United States
Environmental Protection
Agency
Atmospheric Research and
Exposure Assessment Laboratory
Research Triangle Park, NC 27711
Research and Development
EPA/600/S3-91/049 Nov. 1991
EPA Project Summary
Temperature Scenario
Development Using Regression
Methods
Peter J. Robinson
A method of developing scenarios of
future temperature conditions resulting
from climatic change is presented. The
method Is straightforward and can be
used to provide Information about daily
temperature variations and diurnal
ranges, monthly average high and low
temperatures, and the frequency, with
which user-selected high- and low -tem-
perature thresholds are crossed. Linear
regressions between monthly average
temperature and these various attributes
are established by using the observa-
tional record of dally maximum and mini-
mum temperature. These regressions
are then used to estimate values from
the monthly average temperatures esti-
mated by General Circulation Models to
occur as a result of a doubling of atmo-
spheric CO2. Values can be established
for any location having daily tempera-
ture records. For the United States the
station density is sufficient to allow the
creation of detailed regional scenarios
on the spatial and temporal scales re-
quired for Impact assessment. The as-
sumptions, scientific and statistical, in-
herent in this regression-based approach
are reviewed. The method has been in-
corporated In a self-contained PC-based
computer program requiring only the
actual temperature data to be input by
the user. A demonstration of the use of
the program, incorporating discussion
of techniques for evaluating the quality
of the resultant scenario, is provided.
This Project Summary was developed
by EPA's Atmospheric Research and
Exposure Assessment Laboratory,
Research Triangle Park, NC, to announce
key findings of the research project that
Is fully documented In a separate report
of the same title (see Project Report
ordering Information at back).
Introduction
Concern about the potential impacts of
climatic change on the environment and on
human activities has created the need for
estimates of futureclimaticconditions. Since
it is not possible to produce "forecasts"
analogous to daily weather forecasts, it is
necessary to suggest future conditions in
terms of scenarios. A climate scenario is a
possible future condition, but becausemany
assumptions must be used in scenario de-
velopment, no single scenario can give a
definitive prediction of the future. Hence
sets of scenarios, each developed on sound
scientific principles but with no definable
probability of occurrence attached, are usu-
ally produced to indicate the likely range of
possible future conditions. This set of sce-
narios must contain sufficient details for
assessment of possible impacts of climatic
changes. Such assessments could be used
by decision makers to develop strategies to
meet the challenge and opportunities posed
by climatic change.
Most impact-related scenario develop-
ment in the past has been undertaken by a
trained climatologist in close association
with those responsible for assessing the
impacts. As the need for and number of
such assessments increases, however, the
involvement of a climatologist cannot be
guaranteed. Consequently, methods must
be developed that the impact assessors
themselves can use to create and interpret
scenarios. Care must be taken to ensure
that the development and evaluation of the
resulting scenario leads to scientifically jus-
tifiable conclusions. Because no single sce-
nario development technique will meet all
impact assessment needs, a variety of ap-
Printed on Recycled Paper
-------
preaches must be made available. This
study, with the associated software, pro-
vides a method of assessing focal tempera-
ture changes associated with globa\ cli-
matic change. The scenario program that is
described and documented, TEMPSCEN,
is the first program designed specifically to
allow non-climatologists to develop sce-
narios for their own needs in their own local
areas.
The program discussed in the present
document creates scenarios of future tem-
perature conditions for any stations in the
United States having four or more years of
daily maximum and minimum temperature
observations. The scenarios are for a time
when the atmospheric concentration of
greenhouse gases has doubled, currently
estimated to occur sometime towards the
middle of the next century. Regression rela-
tionships are developed between monthly
average temperature and various tempera-
ture attributesfrom observational data which
are provided by the user. These relation-
ships are then used to estimate new values
of the attributes in the monthly average
temperatures postulated for the future by
general circulation models (GCMs). Mod-
eladtemperatures from fourdifferent GCMs
are provided as part of the program pack-
age. At each stage of scenario develop-
ment, opportunities are provided to assess
the strength of the various relationships and
thus the validity and reliability of the result-
ing scenarios.
The temperature conditions, called tem-
perature attributes, considered are the fol-
lowing.
* monthly standard deviation of daily tem-
perature
• monthly average diurnal temperature
range
• monthly mean of daily maximum tem-
perature
« monthly mean of daily minimum tem-
perature
• number of days in the month with maxi-
mum temperature above a user-se-
lected threshold
• number of days in the month with mini-
mum temperature below a user-selected
threshold
The scenario development program itself
is a personal computer-based software pro-
gram; it is menu driven and largely self-
explanatory. The project final report is de-
signed to be used in conjunction with that
software. Itprovidesthe rationaleforchoos-
Ing a particular scenario development tech-
nique, and an assessment of the scientific
problems associated with the one selected.
information on installation and use is given.
A scenario development example is shown.
This example is presented in detail, placing
the somewhat theoretical concerns consid-
ered earlier into a practical context. It pro-
vides the information necessary to ensure
that the technique is used correctly from the
scientific and climatological perspective, so
that valid conclusions can be drawn.
Climate Scenarios
To meet the overall objective of produc-
ingascenariodevelopmenttechnique which
could be used legitimately by impact asses-
sors without outside assistance, the tech-
nique had to
• meet the user's need for specific infor-
mation,
• be capable of use and evaluation by
non-climatologists,
• use easily understood and scientifically
sound techniques,
• use readily available data,
• be applicable to any region of the USA,
• be widely and readily available.
The technique was developed as a PC-
based computer program. In this form it can
be made widely and readily available. Fur-
ther, a variety of U.S. climatic data are
already available in PC-compatible format.
The program is self-contained, needing
only the temperature data of the area of
interest as input. Included in the program
are procedures for assessing the scientific
validity of the results. The project report
amplifies and explains these procedures,
so that the program and documentation
together ensure that the resulting scenario
can be correctly developed and wisely in-
terpreted.
Development Options
Selection of the specific need to be ad-
dressed was based on the results of an
earlier assessment of the strategies forsce-
nario development. That study showed that
those involved in impact assessments had
a clear need for scenarios on local (100 km)
spatial scales containing information on the
frequency with which specific temperature
thresholds were exceeded.
Even with this well-defined need, there
are many possible scenario development
approaches. Most of these approaches were
assessed fortheir ability to meet each of the
program constraints indicated above. The
most suitable was deemed to be a method
using historical observational records to
develop a regression relationship between
the number of threshold crossings and
monthly average temperature. This rela-
tionship could then be used with GCM-
based estimates of future monthly average
temperatures to determine the frequency of
threshold crossings in the changed climate.
The technique used in this approach is
linear regression, a technique almost cer-
tainly familiar to anyone undertaking an
impact assessment. Correlation coefficients
and significance levels for the relationship
are given, and thus the user is able to make
judgments about the quality of the resulting
scenario. Further, the user is also likely to
be in a position to understand the scientific
basis for the method and the scientific un-
certainty associated with its use. Study of
many other nonlinear regression techniques
failed to show that any of them offered
significant improvement over the simple
linear model.
The only data required of the method are
daily maximum and minimum temperatures
forthe stations in the area of interest and the
GCM estimates of future temperatures in
that area. These data are available for any
region of the world. The current version of
the program is restricted to analysis in the
United States. The appropriate GCM data
are provided as part of the program pack-
age. The daily temperature data are widely
available in a number of forms. The pro-
gram is tailored to use temperature data
downloaded from optical disks directly, but
other formats can be accepted, although
they may need some off-line formatting.
The program and method were designed
to emphasize the development of scenarios
for temperature threshold crossings, since
this meets a well defined need in impact
assessment. However, the method allows
calculation of a variety of othertemperature
attributes. The attributes were chosen as a
mixture of those having application in im-
pact assessment and those useful for ex-
ploring the validity of the method and the
character of future climates. The attributes
calculated, with suggestions of their role,
are as follows.
1. Standard deviation of daily temperature
within month:
This explores, within the limits of the
method, the way in which the day-to-
day variability of temperature changes
as the mean temperature changes.
There are indications for some areas
that the variability decreases as the
mean temperature increases. The vari-
ability may give clues to the type of
synoptic weather changes likely under
a changed climate.
2. Monthly average diurnal temperature
range:
This explores how the monthly mean
daily temperature range varies as a
function of monthly mean temperature.
Present GCMs suggest that the green-
-------
house effect should influence minimum
temperatures more than maximum, so
that if this method indicates a decreas-
ing diurnal range with increasing mean
temperature, it suggests a degree of
realism in the scenario.
3. Monthly mean of daily maximum tem-
perature:
The monthly mean of the daily maxi-
mum temperature is often of concern
for energy-related applications, includ-
ing power generation and air condi-
tioner unit sizing. It is also useful in
ecological studies. Consequently,
changes as the monthly mean tem-
perature increases have the potential
for direct impact-oriented scenario de-
velopment.
4. Monthly mean of daily minimum tem-
perature:
This is the converse of the previous
attribute, again providing information
potentially useful for energy and eco-
logical relationships.
5. Number of days per month with maxi-
mum temperature above selected
threshold:
The threshold is selected by the user.
This is the attribute with the most clearly
defined applications in, for example,
agricultural and energy impact assess-
ments.
6. Number of days per month with mini-
mum temperature below selected
threshold:
This threshold is also user-selected.
This is essentially the same attribute as
numbers, and may have applicability in
similar impact assessments.
Regression-Based Scenarios
The scenarios developed here use a com-
bination of GCM results and observational
data. Most of our knowledge about future
climate comes form GCMs. These models,
while the most sophisticated tools avail-
able, can give only general indications of
average conditions over broad regions.
Local, specific detail is often needed, how-
ever, if potential impacts of social and eco-
nomic importance are to be assessed. This
local information is derived from past obser-
vations of the climatic elements. The model
and observational results are then com-
bined to produce scenarios.
The regression approach was chosen as
the scenario development technique be-
cause it came closest to meeting all the
criteria for the project. The choice does not
imply that the approach is always the most
appropriate or scientifically sound. Certainly
when the user's concern is with thresholds
which are extreme values, techniques us-
ing extreme value or time series analyses
are likely to be much more appropriate.
Indeed, a strength of the present method is
that when correctly used it precludes use of
inappropriate results by indicating when the
regression relationships are unsound.
Linear regressions are developed be-
tween monthly average temperature and
each of the temperature attributes for each
station, using the observational period of
record or any portion thereof. These regres-
sions are then used to determine changes
in the attributes arising from changes in
mean monthly temperature. Attribute
changes are calculated forthe mean monthly
temperature changes postulated by the
GCMs of (a) the Geophysical Fluid Dynam-
ics Laboratory (GFDL), (b) Goddard Insti-
tute of Space Sciences (GISS), (c) the United
Kingdom Meteorological Office (UKMO),
and (d) Oregon State University.
Each GCM gives a different temperature
projection so four different scenarios can be
produced. There is no guarantee that these
taken together provide the full range of
possible futures, but they are likely to indi-
cate the inherent uncertainty in the projec-
tions. Further, the temperature changes are
those associated with a doubling of atmo-
spheric CO2, which is projected to occur
sometime in the second quarter of the next
century. This time scale depends on many
nonclimatic factors, such as the level of
global economic growth, or political actions
to curb emission of greenhouse gases.
The effects of a wider range of possible
future conditions, or assessment of the im-
pacts in situations other than doubled CO.,
can be considered by using the option al-
lowing the user to select any temperature
change (cooling as well as warming). The
program also calculates the effect of a 1°
rise, testing the sensitivity of the attributes
to changes in mean monthly temperature.
Program Overview
This section contains information about
the operation of the program, and it includes
some general considerations of program
structure and organization to ensure that
scenario development proceeds efficiently.
The program is self-contained. However,
(a) the daily temperature data needed for
input are not included, and (b) the program
does not produce presentation-quality
graphics.
The large number of observing stations
and the immense volume of temperature
data potentially of interest for scenario de-
velopment preclude the incorporation of the
temperature data. Hence they must be ob-
tained off-line and placed in the appropriate
format for analysis. An increasingly com-
mon and convenient source of temperature
data for analysis by PCs are optical disks.
These are commercially available. Conse-
quently, within the program are routines for
using data downloaded from these optical
disks, although explanations for using other
data sources are also given.
Program outputs were designed to pro-
vide simple on-screen tables and graphics
of the required scenario information. In ad-
dition, all outputs can be saved as ASCII
files, so that more refined data analyses can
be undertaken and higher quality presenta-
tions can be created.
Hardware Requirements
For program operation the following are
required:
PC or Compatible
- 265 KB or more memory.
- A minimum of a 12-MHz 80286 ma-
chine with a math coprocessor is
desirable, or file creation can be a
long and slow process.
Hard Disk
- 500 KB free for program installation.
- Another 500 KB is recommended for
data files storage, although less is
actually required.
Monitor
- For display of on-screen graphics, a
VGA card and color monitor are re-
quired.
Printer (optional)
- Dot matrix or laser
The program has been tested with a
variety of hardware. However, there is no
guarantee that it will work correctly with any
specific combination of components.
Program Operation
The program is entirely menu driven.
Oncestarted, and the "Sign-On" screen has
been cleared, the Main Menu is presented.
This contains five options, the first three for
scenario development, the final two for
housekeeping. Scenario development pro-
ceeds sequentially through Main Menu op-
tions 1,2, and 3. If optical disk data are used
as the temperature input, development starts
with option 1; if other sources are used,
option 2 is the start.
Main Menu option 1 requests a choice of
the temperature scale and the value of the
high- and low-temperature thresholds to be
used. The maximum and minimum tem-
perature files created off-line from the opti-
cal disk data are then selected and used to
produce the monthly and annual statistics.
The second Main Menu option is used to
develop, from the monthly and annual sta-
tistics, the regression relationships between
the various attributes and the monthly aver-
-------
age temperature, and to examine these
regressions. Choosing this options leads to
a submenu. The first submenu option is for
creation of the regression file, which must
be the first action. The file thus created is
then available for viewing, printing, or fur-
ther analysis. This further analysis involves
the creation of new files, either for an indi-
vidual station throughout the year or for the
whole area for a particular month. The re-
sults can be viewed or printed in tabular or
graphic form. These files are designed to
help investigate the confidence which can
be placed in the regression approach.
Main Menu option 3 creates the actual
scenario. Again there is asubmenu, the first
option of which is the creation of the sce-
nario file. This uses the regression results
andthetemperature changes postulated by
four GCMs to determine four scenarios for
the selected attributes. In addition, results
are calculated for a 1 ° temperature change
(in the scale selected in option 1) and a
change which is selected by the user. This
scenario file is also available for further
analysis^ using new files created either for
an Individual station throughout the year or
for the whole area for a particular month.
Like the companion regression files, these
are used to assess the confidence which
can be placed in the results.
Option 4 allows the specification of the
directories where the program and scenario
development files are to be located, and it
provides details of the hardware configura-
tion. Finally, Main Menu option 5 contains
various routines allowing the joining, dele-
tion, or reformatting of the various scenario
development files. These two options are
housekeeping activities designed to allow
efffclentprogram operation while encourag-
ing critical viewing of the scenario results.
The program has been formulated in such
a way that, as far as possible, the user is
forced to consider the validity of the results
at all stages of the scenario development.
This includes both the scientific validity of
the whole approach in a particular situation
and the statistical validity of the individual
regressions. The program is designed to
emphasize this continuous testing of results
so that scientifically credible scenarios are
developed. The practical expression of this
formulation is that the program creates, and
encourages inspection of, a series of files
as the scenario development proceeds.
Scenario Development Example
The report uses the information contained
in example files to develop a scenario. This
is an example designed to show in more
detail the method of program operation,
indicate how a scenario might be devel-
oped, and suggest ways in which the confi-
dence to be placed in the results can be
assessed. The example chapter provides a
detailed guide for using the program and
includes many comments on the interpreta-
tion of results in order to create a valid
scenario. It is assumed that the text and the
example program will be used simulta-
neously.
Conclusions and
Recommendations
A simple linear regression approach can
beusedforthe development of scenarios of
future temperature attributes. Using this
approach, a scenario development soft-
ware program, TEMPSCEN, for use on a
personal computer, was created. This pro-
gram can be used by knowledgeable per-
sons concerned with the potential impact of
climatic change in a specific region. The
program is self-contained, the only data to
be proved by the user being the daily tem-
perature information for the places of inter-
est. Throughout the scenario development,
opportunities are provided for testing the
validity of the results and assessing the
degree of confidence which can be placed
in them.
The temperature attributes used are
monthly values of the standard deviation of
daily temperatures, the diurnal temperature
range, the mean maximum, the mean mini-
mum and the number of days exceeding
user-selected high- and low-temperature
thresholds. These are regressed against
monthly average temperature, using the
historical observational climate record. Sce-
narios can be developed for any station
having at least four years of daily maximum
and minimum temperature data. Scenarios
are developed forthe temperature changes
postulated by four general circulation mod-
els and for a temperature change selected
by the user. Hence a suite of scenarios,
indicating a variety of possible future condi-
tions which are equally likely, is produced.
The program is menu-driven. Through-
out the program frequent analysis of the
output as development proceeds is encour-
aged. Context sensitive help is always avail-
able. This includes specific information about
the steps in program operation and more
general suggestions pertaining to the inter-
pretation and analysis of scenarios.
The TEMPSCEN program allows simple,
straightforward development of scenarios.
It is recommended that the method be used
when general estimates of future tempera-
tures are needed.
There is no guarantee that future
conditions will be anywhere near
those projected using this method.
Scenarios are limited to ranges of
possible conditions. Results must
not be used uncritically. The Inter-
mediate program products should
be examined closely before any con-
clusions are drawn.
&U.S. GOVERNMENT PRINTING OFFICE: 1991 - 648-080/40104
-------
-------
Peter J. Robinson is with the University of North Carolina, Chapel Hill, NC 27599.
Peter L Flnkalateln is the EPA Project Officer (see below).
The complete report, entitled "Temperature Scenario Development Using Regression
Methods (Order No. PB91- 231506/AS; Cost: $19.00, subject to change) will be
available only from:
National Technical Information Service
5285 Port Royal Road
Springfield, VA 22161
Telephone: 703-487-4650
The EPA Project Officer can be contacted at:
Atmospheric Research and Exposure Assessment Laboratory
U.S. Environmental Protection Agency
Research Triangle Park, NC 27711
United States
Environmental Protection
Agency
Center for Environmental
Research Information
Cincinnati, OH 45268
BULK RATE
POSTAGE & FEES PAID
EPA
PERMIT No. G-35
Official Business
Penalty for Private Use $300
EPA/600/S3-91/049
------- |