * (temperature driven emission rate per grid cell) * (hours of
daylight per grid cell).
The assignment of vegetation type to each grid cell was based on the global vegetation
database of Olson et al. (1983). Because of limited knowledge about the ranges of foliar
biomass and the proportion of isoprene emitters, the 52 vegetation classifications in this
database were aggregated into 19 vegetation types as illustrated in Figure 3. The areal
extent, maximum foliar biomass and proportion of foliar biomass emitting isoprene for
these vegetation types are given in Table 1.
The spatial and temporal patterns in active foliar biomass density (FED) were estimated
using the Global Vegetation Index (GVI). The GVI satellite imagery was obtained from
The U.S. Army Corps of Engineers Construction Engineering Research Laboratory for
the year 1988 at a weekly time step. It originated from the National Oceanic and
Atmospheric Association (NO A A) satellite series'which carries the Advanced Very High
Resolution Radiometer (AVHRR) and generates daily global coverage. The GVI is
IX- 10
-------
essentially the NDVI (Normalized Difference Vegetation Index), a "greenness" index,
brought up to a 16 Km spatial resolution and a one week time step by use of a maximum
value compositing procedure (Holben and Fraser 1986, Holben 1986). Weekly GVI data
were composited using the maximum value proceedure to produce monthly maximums.
An example of the GVI surface for the month of July, 1988 is displayed in Figure 4.
NDVI and GVI imagery have considerable value for monitoring the seasonal and annual
variability of vegetation at the global scatej (Goward et al. 1985, Townshend and Justice
1986, Tucker et al. 1986, Goward 1989), as illustrated in Figure 5. NDVI and GVI have
been correlated! with vegetation properties that include leaf area index, primary
production, and annual evapotranspiration (Asrar et al. 1984, Sellers 1985, and Tucker
and Sellers, 1986, Box et al. 1989). Nevertheless there are definite limitations in its use,
including its sensitivity to factors such as topography, atmospheric turbidity and
illumination angle (Townshend and Justice 1986). These problems will be addressed to
some degree by the satellite sensors currently being developed for the EOS Platforms.
For the present analysis, NDVI (GVI) was used to estimate the amount of active foliar
biomass. The intent was to use available satellite imagery to gain an indication of foliar
biomass which was potentially emitting isoprene.
To create an active foliar biomass density (FBD) surface based on GVI, an equation
relating the two was developed for each vegetation type. Empirical studies have found
GVI to be generally related to the natural logarithm of variables such as leaf area index
(Box et al. 1989, Running 1988, Running et al. 1989). This relationship has been
expressed in the following form:
GVI (or NDVI) = b * ln(X/a)
where
X = Leaf area index, annual evapotranspiration, or net primary production
a, b = empirically determined constants.
IX- 11
-------
Table 1.
Areal extent (Olson et al. 1983), maximum foliar biomass (Box
1981; Cannell 1982), and proportion of biomass emitting isoprene
(Rasmussen and Khalil 1988) for the vegetation types.
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
Vegetation Type
Ice
Desert
Tundra
South Temperate Broad-
Leaved Forest
Grassland
Farms and Towns
Non-Paddy Irrigated
Dryland
Forest/Field/Woods
North Temperate Broad-
Leaved Forest
Cool Conifer & Hardwood
Tropical Montane
Wetlands, Shore & Hinter-
Lands
Woodlands
Warm Conifer
Paddyland
Taiga
Tropical Seasonal Humid
Forest
Tropical/Subtropical
Humid Forest
Cool Conifer
TOTAL:
Area
(106 Km2)
1.24
18.36
11.68
0.71
21.31
12.21
1.57
9.19
0.78
3.54
1.17
3.15
19.87
0.40
1.94
11.50
6.12
4.22
3.09
132.05
Maximum Foliar
Biomass Density
(Kg/m2)
0
0.09
0.10
0.40
0.50
0.50
0.50
0.50
0.35
0.64
0.75
0.80
0.80
0.89
1.00
1.00
1.05
1.05
1.46
Proportion
Emitting
Isoprene
(%)
0
25
20
50
10
10
10
30
50
40
30
25
30
40
10
60
50
50
60
IX- 12
-------
Figure 3. Global Vegetation Distribution (after Olson et al. 1983).
-------
Figure 4. Maximum-value composite of Global Vegetation Index (GVI) data for July, 1988.
-------
Taiga
60
50-
1 4C'"
* 30-
« 20-
10-
0
5%
Mean
95%
Jan Feb Mar Apr May Jun Jul Aug Sept Oct Nov Dec
Temperate Broadleaf Forest
60
50-
§ 40-
1 30-
W 20-
10-
0
I I I I I I I
Jan Feb Mar Apr May Jun Jul Aug Sept Oct Nov Dec
Tropical/Subtropical Broadleaf Forest
60
50 I
S 40-j
•»
> 30-
20-
10
5%
Mean
95%
Jan Feb Mar Apr May Jun Jul Aug Sept Oct Nov Dec
Figure 5. Annual time series of Global Vegetation Index for three vegetation types. All the G VI values
of 0 (unreliable data) were deleted. The values are otherwise area weighted. These graphs
display the mean, 5th, and 95th perccntile values by month for the year 1988. They represent
pnly the northern hemisphere portion of these biomes.
IX- 15
-------
To estimate FEE' from GVI, we inverted this relationship and scaled the observed range of
GVI to literature values for the range of FED within each vegetation type. The resulting
model for the relationship of GVI to FED mimics results of the earlier empirical studies in
that foliar biomass change is most sensitive to GVI at the higher GVI values. The equation
relating GVI and FED was of the form:
FED = a-*e(GVi/b)
where
FED = foliar biomass density
b = '-Jmax ~ Gmin / lri(rmax /
a = Fmax exp(-Gmax / b)
= 95 percentile of GVI values per vegetation type
= 5 percentile of GVI values per vegetation type
Fmax = FED maximum per vegetation type
= FED minimum per vegetation type
FED maximums were taken from Box (1981), Cannell (1982), and Whittaker (1975).
Literature values also indicated a range of foliar biomass of about an order of magnitude
within a vegetati.on type, so Fmin was set an order of magnitude lower than Fmax. Gmax and
Gmin for each vegetation type by month were determined from annual time series of mean
monthly GVI values (as shown in Figure 5). For GVI that are greater than the value of
the 95 percentile, FED is set to a constant value in order to eliminate anomalous
occurrences of exceedingly high GVI that would convert to exaggerated FED. This upper
FED limit was specific for each vegetation type. Figure 6 presents an example of the FED
model used for cool conifer forests.
IX- 16
-------
The global FED surface for the month of July is displayed in Figure 7. Using the FBD for
July in the northern hemisphere and January in the southern hemisphere, total active foliar
biomass using this approach was 60 Pg. This estimate compares well to a published
estimate of 75 Pg (Box 1981).
The percentage of isoprene emitting plants from particular vegetation types was adapted
from Rasmussen and Khalil (1988). Their values were based on surveys of species-
specific isoprene emission rates. A number of such surveys have been published
(Zimmerman 1979, Evans et al. 1982, Cronn et al. 1982, Winer 1989).
Several equations have been developed that relate the rates of vegetation isoprene emission
to air temperature (Tingey et al. 1979, Lamb et al. 1985). These are based on laboratory
chamber studies where temperature is regulated, or field enclosure studies where
temperature is monitored. Comparisons of field emission rates based on enclosure and
non-enclosure techniques reveal reasonable agreement (Lamb et al. 1986). Lamb et al.
(1987) compiled existing data across a broad range of plant types and temperatures and
found the best fit correlation to be the following exponential relationship:
log (Rate) = -0.109 + 0.0416 * T
where
Rate =: isoprene emission rate (ug/g/hr)
T == ambient temperature in C
IX- 17
-------
a
o
5
u
I .5
I . 4
1 . 3
1 . 2
I . I
1 .0
0 . 9
0.8
0. 7
0 .6
0.5
0 . 4
0.3
0. 2
0 . I
0 .0
Foliar Biomass versus GVI
Cool Conifer
|irii|Tiii|-iiii|iiTr|iiii|iri>f T I rTj-riii[iiir|iiii,rT-|i[
5 10 15 20 25 30 35 40 45 50 55 60
GVI * 100
Figure 6. A graphical example of the model used to estimate foliar biomass density
based on observed Global Vegetation Index values. The above case is for the
cool conifer vegetation type. The 5th and 95th percentiles of GVI values are
indicated by stars.
IX- 18
-------
Figure 7. Global distribution of estimated active foliar biomass for July, 1988.
-------
Degrees Latitude versus Hours of Daylight
90° N
45° N- -
•a
3
0-
45° S-
90C S-
March 15
June 15
September 15
December 15
12
16
20
24
Hours of Daylight
Figure 8. Hours of daylight at 0.5° latitude increments. Light curves shown for
the 15th day of four months.
For the present study we have used this relationship across all vegetation types and at a
monthly time step. A global air temperature data base developed by Legates and Willmott
(1990) was used for mean monthly temperatures. The data base is comprised of mean
monthly air temperatures for each 0.5 by 0.5 cell covering the globe. Legates and
Willmott (1990) interpolated the land surface air temperatures into each cell based on data
collected from 17,986 terrestrial air temperature stations. The photoperiod during each
monthly time step (Figure 8) was computed using spherical-geometric equations and
parameters as presented in Sellers (1985).
IX-20
-------
Isoprene Emissions Under A Doubled CC>2 Climate
f
An equilibrium approach, described previously (Turner 1990, King 1990), was used to
estimate future isoprene emissions. Area weighted mean emissions values were assigned to
current vegetation types and potential changes in the areal extent of those types were
determined based on General Circulation Models (GCMs) and current vegetation-climate
correlations. The Holdridge climate-vegetation correlation system and its application to
prediction of vegetation redistribution is discussed in King (1990). Vegetation-specific
emission factors for the Holdridge types were determined by overlaying the current
distribution of the aggregated Holdridge vegetation types (Table 2, Figure 9) on the
current annual emissions surface. All areas within each aggregated Holdridge vegetation
type were used to calculate an area weighted mean emission. These means were then
multiplied by the; areal extent of the associated vegetation type under doubled CO2 climate
scenarios (Table 3, Figure 10) as determined by Leemans (1990). The four GCMs were
those of Oregon State University, Goddard Institute for Space Studies, Geophysical Fluid
Dynamics Laboratory and United Kingdom Meteorological Office.
IX-21
-------
Table 2 Aggregation Scheme for Combining Holdridge
Life Zones into Biomes.
Biome
Holdridge Zone
1 Tundra
2 Cold Parklands
3 Foresi: Tundra
4 Boreal Forest
5 Cool Desert
6 Steppe
7 Temperate Forest
8 Hot Desert
9 Chapparal
10 Warm Temperate Forest
11 Tropical Semi-Arid
12 Tropical Dry Forest
13 Tropical Seasonal Forest
14 Tropical Rain Forest
Ice
Polar Desert
Subpolar Dry Tundra
Subpolar Moist Tundra
Subpolar Wet Tundra
Subpolar Rain Tundra
Boreal Desert
Boreal Dry Scrub
Boreal Moist Forest
Boreal Wet Forest
Boreal Rain Forest
Cool Temperate Desert
Cool Temperate Desert Scrub
Cool Temperate Steppe
Cool Temperate Moist Forest
Cool Temperate Wet Forest
Cool Temperate Rain Forest
Warm Temperate Desert
Warm Temperate Desert Scrub
Subtropical Desert
Subtropical Desert Scrub
Tropical Desert
Tropical Desert Scrub
Warm Temperate Thorn Steppe
Warm Temperate Dry Forest
Warm Temperate Moist Forest
Warm Temperate Wet Forest
Warm Temperate Rain Forest
Subtropical Thorn Woodland
Tropical Thorn Woodland
Tropical Very Dry Forest
Subtropical Dry Forest
Tropical Dry Forest
Subtropical Moist Forest
Tropical Moist Forest
Subtropical Wet Forest
Subtropical Rain Forest
Tropical Wet Forest
Tropical Rain Forest
IX-22
-------
i 3
s-
LEGEND
0) no data
CM Cold Parklands
U> 3) Forest Tundra
_ 4) Boreal Forest
L 5) fool Desert
: 6) Stepp^
7) Tenperate Forest
, 8) Hot IVsf-rt
L_, 9) Chaparral
_,!()) Waim Tiip Forest
11) Trop Semi-Arid
I-') Trop Drv Forest
|"|13) Trop Seas Forest
— 14) Trop Rain Forest
LONGITUDE
Figure 9. Global distribution of modified Holdridge vegetation types for current climatic
conditions (adapted from Leemans, 1990).
-------
Table 3. Annual Isoprene Flux by Different Vegetation Types. Values Are
Area-Weighted Means Rounded to the Nearest 100 Kg/Km2/yr.
Vegetation Type
1 Ice
2 Desert
3 Tundra
4 South Temperate Broad-
Leaved Forest
5 Grassland
6 Farms and Towns
7 Non-Paddy Irrigated
Dryland
8 Forest/Field/Woods
9 North Temperate Broad-
Leaved Forest
10 Cool Conifer & Hardwood
1 1 Tropical Montane
12 Wetlands, Shore & Hinter-
Lands
13 Woodlands
14 Warm Conifer
15 Paddyland
16 Taiga
17 Tropical Seasonal Humid
Forest
1 8 Tropical/Subtropical
Humid Forest
19 Cool Conifer
Mean
(Kg/Km2/yr)
0
500
100
2,000
900
1,000
900
4,300
2,200
4,200
7,700
5,300
7,700
7,800
3,600
3,100
19,400
21,500
7,000
Isoprene Flux
Standard
Deviation
(Kg/Km2/yr)
0
300
100
1,100
700
600
600
3,200
900
2,900
3,400
4,800
3,300
4,200
1,600
800
5,200
5,500
2,500
Total Emissions
(Tg/Yr)
0
10.07
0.97
1.40
19.64
12.32
1.47
39.67
1.74
14.88
9.06
16.84
153.87
3.13
7.04
35.45
118.77
90.87
21.53
IX-24
-------
x
LEGEND
-j 0) no data
i 1) Tundra
S CM Cold Parklands
t 3) Forest Tundra
L-J 4) Boreal Forest
, 5) t'ool Desert
| 6) Steppe
i 7> TTIP Forest
ft) Hot Forest
9) Chappaxal
10) Warm T
-------
RESULTS
Annual isoprene emissions by vegetation type are listed in Table 4. On a per unit land
area basis, emissions are highest from the equatorial humid forests and lowest in the arctic
tundra. There is a general trend towards increasing emissions with increasing foliar
biomass and me;in annual temperature. The standard deviation is a larger proportion of
the mean for the lower biomass vegetation types, probably reflecting the aggregation of
somewhat dissimilar types. The estimated global isoprene emissions for the month of July
1988 is given in Figure 11. Using the methods described above, the global annual isoprene.
emissions for 1988 was estimated at 560 Tg (Figure 12).
The overlay of spatially distributed Holdridge biomes on the annual total emissions surface
(Figure 12) yielded average isoprene fluxes for each biome (column one of Table 4). The
mean fluxes for different biomes are roughly consistent with patterns of biomass and
temperature, with a range from 300-16,000 kg km-2y-i. These means reflect current
patterns of land use and foliar biomass.
The change in areal extent of the different Holdridge vegetation types (Table 5) indicates a
general trend towards reductions in the area of boreal forests and tropical seasonal forests,
and increases in tropical rain forests and temperate forests (Leemans 1990; King et al.
1990, Leemans and Prentice 1990). The global isoprene emission totals (Table 4) for the
vegetation distributions as predicted by the different GGMs were quite similar, ranging
from 670-710 Tg. Because of the relatively high annual emissions from tropical forests,
the increase in their area tends to account for most of the predicted increase in emissions.
IX-26
-------
Table 4. Annual Isoprene Emission (Tg) from Holdridge Life
Current and Doubled CO2 GCM Scenarios.
Smith & Leemans Aggregated
Zones for
Average*
Holdridge Life Zones (Kg/Km2/yr) Current GFDL
1 Tundra
2 Cold Parklands
3 Forest Tundra
4 Boreal Forest
5 Cool Desert
6 Steppe
7 Temperate Forest
8 Hot Desert
9 Chaparral
10 Warm Temperate Forest
1 1 Tropical Semi- Arid
12 Tropical Dry Forest
13 Tropical Seasonal Forest
14 Tropical Rain Forest
TOTAL
297
1373
1446
2807
546
100
2245
984
1898
4128
4197
7519
9859
15823
2.77
3.83
12.87
42.20
2.19
7.41
22.32
20.52
10.60
13.09
40.13
111.75
149.08
133.87
572
0.95
3.89
5.61
26.90
1.66
11.63
26.63
20.33
14.07
8.09
58.72
147.17
98.73
243.85
668
GISS
1.27
3.27
8.49
37.88
1.29
6.96
30.15
17.35
10.33
7.88
70.26
145.51
77.74
273.91
692
OSU
1.42
3.69
8.68
39.70
1.74
8.72
25.96
19.12
9.29
10.11
50.95
111.75
100.01
316.95
708
UKMO
0.86
2.35
4.90
28.58
1.14
7.20
29.14
19.63
16.27
11.89
69.80
195.90
75.38
203.50
666
* Average isoprene flux for individual life zones.
IX-27
-------
LEGEND
2> 510-1 (XX) kc/TariL'
3) 1001-1500 ks/lait;
4) 1501-CrCKX) k.«./kni.'
, 5) 2
-------
/
! -
C
LEGEND
J 0) 0 ke/kn£
!i 1 i I --'000 kg/krcC
! 2) 2001-4000 kg/knii
JSJ 3) 4001-«000 kg/knC
4) 8001-8000 kc/kni.1
'„ 5) 8001-10000 kgAnt:
] (J) 10001-13000 ke/\a£
£J 7) 12001 -1 4000 kgAni-'
8) 14001 -IttOOO keAiii;
;>) 1«001- 18000 kg/kn£
10) 18001-20000 k«/kniJ
00-220
j_|ll) 20001-22000 kpAnC
- '
L13) 24001-2ROOO kgAra2
•114) 20001-28000 ke/krrf-'
• 15> 28001-30000 kg/koi.1
LONG I TIDE
Figure 12. Global distribution of total annual isoprene emissions for 1988.
emission summed over the globe equalled 560 Tg.
Total
-------
Table 5 Changes in Areal Extent of Different Vegetation
Types as Predicted by 4 GCMs.
Area
Biome (106 Km2)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
Tundra
Cold Farmlands
Forest Tundra
Boreal Forest
Cool Desert
Steppe
Temperate Forest
Hot Desert
Chaparral
Warm Temperate Forest
Tropical Semi-Arid
Tropical Dry Forest
Tropical Seasonal Forest
Tropical Rain Forest
9.30
2.79
8.90
15.03
4.01
7.39
9.94
20.85
5.58
3.17
9.56
14.86
15.13
8.46
Difference in Biome Area (106 Km2)
GFDL
-6.11
0.03
-5.02
-5.45
-0.97
4.20
1.92
-0.20
1.83
-1.22
4.43
4.71
-5.11
6.95
GISS
-5.05
-0.41
-3.03
-1.54
-1.67
-0.46
3.49
-3.22
-0.13
-1.25
7.18
4.49
-7.24
8.85
OSU
-4.56
-0.10
-2.90
-0.89
-0.82
1.30
1.63
-1.42
-0.69
-0.72
2.58
-0.00
-4.98
11.57
UKMO
-6.43
-1.09
-5.50
-4.85
-1.93
-0.21
3.04
-0.92
2.99
-0.29
7.07
11.19
-7.48
4.40
Total: 134.97 x 106 Km2
IX-30
-------
DISCUSSION -
Uncertainty Analysis
The largest uncertainties in this analysis lie in the estimation of active foliar biomass, the
use of a mean monthly temperature rather than a diurnal temperature cycle, the
application of the same temperature/emissions relationship to all emitting vegetation, and
the estimation of the vegetation-specific proportion of foliar biomass which emits
isoprene.
The approach in mis-paper insures that foliar biomass is not overestimated in comparison
with literature values. However, it is just a first step in getting away from assigning a
constant foliar biomass to a given vegetation type. Continued research linking remote
sensing to vegetation characteristics is needed. Specific uncertainties include the shape of
the curve relating GVI to FED, which may not be the same for all vegetation types and the
same at all times of the year. In addition, active foliar biomass would tend to be
overestimated using the approach described here in that the scheme by which GVI is
assigned to each pixel selects for the greenest (highest NDVI) 1 Km2 pixel within a 16 x 16
Km square. Thus spatial heterogeneity at scales less than 16x16 Km is not accounted for.
Higher resolution NDVI data sets are available and an important sensitivity analysis to be
conducted in the near future will be to compare results for limited areas using a 1 Km2
spatial resolution and the results using GVI.
The use of mean monthly temperatures instead of hourly temperatures to drive emission
rates is likely to produce fluxes which are artificially low. Since the relationship of
emissions to temperature is exponential, a model using an hourly time step and a diurnal
temperature cycle, which had the mid-day high temperatures, would result in higher
predicted emissions. We have recently obtained a global database with daily temperature
ranges and that will allow us to more fully evaluate the impact of this factor on regional
and global emissions estimates.
IX-31
-------
Because cloudiness is not accounted for, the light regime used in this model will tend to
produce an overestimate of emissions. This factor is balanced to some extent by not
attempting to correct for low GVIs due to cloudiness. Estimates of cloud cover for
specific sites, and globally, are available and may be used to evaluate the sensitivity of the
emissions estimates to this factor.
There is no obvious direction which the use of the same relationship between temperature
and emissions for all vegetation types would tend to drive emission estimates. Studies are
in progress to extend the range of observations to a greater number of species. If there
are clear patterns with respect to vegetation types we will develop separate curves for each
vegetation type. The uncertainty relative to the proportion of biomass emitting isoprene in
each vegetation type is bounded by the range from 10 - 60 %. In general, the limited
number of field and laboratory measurements on which the vegetation factors in this
model are based., highlights the need for extensive and intensive measurement programs.
There are of course large uncertainties in current estimates of the magnitude of climate
change to be expected under doubled-COi conditions. These uncertainties relate to oceanic
factors and to feedbacks involving water vapor, clouds, sea ice, and the biosphere. There
are likewise large uncertainties in using vegetation-climate correlations approaches to
predicting the distribution of vegetation types (King 1990). In both cases research is
progressing on a broad front to reduce uncertainties. In the meantime, sufficient progress
has been made to begin making the type of analysis described in this paper.
Comparisons to Other Estimates
The estimate for global annual isoprene emission using the present model was 560 Tg,
expressed as carbon. This estimate compares with earlier estimates of 350 Tg
(Zimmerman et al. 1978) and 450 Tg (Rasmussen and Khalil 1988). Zimmerman et al.
(1978) based their estimate on a fixed ratio between isoprene production and net primary
productivity (0.7%) and a global net primary productivity of 118 Gt yr-i. Rasmussen and
Khalil (1988) broke out the land surface by vegetation type, then used literature values for
IX- 32
-------
foliar leaf surface area, reduced that by their estimate of the proportion of non-isoprene
emitters, estimated annual hours of daylight during the growing season, and predicted
emissions by applying a universal emission factor (648 g km-2 of leaf area hr-i). At a
continental scale, our estimates are also generally greater than those of earlier
investigators for the U.S. (Zimmerman 1979b, Lamb et al. 1987) but lower than an
estimate made by Ayers and Gillett (1988) for tropical Australia (16 Tg vs 25 Tg).
Others have estimated hydrocarbon emissions by using atmospheric chemistry models to
calculate how large a CO source would be needed to balance the global CO budget.
Crutzen and Gidel (1983) estimated a CO source on the order of 1400 Tg (as carbon)
which was thought to originate largely from vegetation NMHCs. Similarly derived
estimates ranged from 560 - 1250 Tg in the study of Logan et al. (1981). Assuming that
the CO yield of isoprene.oxidation is two CO molecules per isoprene molecule (Logan et
al. 1981) and about half of the global NMHC-CO is from isoprene oxidation (Zimmerman
et al. 1978), then the associated isoprene source expressed in terms of carbon mass would
be 700 - 1750 Tg. Hanst et al. (1980) and Zimmerman et al. (1978) have suggested higher
CO yields for isoprene oxidation. If the yield is actually more on the order of four CO
per isoprene, then the derived source would range from 350 - 875 Tg. The 560 Tg
estimate from our model falls between an apparent lower bound based on extrapolating
from enclosure studies and an upper bound based on balancing the global CO budget.
Only a few estimates of isoprene emissions have been made on local scales using
atmospheric concentrations rather than extrapolation from enclosure studies. Zimmerman
et al. (1988) estimated the isoprene flux from a "tropical forest" based on changes in
concentration in the boundary layer under a capping inversion over a diurnal time frame.
His estimate was 25 mg.m-2 d-i during July. The equivalent estimate using an emissions
model coupled to a photochemical model was 38 mg m-2 d-i (Jacob and Wofsy 1987). Our
July estimates for tropical forests ranged from means of 45 to 54 mg m-2 d-i. As more
measurements of ambient concentrations are made there will be increased opportunity for
calibration and validation of this emissions model.
IX-33
-------
Perhaps as important as getting another estimate of total annual emissions, the present
model provides the spatial and temporal patterns in global isoprene emissions at a
reasonably high spatial resolution. The value of this information is in revealing areas
where more intensive study is needed and providing the basis for hypotheses that can be
tested via field measurements. These emission surfaces can also be used in globally
distributed atmospheric chemistry models to begin evaluating the role of biogenic
hydrocarbon emissions in local and global tropospheric chemistry.
Doubled-CCh Climate Emissions
The 150 Tg increase in annual isoprene emissions that is predicted for the doubled-COi
climate scenarios represents an increase in annual emissions of about 25%. To a great
degree that increase is concentrated at tropical latitudes. It seems likely that total
nonmethane hydrocarbons (TNMHC) will also increase because of the similar relationship
of emissions to vegetation type. If they increased at a rate similar to isoprene alone,
TNMHC emissions could rise by 400 or more Tg, using the mid range values of Logan et
al. (1985) for current emissions. Back calculating to CO produced, as in the analyses
above, the total increased CO source due to NMHC could be over 200 Tg.
The potential impact on the climate system of increased NMHC emissions is difficult to
assess. The added CO might not be large relative to the global CO budget, but it will
probably be accompanied by increased CO from other sources and contribute to a
downward trend in on OH concentration (Crutzen 1988) and increased atmospheric
lifetime of some greenhouse gases. Possible compensatory factors include increases in
concentrations of water vapor and ozone which would favor OH production. However,
there is great uncertainty about the interrelationships between these various factors
(Thompson et al. 1989, Thompson et al. 1990, Lelieveld and Crutzen 1990).
As noted previously, photochemical production of ozone is largely controlled by the
availability of NOX. Results of the GTE-ABLE (Global Tropospheric Experiment-
Atmospheric Boundary Layer Experiment) studies in the Amazon Basin suggested that the
IX-34
-------
strength of the isoprene flux has little local influence on ozone production because of high
background levels of CO and isoprene (Jacob and Wofsy, 1987). Limited NOX apparently
restricts the rate of ozone formation such that local ozone levels rise to only about 20 ppm
during the day. Thus, a greater areal extent of tropical forest would probably not produce
substantial increases in ozone formation unless the ambient NOX levels were to increase.
Not considered here in detail, but of potential importance to the total tropospheric ozone
burden and climate, is the transport of NMHC to the free troposphere (that area above the
PBL and relatively isolated from the surface, i.e. low deposition rates) by convection,
especially in the tropics (Garstang et al. 1988). This convection may be accompanied by
lightning, which forms NOX, leading to potentially significant ozone production in the
upper atmosphere above the tropics. Thus the normally low NOX and short lifetime for
isoprene that are expected over the Amazon basin would not apply to the NMHC
transported upward via convection. The GTE-ABLE results, discussed above, apply to the
dry season when this upward transport is at a minimum. A second GTE-ABLE
experiment was to examine wet season fluxes and transport. If coupling between the
Amazon biosphere (NMHC flux) and the global troposphere is established, it is likely that
the importance of NMHC to tropospheric ozone production, and hence climate feedbacks,
will increase. Any additional ozone formation in the free troposphere is very effective in
causing an increase in surface temperature (Wuebbles et al. 1989).
Overall, these analyses suggest a positive feedback to climate warming is likely via impacts
on biogenic emissions of NMHC. Progress in evaluating the potential change in the
oxidation state of the troposphere and the concentrations of greenhouse gases will come in
part from improved 3-D atmospheric chemistry models. Specific problem areas include
the spatial and temporal pattern in source strengths for reduced species, the kinetics and
products of NMHC oxidation, the heterogeneous chemistry of cloud droplets, and patterns
in vertical and horizontal transport. Increased coupling of general circulation models,
atmospheric chemistry models and ecosystem emissions models will promote rapid
advances on these fronts in the coming years.
IX-35
-------
Transient Increases in NMHC Emissions
The equilibrium approach to estimating isoprene emissions under a doubled-CO2 climate
has several obvious limitations (see Turner 1990) including the fact that there may be a
significant transient effect. Trees often live hundreds of years, yet climate change will
occur on a much shorter time scale. As temperatures rise, climate will gradually drift out
of equilibrium with the vegetation. Given the exponential increases in NMHC emissions
with increases in temperature, there may be many situations where NMHC emissions rise
substantially before the vegetation changes.
An evaluation of this prospect requires process based models which can be run using
particular climate scenarios. The Global Change Research Program at ERL-C is currently
developing such a model. The Forest BGC model (Running and Couglan 1988), which
accounts for hydrology, nutrient cycling and photosynthesis, is being modified to include
NMHC emissions. Laboratory and field measurements are now underway to develop
response surfaces relative to light and temperature for particular tree species. Ultimately
the model will be used for site-specific simulations and will be spatially distributed using
remote sensing to initialize variables such as vegetation type and foliar biomass. The
prospects of coupling a GCM to an ecosystem model that predicts NMHC emissions are
good.
Conclusions
Isoprene emissions have significant effects on atmospheric chemistry and ultimately on
climate via their influence on the concentrations of greenhouse gases such as methane and
ozone. The global model presented here for isoprene emissions under the current climate
indicates that highest emissions occur at low latitudes, particularly hi wet tropical forests.
Equilibrium analyses of potential vegetation change suggest that emissions under a doubled-
CO2 climate may rise by about 25%, based largely on predicted increases in the areal
extent of wet tropical forests. Higher levels of isoprene emission are expected to increase
concentrations of methane and ozone but 3-D atmospheric chemistry models are needed to
IX-36
-------
evaluate these effects. Future studies incorporating transient responses and impacts of land
use practices are needed to refine these equilibrium analyses.
Acknowledgements
Special thanks to Tom Smith and Rik Leemans for sharing their doubled-CO2 climate data
used here for the vegetation redistribution analysis.
IX-37
-------
References
ACE. 1988. GRASS Users and Programmers Manual. U.S. Army Corps of Engineers
Construction Engineering Research Laboratory, Champaign, Illinois.
Asrar, G., M. Fuchs, E.T. Kanemasu, and J.L. Hatfield. 1984. Estimating absorbed
photosynthesis radiation and leaf area index from spectral reflectance in wheat. Agron. J.
76:300.
Ayers, G.P. and R.W. Gillett. 1988. Isoprene Emissions from vegetation and
hydrocarbon emissions from brushfires in tropical australia. J. Atmos. Chem. 7:177-190.
Box, E.G., B.N. Holben & V. Kalb. 1989. Accuracy of the AVHRR vegetation index as a
predictor of biomass, primary productivity and net CO2 flux. Vegetation. 80:71-89.
Box, E.G. 1981. Foliar biomass: Data base of the international biological program and
other sources. In: J.J. Bufalini and R.R. Arnts, Eds., Atmospheric Biogenic
Hydrocarbons, Vol. 1. Ann Arbor Science Publishers, Inc., Ann Arbor, MI.
Cannell, M.G.R., ed. 1982. World Forest Biomass and Primary Production Data.
Academic Press. New York, NY.
Cronn, D.R., and Nutmagul, W. 1981. Analysis of atmospheric hydrocarbons during
winter MONEX. Tellus 34:159-165.
Crutzen, P.J. 1988. Variability in atmospheric-chemical systems. In: T.Rosswall, R.G.
Woodmansee arid P.G. Risser, eds. Scales and Global Change. John Wiley & Sons Ltd.,
New York, NY.
Crutzen, P. J., and L. T. Gidel. 1983. A two-dimensional photochemical model of the
atmosphere 2: The tropospheric budgets of the anthropogenic chlorocarbons CO, CH4,
IX-38
-------
CH3C1 and the effect of various NOX sources on tropospheric ozone. J. Geophys. Res.
88:6641-6661.
Delany, A.C., P.J. Crutzen, P. Haagenson, S. Walters, and A.F. Wartburg. 1985.
Photochemically produced ozone in the emission from large-scale tropical vegetation fires.
J. Geophys. Res. 90:1425-2429.
Dignon, J., and S. Hameed. 1985. A model investigation of the impact of increases in
anthropogenic NOX emissions between 1967 and 1980 on tropospheric ozone. J. Atmos.
Chem. 3:491-506.
Evans, R.C., D.T. Tingey, M.L. Gumpertz, and W.F. Burns. 1982. Estimates of isoprene
and monoterpene emission rates in plants. Bot. Gaz. 143:304.
Garstang, M., J. Scala, S. Greco, R. Harriss, S. Beck, E. Browell, G. Sachse, G. Gregory,
G. Hill, J. Simpson, W.-K. Tao, and A. Torres. 1988. Trace gas exchanges and convective
transport over Amazonian rain forest. J. Geophys. Res. 93:1528-1550.
Goward, S.N. 1989. Satellite Bioclimatology. J. Climate 2:710-720.
Goward, S.N., C.J. Tucker, and D.G. Dye. 1985. North American vegetation patterns
observed with the NOAA-7 advanced very high resolution radiometer. Vegetation 64:3-
14.
Guthrie, P.D. In Press. The CtLj-CO-OH Conundrum: A simple analytic approach.
Biogeochemical Cycles.
Hanst, P.L., J.W. Spence and E.O. Edney. 1980. Carbon monoxide production in
photooxidation of organic molecules in the air. Atmos. Environ. 14:1077-1088.
Harborne, J.B. 1988. Introduction to Ecological Chemistry. Academic Press, New York.
IX-39
-------
Holben, B.N. 1986. Characteristics of maximum-value composite images from temporal
AVHRRdata. Int. J. Remote Sensing 7:1417.
Holben, B.N., and R.S. Fraser. 1984. Red and near-infrared response to off-nadir
viewing. IntJ. Remote Sensing 5:145.
Isaksen, I.S.A., ed. 1988. Tropospheric Ozone, Regional and Global Scale Interactions.
D.Reidel Publishing Co., Dordrecht, FRG.
Jacob, D.J., and S.C. Wofsy. 1988. Photochemistry of biogenic emissions over the
Amazon forest. .1. Geophys. Res. 93:1477-1486.
Kavanaugh, M. 1987. Estimates of future CO, NiO and NOX emissions from energy
combustion. Atmos. Environ. 21:3463-3468.
Khalil, M.A.L., and R.A. Rasmussen. 1990. The global cycle of carbon monoxide:
Trends and mass balance. Chemosphere 20:227-242.
Kidwell, K.B. 1984. NOAA polar orbital data user's guide. NOAA National Climate
Data Center, Washington, DC.
Kidwell, K.B. 1990. Global Vegetation Index User's guide. NOAA National Climate Data
Center, Washington, DC.
King, G.A. 1990. Effects of global climate change on global vegetation. Contained within
this document.
Lamb, B., A. Guenther, A. Gay, and H. Westberg. 1987. A national inventory of
biogenic hydrocarbon emission. Atmos. Environ. 21:1695-1705.
Lamb, B., H. Westberg, G. Allwine, and T. Quarles. 1985. Biogenic hydrocarbon
emissions from deciduous and coniferous trees in the U.S. J. Geophys. Res. 90:2380.
IX-40
-------
Lamb, B., H. Westberg, and G. Allwine. 1986. Isoprene emission fluxes determined by
an atmospheric tracer technique. Atmos. Environ. 20:1-8.
Legates, D.R., and C.J. Willmott. 1990. Mean seasonal and spatial variability in global
surface air temperature. Theor. Appli. Climatol. 41:11-21.
Leemans, R. 1990. Possible changes in natural vegetation patterns due to global warming.
IIASA WP-90-08, International Institute for Applied Systems Analysis.
Leemans, R. and I.C. Prentice. 1990. Possible changes in natural vegetation patterns
using GCM climate scenarios. In preparation.
Lelieveld, J. and P.J. Crutzen. 1990. Influences of cloud photochemical processes on
tropospheric ozone. Nature 343:227-233.
Lieth, H., and R.H. Whittaker. 1975. Primary productivity of the biosphere. Springer-
Verlag. New York.
Liu, S. C., M. Trainer, F. C. Fehsenfeld, D. D. Parrish, E. J. Williams, D. W. Fahey, G.
Hubler, and P. C. Murphy. 1987. Ozone production in the rural troposphere and the
implications for regional global ozone distributions. J. Geophysical Research 92:4191-
4207.
Logan, J. A. 1983. Nitrogen oxides in the troposphere: global and regional budgets. J.
Geophys. Res. 88:10,785-10,807.
Logan, J.A. 1985. Tropospheric ozone: Seasonal behavior, trends and anthropogenic
influences. J. Geophys. Res. 90:10463:10482.
Logan, J.A., MJ. Prather, S.C. Wofsy, and M.B. McElroy. 1981. Tropospheric ozone: A
global perspective. J. Geophys. Res. 86:7210-7254.
IX-41
-------
Monson, R. K., and R. Fall. 1989. Isoprene emission from Aspen leaves: Influence of
environment and relation to photosynthesis and photorespiration. Plant Physiol. 90:267-
274.
NASA. 1988. Earth System Science: A Closer Look. Report of the Earth System
Sciences Committee. NASA. 208pp.
NRC. 1984. Global Tropospheric Chemistry: A Plan for Action. National Academy
Press, Washington, DC. 194pp.
Olson, J.S., J.A. Watts, and L.J. Allison. 1983. Carbon in live vegetation of major world
ecosystems. ORNL-5862. Oak Ridge National Laboratory, Oak Ridge, TN. 180 pp.
Ramanathan V., R.J. Cicerone, H.B. Singh, and J.T. Kiehl. 1985. Trace gas trends and
their potential role in climate change. J. Geophys. Res. 90:5547-5566.
Rasmussen, R.A., and M.A.K. Khalil. 1988. Isoprene over the Amazon basin. J.
Geophys. Res. 93:1417-1421.
Rasmussen, R.A., and F.W. Went. 1965. Volatile organic material of plant origin in the
atmosphere. Proc. Nat. Acad. Sci. 53:215-220.
Running, S.W., and R.R. Nemani. 1988. Relating seasonal patterns of the AVHRR
vegetation index to simulated photosynthesis and transpiration of forests in different
climates. Remone Sens. Environ. 24:347-367.
Running, S.W., R.R. Nemani, D.L. Peterson, L.E. Band, D.F. Potts, L.L. Pierce, and
M.A. Spanner. 1989. Mapping regional forest evapotranspiration and photosynthesis by
coupling satellite data with ecosystem simulation. Ecology 70:1090-1101.
IX-42
-------
Running, S.W., and J.C. Coughlan. 1988. A general model of forest ecosystem processes
for regional applications. I.Hydrological balance, canopy gas exchange and primary
production processes. Ecol. Mod. 42:125-154.
Sellers, P.J. 1985. Canopy reflectance, photosynthesis and transpiration. Int. J. remote
Sensing 6:1335.
Sellers, W.D, 1965. Physical Climatology. University of Chicago Press, Chicago, 272 pp.
Singh, H. B. 1987. Reactive nitrogen in the troposphere. Environ. Sci. Technol. 21:320-
327.
Talbot, R.W., M.O. Andreae, T.W. Andreae, and R.C. Harriss. 1988. Regional aerosol
chemistry of the Amazon Basin during the dry season. J. Geophys. Res. 93:1499-1508.
Tarpley, J.P., S.R. Schneider, and R.L. Money. 1984. Global vegetation indices from
NOAA-7 meteorological satellite. J. Climate appli. Meteorol. 23:491.
Thompson, A.M., and RJ. Cicerone. 1986. Possible perturbations to atmospheric CO,
CHiandOH. J. Geophys. Res. 91:10853-10864.
Thompson, A.M.., R.W. Stewart, M.A. Owens, and J.A. Herwehe. 1989. Sensitivity of
tropospheric oxidants to global chemical and climate change. Atmos. Environ. 23:519-
532.
Thompson, A.M., M.A. Huntley, and R.W. Stewart. 1990. Perturbations to tropospheric
oxidants, 1985-2035. 1. Calculations of ozone and OH in chemically coherent regions. J.
Geophys. Res. 95:9829-9844.
Tingey, D.T., M. Manning, L.C. Grothaus, and W.F. Burns. 1979. The influence of light
and temperature on isoprene emission rates from live oak. Physiol. Plant. 47:112-118.
IX-43
-------
Townshend, J.R.G., and C.O. Justice. 1986. Analysis of the dynamics of African
vegetation using the normalized difference vegetation index. Int. J. remote Sensing
7:1435.
Trainer, M., E. J. Williams, D. D. Parrish, M. P. Buhr, E. J. Allwine, H. H. Westberg, F.
C. Fehsenfeld, and S. C. Liu. 1987. Models and observations of the impact of natural
hydrocarbons on rural ozone. Nature 329:705-707.
Tucker, C.J., C.O, Justice, and S.D. Prince. 1986. Monitoring the grasslands of the Sahel
1984-1985. IntJ. Remote Sensing 7:1571.
Tucker, C.J., and P.J. Sellers. 1986. Satellite remote sensing of primary production. Int.
J. Remote Sensing 7:1395.
Turner, D.P. 1990. Effects of climate change on carbon storage in terrestrial ecosystems:
equilibrium analysis at the global level. (Contained within this Document)
Wardley, N.W. 1984. Vegetation index variability as a function of viewing geometry.
IntJ. Remote Sensing. 5:861-870.
Whittaker, R.H. 1975. Communities and Ecosystems, 2nd Edition. MacMillan Publishing
Co., New York, NY. 385 pp.
Winer, A.M. 1989. Hydrocarbon Emissions from Vegetation Found in California's
Central Valley. Final Report, A732-155, for California Air Resources Board,
Sacramento, CA.
Wuebbles, D.J., K.E. Grant, P.S. Connell, and J.E. Penner. 1989. The role of
atmospheric chemistry in climate change. J. Air Poll. Control Agency. 39:22-28.
IX-44
-------
Zimmerman, P., Chatfield, R.B., Fishman, J., Crutzen, P.J. and Hanst, P.L. 1978.
Estimates on the production of CO and H2 from the oxidation of hydrocarbon emissions
from vegetation. Geophy. Res. Lett. 5:679-682.
Zimmerman, P.R. 1979a. Tampa Bay Area Photochemical Oxidant Study: determination
of emission rates of hydrocarbons from indigenous species of vegetation in the Tampa/St.
Petersburg, Fl. area. Final Appendix C, February 2, 1979. EPA 904/9-77-028.
Zimmerman, P.R,, R.B. Chatfield, J. Fishman, P.J. Crutzen, and P.L. Hanst. 1978.
Estimates on the production of CO and H2 from the oxidation of hydrocarbon emissions
from vegetation. Geophys. Res. Lett., 5:679.
Zimmerman, P.R., J.P. Greenberg, and C.E. Westberg. 1988. Measurements of
atmospheric hydrocarbons and biogenic emission fluxes in the Amazon boundary layer. J.
Geophys. Res. 93:1407-1416.
IX-45
------- |