EPA's Response to
External Peer Review and Public Comments of
'Preliminary Steps Towards Integrating Climate and Land Use: the Development of Land-use
Scenarios Consistent with Climate Change Emissions Storylines"1
(EPA/600/R-08/076A)
1 Final product title, "Land-Use Scenarios: National-Scale Housing-Density Scenarios Consistent with Climate
Change Storylines"
-------
EPA Responses to Peer Reviewers' Comments
(1) Does the report address its stated goals and if not, what are your recommendations for improving
the report?
Reviewer
Reviewer Comments
EPA Response
Daniel
Brown
Daniel
Brown
The report could be a little bit clearer in its body about its specific
objectives. I find it somewhat unusual that there is information about the
goals and content of the report in the executive summary that I don't see
in spelled out in the introduction. The summary should summarize the
content, meaning that the same content should be available only in more
detail in the body. As the body of the report reads now, it is a bit abrupt
in its movement from general statements about climate change and land-
use change to specifics about what was done in the project, with little in
the way of content that would motivate or frame the rest of the document.
So, one suggestion would be to include in the introduction to the
document a clearer statement of the objectives of the project and of this,
apparently interim, report. This statement could include an indication of
why this is a good stage at which to produce the report and what the next
stages will be. Of course, this later question is spelled out in general
terms in the section on "Options for Future Study," but one wonders
whether there is already some work underway along any of the lines
mentioned.
As for achieving the overall goals of the project, one question I have is
whether the computer code and data sets that are referenced in this
document are publicly available and now in an easier-to-use format.
Clearly quite a lot of work went into the completing the project and, as
the goal is to "enable us, our partners, and our clients to conduct
assessments of both climate and land use change effects," important steps
in achieving the goals are (a) making this process simple to implement so
that various alternative scenarios could be explored and (b) distributing
the tools that are produced by the project to a wide range of potential
users. A number of tools, data sets, and conversion processes are
outlined, but nowhere is a URL specified for accessing these tools. As a
report on methodology, distributing these is critical to achieving the
stated goals.
The introduction was revised
to reflect comments about
goals and stage of project.
A GIS-based tool was
developed based on this study
and is currently in review. This
tool is mentioned in the
Preface.
Steven The report meets its chief goal of providing a model for characterizing
Manson and assessing the changes in land use in the United States into the future,
as measured by housing density and impervious surface cover. This
research is especially valuable in how it downscales the widely accepted
Intergovernmental Panel on Climate Change (IPCC) Special Report on
Emissions Scenarios (SRES). The social, economic, and demographic
storylines that drive SRES are tied to specific processes at fine scales,
such as migration to the county and imperviousness/housing density
down to the hectare. There are areas where the report could provide
additional details, and there are others where the model could be
No response necessary
-------
Reviewer
Reviewer Comments
EPA Response
extended in the course of future research.
Dawn
Parker
I'm basing the stated goals on these outlined in the charge:
The Global Change Research Program (GCRP), within EPA's Office of
Research and Development, focuses on assessing how potential changes
in climate and land use may affect water quality, air quality, aquatic
ecosystems, and human health in the United States. The GCRP has
completed an internal EPA report describing the methodology used to
develop future land-use scenarios for the United States by decade to the
year 2100.
and on the "Preface" text from page vii:
... The report describes the methodology used to develop and modify the
models that constitute the EPA Integrated Climate and Land Use
Scenarios (ICLUS). The scenarios and maps resulting from this effort are
intended to be used as benchmarks of possible land use futures that are
consistent with socioeconomic storylines used in the climate change
science community. The two-way feedbacks that exist between climate
and land use are not yet fully understood and have consequences for air
quality, human health, water quality, and ecosystems. In this report we
describe the first steps towards characterizing and assessing the effects of
these feedbacks and interactions by developing housing density and
impervious surface cover scenarios. These outputs facilitate future
integrated assessments of climate and land-use changes that make
consistent assumptions about socioeconomic and emissions futures.
EPA's intention is to use the results of this first phase of modeling to
inform and facilitate investigation of a broader set of impacts scenarios
and potential vulnerabilities in areas such as water quality, air quality,
human health, and ecosystems. More specifically, this research will
enable more sophisticated model runs that will evaluate the effects of
projected climate changes on demographic and land use patterns and the
results of these changes on endpoints of concern.
I would like to make a careful distinction between 1) whether the report
addresses its stated communication goals and 2) whether the model
developed by the authors meets the goals set out by the EPA and the
authors.
With a few exceptions described under question 2, the report meets it
stated goal to "describe the methodology used to develop and modify the
models that constitute the EPA Integrated Climate and Land Use
Scenarios (ICLUS)."
No response necessary
-------
Reviewer
Reviewer Comments
EPA Response
Dawn
Parker
The model presented is, as the authors state, a first step. The team is to
be commended for developing an approach, based on existing and
available data, that attempts to project migration and residential land-use
change for the entire US. However, for reasons that are described in
greater detail below, it is my assessment that the model that the have
developed is not yet ready to be used to project land-use changes that can
then serve as inputs into other environmental assessment models for the
purpose of policy analysis. My main reasons for this conclusion are:
• The report does not review, make reference to, or make
comparisons to other regional and national land-use change
models that were developed elsewhere with similar goals.
Dawn
Parker
• No estimates of the error and uncertainty of the integrated
projections (based on the coupling of three models and on many
assumptions) are provided.
Theobald (2001, 2003, 2005)
compared SERGoM to other
modeling efforts, including a
couple suggested by the
reviewer. The general
difference is that most models
are based on assumptions that
require very spatially detailed
data such as those at the parcel
level (e.g., specific land use
type and zoning). The more
general models that have been
developed for countrywide
scale by Europeans are more
general. None of the models
developed in the US suggested
by the reviewer have been
developed for the entire US ~
they are too data intensive...
that is why SERGoM is
unique. Also, some of the work
and citations post-dates the
beginning of this project. See
report for specific changes and
citations.
Added text to introduction
about how this study is
intended to explore scenarios,
and that uncertainty with any
of the outcomes is very high.
Dawn
Parker
No model validation has been conducted to compare the
projections of the ICLUS model to real-world land-use change
data. At a minimum, in-sample model validation should have
been conducted. It is important for policy makers to have
information on how well a model designed specifically to
produce realistic temporal and spatial change projections
performs against real-world data, so that users can assess the
level of confidence that they should have in model predictions.
Validation also provides important information on next steps for
model modification and improvement. (Verburg et al. 2006).
We agree that some type of
validation is important. We
provided additional discussion
of validation in the text. We
also added a recommendation
that forecasted housing density
patterns should be compared to
more local and specific
models.
-------
Reviewer
Reviewer Comments
EPA Response
Dawn I am concerned that the coupled models are over fitted and contain too
Parker few explicit representations of land-use change processes and the drivers
of land-use change. A model that contains few structural elements and is
very closely calibrated to a particular place and time is unlikely to
perform well outside of the range of calibration data (Verburg et al.
2006).
All models face this criticism -
- and it comes back to
understanding the assumptions
of the model. Again, these are
forecasts that reflect specific
assumptions that are described
in the SRES scenarios. And,
other reviewers recognize that
this may be a necessary
tradeoff. We also added
Appendix F, which lists the
main assumptions of the
models.
David
Skole
Yes, but not as well as one would have liked. Indeed, the goals of the
project are never made explicit. There is some indication from the
Introduction on page 1 lines 14-17 and lines 24-28. The impression is left
to the reader that the modeling project will lay a foundation for integrated
assessment of the complex relationships between land use change and
climate change: "The motivation for the EPA-ICLUS project was derived
from the recognition of the complex relationships between land use
change and climate change impacts and the absence of an internally
consistent set of land use scenarios that could be used to assess climate
change effects." This insight into complexity and integrated assessment is
never achieved. The report suggests there will be a way to assess
feedbacks from climate on land use change and this insight is never
achieved in the report. The report should state very clearly that its main
intention is to model one form of land use change (housing density and
impervious surface) to estimate its effect on greenhouse gas emissions.
Revised Executive Summary
and Introduction to make goals
clear.
David It is not possible from this report and the methods it used to even make a
Skole statement on land use affects on climate since the study does not include
an explicit method for linking resulting land use changes with surface
conditions, sensible and latent heat flux or other similar biophysical
parameters. It is probably not readily possible to link the results of this
sources. For instance the land use effect on carbon stocks (e.g. forest
study to greenhouse gas emissions since there are many non-modeled
loss) and on gas exchange (e.g. nitrous oxide in agriculture) are not
considered. Hence the reports needs to make it very clear what it can and
cannot achieve, starting with an clear statement of goals - and not mis-
represent this approach as "complex" or "integrated".
Added this clarification to
Introduction.
David A very clear example of the lack of integration is shown in the migration
Skole gravity model development, in which historical average climate
conditions are used. The report suggests a literature that shows strong
relationships between migration locations and climate, yet, rather
surprisingly, the model uses past conditions - no attempt is to incorporate
Added text to the introduction
clarifying that this report
describes the first phase of the
project, and that such
integration is a likely future
-------
Reviewer
Reviewer Comments
EPA Response
new climate parameters.
step.
David This reviewer's suggestion is to re-craft the entire Introduction with a
Skole very clear overview of Goals and Objectives. Be very clear that the rather
simplified, if not simple, method to spatialize demographic trends is
merely a first attempt to recognize the geographical variations in
potential land use changes derived from population density location.
Clarified goals and objectives
in Introduction.
David There is confusion between housing density and the cited work by Lui et
Skole al (2003). Lui's work, which I have many problems with anyway, focuses
on households not housing density. Their claim is that the number of
households is a better predictor of land use change and other
environmental impacts than population alone. The household metric is
not necessarily coupled to population density, and they have used this
concept extensively to model a far ranging array of impacts, including
such things as divorce which tends to make two households out of one
and has no bearing on population dynamics. The EPA analysis includes a
variation in household size but only as a function of fertility, which is
reasonable but not the same concept as Lui et al. (2003). I suggest the
authors steer clear of associating their approach to that of Lui et al
(2003) and focus, as they have, on trends in housing density derived from
population characteristics.
The Liu, et al. reference was
removed.
David There are a number of land use change modeling efforts underway and
Skole many different approaches and methods. Usually the method used is a
function of the goals of the analysis. It is not at all clear that the methods
selected are related to any of the goals, what ever they are. Since the
methods selected for this study have some obvious limitations, it is
important that the precise logic for selection of the methods is clearly
traced to the goals. The authors should, again, write the goals clearly up
front - and perhaps early in the text also include expected outcomes.
The Introduction was revised
to reflect goals. Other
reviewers commented on
methods as appropriate.
David Lastly the same can be said for the rationale to select the SRES so-called
Skole storylines as the basis for scenario analysis. Especially because they had
to be changed so much for the downscaling, it is not at all clear that this
was the best way to select scenarios. Again, an improved description of
goals and objectives would be warranted.
The Introduction was revised
to state goals more clearly and
describe the rationale for
selecting and modifying the
SRES storylines.
David The bottom line for this reviewer is that the approach taken may not be
Skole suitable for all climate impacts or emissions modeling and as such would
seem flawed and inaccurate in the context of many requirements that I
can think of. The approach appears to take a population growth model,
The goals of this study were
clarified in the Introduction.
Regarding scale, the land use
change model operates at a 1
-------
Reviewer
Reviewer Comments
EPA Response
modified by migration at a county scale and then spreads these people
over the landscape using some simple spatial allocation model mostly
derived from allocation weights based on urban travel distance. It is hard
to rationalize that this is a good approach - and that the 1 ha
spatialization appears to be much finer resolution than can actually be
modeled. To be honest the first order impression I had is that it's a
simple, unrealistic depiction of land use change that lacks any theory or
processes. Yet, such an approach may be perfectly reasonable for a
simple county-based assessment of settlement patterns and density over
time in order to estimate, for example, transport emissions or household
energy distribution and consumption and associated carbon dioxide
emissions. It may be practically ineffective for estimating nitrous oxide
emissions from agriculture (without knowledge of fertilizer application
rates) or the effect of forest conversion for bio-fuels or intensive forest
management on carbon dioxide emissions, or the effect of climate on Net
Primary Productivity. It is hard to form an opinion on approach without
knowing exactly what the goals, objectives, and expected outcomes are
thought to be.
ha resoultion ~ but we are
clear that any analysis of the
model should be aggregated up
to at least 1 km2 (which is why
we did this for the impervious
surface analysis). The 1 ha
resolution allows for major
land cover and transportation
structure to be better modeled.
-------
(2) Is the methodology explained in sufficient detail? What additional details or information should be
added to the report?
Reviewer
Reviewer Comments
EPA Response
Daniel By and large, I think the methods are well described. I admit to some
Brown confusion in the description of the SERGoM processes, especially in
section 4.3 on adapting SRES scenarios into SERGoM. It appears to me
that the modifications to the model outlined in that section are not
depicted in Figure 4-1. If that interpretation is correct, I think it's an
important oversight and should be corrected. I think the modifications on
changes in household size and urban form are important innovations in
this project and that their implementation within the context of the
SERGoM approach should be made absolutely clear. As it is now,
section 4.3 talks about a new allocation weight to reflect different
scenarios based on travel times, but it it's not clear how it combines with
allocation weights discussed earlier.
Good point, added clarifying
sentences in section 4.3. Figure
4-1 conveys the overall model
structure ~ there are a number
of details left out of the
flowchart, but that's the
balance between a general,
overall depiction of a model
and the technical details. More
boxes and arrows could be
added to the figure, but it
would obfuscate the overall
operations of the model.
Steven Overall the level of detail is adequate, although there are specific
Manson instances noted below where more detail would help in interpreting the
model. These relate to the gravity model, apportioning PUMA data, and
migration modeling (see comments under Question 3, 5). There is also
room for more description of impervious surface modeling and
importance of compactness in urban areas (see comments for Question 6)
The IS and compactness
discussions have been edited to
add more detail. The gravity
model and PUMA
apportionment discussions
were also improved.
Dawn The methodology is explained in detail. The one point that is not clear is
Parker the calibration and role of the housing density and impervious surface
cover model described in Appendix C. Was this model developed using
modeled housing densities, rather than real-world densities? Did a
statistical model of the relationship between real-world housing density
and impervious surface feed into the model at some level? Were any of
the land cover layers used for model calibration classified based on
impervious surface, or were urbanized land uses derived in some other
way? Is the evaluation of the model designed to validate how well the
residential housing model projects changes in housing density, or how
well the model projects changes in impervious surface?
Appendix C was revised to
include more detail and
clarification for impervious
surface.
Dawn It would also be helpful to have a section that summarizes the many
Parker assumptions made in each of the model components.
We added tables summarizing
major inputs and assumptions
to Appendix F, with references
to these tables in the text.
David The methods were sufficiently explained for this reviewer. I comment on
Skole the actual sufficiency of the methods selected in another response. There
are two important missing elements that require further elaboration. The
first should be a detailed table of all the input datasets and a clear
description of them.
We added tables summarizing
major inputs and assumptions
to Appendix F, with references
to these tables in the text.
-------
Reviewer
Reviewer Comments
EPA Response
David The second should be an explicit discussion of accuracy and validation.
Skole The absence of any kind of validation exercise makes this reviewer
skeptical of the methodology. There are some very elegant looking
approaches with simple implementations (e.g. the gravity model) that
have: 1) no clear elaboration in literature, and 2) no validation from
historical data, or other suitable validation dataset. The only approximate
validation is for five state population estimates, which by review of
Figures 3-4 to 3-8 appear to be very poor fits, with the EPA model
consistently under estimating the state models. The text dismisses this
rather simply. The spatialization model seems to be run on blind faith.
This report must make every effort to provide a mapped validation for a
location or region.
Regarding the spatialization
model, please see responses to
Dawn Parker's comments in
Question 1. Regarding the
demographic model, we
provide several references that
discuss the widespread use of
gravity models for spatial
interaction studies. Validating
the future projections could
only be done against other sets
of projections, as sufficient
data were not available to run
the model for an historical
period. Due to the uncertainty
associated with any projection
effort, we chose to explore
multiple possible scenarios.
David The impervious surface calculation is one parameterization where there is
Skole an attempt to provide a validation, but the text on page 39 lines 10-20 is
hard to understand, even though there is more description in Appendix C.
(The reference to Theobald et al in press is useless). For instance the text
beginning on page 39 line 12 is confusing: "A brief comparison of our
modeled IS to existing fine-grained (from high-resolution photography)
validation datasets resulted in an R2=0.69 (Elvidge et al. 2004) and
R2=0.69 and R2=0.96 for Frederick County, Maryland and Atlanta,
Georgia (Exum et al. 2005)." For one, Elvidge et al use satellite imagery,
and it is not clear how the spatial resolution differences between Elvidge
et al and this report matter. I imagine, although cannot tell, that the
comparison is done with non spatial data.
Thanks, these comments were
helpful in editing this section.
We revised the text to clarify
the language, and updated the
citation to Theobald et al.
2009.
David Lastly, related to Question 1, what role does impervious surface play in
Skole emissions or climate analysis. I can imagine it could - but its not clear
from this report how the EPA intends to make the connection.
Although there are a variety of
ways that impervious surface
(IS) plays a role in emissions
or climate, in this document we
pursued only the use of IS as a
general indicator - not
specifically tied to possible
changes in carbon cycling,
emissions, or heat island
effects.
We've addressed this in
general, e.g. in Section 5.4:
Groisman et al (2005) suggest
that one potential impact of
climate change is an increase
in the intensity of individual
storm events. Since these
10
-------
Reviewer
Reviewer Comments
EPA Response
events are responsible for the
majority of impacts to water
quality from stormwater
runoff, examining the possible
extent of impervious surfaces
become even more important
given the anticipated impacts
of climate change.
David There are some other poorly described section that could be better
Skole elaborated. For instance how is the modeled output from this study
matched up against the MRLC dataset to derived changes in land cover
types, and what happens when there are in consistencies between them.
We added more detail here
about the resolution and
methods, but it is a simple
overlay operation that involves
two different data layers.
-------
(3) Given the goals of this study, comment on the technical merit of the modeling approaches used, as
compared with other available approaches. Please comment on strengths and weaknesses of the
modeling approaches used.
Reviewer
Reviewer Comments
EPA Response
Daniel This is a very important question. While the interactions of land use
Brown change and climate change on ecosystems and human societies are
important, and there remain a number of open questions to be explored
regarding these interactions, the choice of modeling approach here
clearly directs this line of research towards answering a particular subset
of these questions. While there are clearly a number of simplifying
assumptions contained within this analysis, I view this approach less as a
modeling exercise and more as a data assimilation and projection
exercise. What is being modeled is demographic change, though the
project takes those as projections given from the census bureau. Beyond
that, conversions of demographic projections to land use and land cover
impacts are based largely on empirical regularities and stated
assumptions. I describe the process in this way to distinguish it from
process-oriented models.
A clear advantage of the approach taken here, as discussed in the section
on SERGoM, is the ability to generate bounded and comparable
estimates on a national scale. The assumptions that go into the different
scenarios are reasonably clear. The authors have made a case for how
well these assumptions represent the SRES scenarios and, while I
suppose reasonable people might disagree on these arguments, there is a
reasonably high level of clarity on what the assumptions are. If there
were a computer interface available for manipulating the assumptions
and evaluating the outcomes in real time, it might be more useful for
exploratory purposes. No where are the computer resources required
produce a scenario identified, but these may be limiting on the utility of
such an approach. This is a reasonable approach when the goal is to
assess land-use and climate-change interactions in the sense of joint
effects for impact assessment. This seems to be to approach being taken
here. The approach could conceivable be used to evaluate the relative
independent impacts of plausible future land-use changes and plausible
future climate changes on a system of interest, as well as their joint
effects.
Thank you for these valuable
comments. The availability of
the GIS tool is now discussed
in the Preface and Section 5.4.
Daniel The most obvious limitation of the approach is its reliance on past
Brown experience and data to parameterize future dynamics and outcomes.
This assumption of stationarity is very limiting when it comes to land use
processes. The authors acknowledge the difficulties of projecting the
economic aspects of land use (e.g., due to changes in credit availability,
fuel prices, job markets, trade, etc) and use that as an argument for
focusing on the demographic drivers. This is a reasonable argument, but
it doesn't make the possibility of huge disruptions in past dynamics into
the future as a result of changes in these broader economic conditions go
away. The fact remains that the approach involves projection of past
dynamics into the future, assuming that the future will look much like the
past. The manipulation of parameters to match the SRES scenarios is a
These are useful comments,
and we added some of these
points to our caveats.
-------
Reviewer
Reviewer Comments
EPA Response
great start towards imagining different futures. However, even with this
important activity going on in this project, there are a number of
processes or relationships (for example, intercounty migration patterns,
association between housing density and imperviousness) that are
assumed to be unchanged into the future. It's difficult to avoid such
assumptions with this data-driven approach - unfortunately, we just can't
get data about the future. Nonetheless, the authors have made a great
start towards tweaking a data driven model to represent alternative
scenarios.
Daniel There are other important interactions, including those between land use
Brown and climate, that this approach is not particularly well suited to address.
Those are partially acknowledged in the "Options for Future Work"
section and involve impacts of climate on land-use change and impacts of
land-use on climate change.
We have made edits to both the
Introduction and Options for
Future Work about the
limitations of this approach.
Daniel The examples mentioned describe how sea level rise or changes in
Brown amenity values associated with climate could cause changes in migration
patterns and other land-use changes that are not included in the
demographic scenarios driving the land-use scenarios. I'm not sure I see
a straightforward path to evaluating this scenario with the model as
currently structured. Because the model is so closely parameterized with
prior observations (e.g., to set the county-county migration flows),
incorporating a process that hasn't yet been encountered on a large scale,
like coastal inundation, would be difficult.
Thank you for the comment.
The authors agree that
incorporation of these
processes for the purpose of
predicting demographic
patterns in out-years would be
very difficult. Rather, we
might use the information
developed by ICLUS to help
gauge the extent of the
problem from the standpoint of
how that land is being used. In
the case of coastal inundation,
if we overlay sea level rise
maps in 2050 over SERGoM
outputs, how many people will
have to live somewhere else?
This model will not be able to
(nor was it intended to) predict
where those people will go
instead and when.
Daniel An alternative direction not mentioned is the possible effects of land-use
Brown change on climate. For example, urban heat islands and other large scale
land alterations on latent and sensible heat budgets can create significant
forcings on climate. In order to evaluate these effects, the land-surface
model would presumably need to be linked dynamically to the climate
model, so that updated land-surfaces are fed to the climate model at each
step. If there is no effect of climate on land-use, then the land-surface
series already created could serve this purpose (with more variables
generated). If there is climate effects on land use (e.g., through flooding,
drought, changes in crop productivity or other effects) then it would be
Examining the effects of land-
use change on climate was not
an explicit goal of this project,
but this is an interesting
comment to consider in future
studies.
14
-------
Reviewer
Reviewer Comments
EPA Response
more complicated.
Daniel
Brown
Also, are there conceivable futures in which large swings in land use that
could result in significantly more or less sequestration of carbon in the
landscapes? This can't really be evaluated, but might be important,
especially if there are policies aimed at climate mitigation are
implemented specifically for this purpose.
Examining the effects of land-
use change on carbon
sequestration was not an
explicit goal of this project, but
this is an interesting comment
to consider in future studies.
Daniel Other interactions within the land-use system are also important. The
Brown positive feedback that causes larger places to grow more rapidly (which
Paul Krugman recently won the Nobel Prize in economics for formally
describing) is represented in the demographic model (perhaps too well).
However, the model doesn't account for changes in industrial and
commercial activity associated with these changes and how they might
result in different kinds of new attractions in a place. The urban form
manipulations in SERGoM could conceivably be used to approximate the
observed negative feedback within exurban areas, where nearby
development decreases the likelihood of development, through use of
varying densities, but these processes are not represented explicitly as far
as I can tell.
This is an interesting comment
and an area for potential model
improvement in the future.
Daniel Another aspect of the model that is limiting is its deterministic nature,
Brown i.e., it produces only one outcome based on the number of estimated
migrants between counties and the most suitable locations within
counties. This assumes both a high level of certainty that these factors
are well modeled and that the people moving and locating have good
information and behave uniformly rationally. Variation in outcomes is
not admitted to the model, except through the scenarios. In fact, there is
quite a bit of both variability and uncertainty within the context of any
given scenario that is not represented at all. The outputs of the
scenarios, therefore, give the users no information about likelihood or
probability or variance of outcomes. Adding stochastic variation to the
models would go some way towards providing some of this information.
We created the scenarios to
look at different possible
outcomes, and acknowledge
that the outputs represent only
a small range of the infinite
potential outcomes. We will
explore possibilities of adding
stochastic variation in future
improvements. The
Introductions and Options for
Future Study sections were
revised to express this.
Daniel
Brown
Along these lines, there a few mentions throughout to the "likely"
outcomes under land-use change (see pgs. x, 3, 39). I think this word
should be assiduously avoided in describing the outcomes from the
model and the project. All the authors can say is what is plausible if we
accept the assumptions.
Replaced most occurrences of
"likely" with some form of
"plausible," "possible," or
"might," depending on the
context. Some are left
unchanged where appropriate.
Steven Overall. The model uses appropriate methods, especially in so far as they
Manson are standard and well-understood approaches being used in new ways to
address outstanding research questions (e.g., downscaling, spatial
allocation at fine scales across broad extents). Other commonly used
methods that could be used in this situation tend to center on 'black box'
approaches such as very complicated systems dynamics models or
computational intelligence methods such as artificial neural nets. These
approaches could conceivably produce better model fit, but at the
expense of transparency and maintaining the assumption of statistical
No response necessary
-------
Reviewer
Reviewer Comments
EPA Response
stationanty over time.
Steven Internal migration. The gravity model is an effective approach to
Manson migration modeling. Other approaches that could be used, or be used in
conjunction with gravity modeling, include spatial statistical estimation
or a more process-based model of migration based on survey responses
(although some of the literature cited relies on these data). This said,
these other approaches would likely run afoul of the limited nature of
data available at necessary scales.
No response necessary
Steven International migration. The population model could have a better
Manson international migration component that moves beyond the uniform
distribution of migrants among counties. This would likely involve
county-specific (or perhaps just state-specific) estimates that are driven
by past migration patterns or features of counties that appeal specifically
to immigrants. In aggregate, the current schema is adequate to the task,
but the site-specificity of the model would be better served given the
importance of gateway cities and social networks in guiding where
international migrants find themselves.
Added discussion of
limitations to Section 3.4
Dawn Again, the authors of the report should be commended for undertaking a
Parker first effort at this very challenging modeling task, and also for providing
sufficient detail on their modeling methodology so that I am able to make
detailed comments and criticisms.
No response necessary
Dawn In a report such as this one, I would expect to see a brief review of other
Parker related models, along with a specific discussion of how their model
compares to other approaches. Several quite sophisticated national and
regional level models have been developed in European study areas, and
some of them have even been coupled with the IPCC scenarios (Engelen,
White, and de Nijs 2003; Verburg, Rounsevell, and Velkamp 2006). It
would also be helpful to see comparisons to projections from regional
models done in the US (Jantz, Goetz, and Shelley 2003; Landis and
Zhang 1998; Waddell 2002). Many different approaches are available to
model land-use and land-cover change, and the choice of approach is
often constrained by available data and research resources. It is also an
open question which approaches will be most effective at regional and
national scales and over long time frames. Thus, rather than focus on a
detailed comparison between the ICLUS approach and previous
approaches, I will comment on specific concerns that I have with the
ICLUS approach. Some are due to data constraints. The data constraints
represent an important policy issue that I will discuss further in question
Similar to Ql, we have added a
short discussion of how
SERGoM compares with other
modeling efforts, including
efforts that have integrated
SRES scenarios with land use
change modeling.
16
-------
Reviewer
Reviewer Comments
EPA Response
Dawn I support the use of the IPCC scenarios. These are well understood by
Parker the international community and have been used in other, similar
modeling efforts. Certainly they should be seen as a starting point, but
they are a reasonable one. They have also been used for very coarse-
scale economic integrated assessment models. Have any other scholars
attempted to downscale these scenarios for the US?
Several downscaled models
that look purely at climate
(Univ. of WA and NASA, for
example) exist and Columbia
U has some downscaling to
address economy/GDP for the
country as a whole. Urban land
use study using downscaled
SRES is available from the
Journal of Environmental
Management. Further
investigation of these models is
a possibility for future study.
Dawn I am not a demographer, and so cannot fully assess the gravity model
Parker used here. However, I am concerned that the county-to-county approach
used here fails to capture the multi-scale dynamics of regional vs. local
migration and the land-use change that results. The drivers of inter-urban
and intra-urban migration differ (Clark and Van Lierop 1987). Drivers
such as regional amenity values, employment opportunities, and life
cycle stage can trigger inter-urban migration. Once a household has
relocated, preferences, income, and transportation networks will
influence where the household locates within an urban area. Location
decisions of those migrating within an urban area may also be very
different than those in-migrating from another region. Would it be
possible using the available data to estimate a two-stage migration
model, one for example from MSA to MSA, and the second within
MSA?
The migration data used to
develop this model included a
large proportion of intra-MSA
migrations, and such migration
were built into the regression.
It may be possible to develop a
two-stage model, though it was
not in the scope of this first
study.
Dawn p. 9 section 3.2: Perhaps it would make more sense to distinguish
Parker between "immigrant and non-immigrant populations," rather than by
ethnicity. What factors drive patterns of ethnic migration?
The race/ethnicity categories
we used were driven by the
population and rates of change
data. The initial population
data was not detailed enough to
distinguish in this way, and
they fertility and mortality
rates do not distinguish
between foreign- and native-
born.
Dawn p. 11 23-24: How much confidence can we have that the current trends
Parker and distributions of migration will continue? It is a concern that the
census migration projections seem unrealistic, since they are a model
input.
We added text about
uncertainty in this area, since
changes driven by policy and
economics can easily disrupt
patterns and projections.
-------
Reviewer
Reviewer Comments
EPA Response
Multiple scenarios were
considered given the high
uncertainty.
Dawn p. 12 17-18: Are these "R2" actually pair-wise correlation coefficients
Parker (r2)? How do your results compare to other gravity models?
By definition, the Pearson
coefficient correlation is
calculated in a pair-wise
fashion.
Dawn I agree that the stepwise regression techniques are not appropriate for this
Parker application. It is important to keep known theoretical and empirical
drivers of land-use/cover change (LUCC) in the model. If the goal of the
model is an aggregate prediction, colinearity between variables, within
reason, will have minor effects of the predictive power of the entire
model, especially with a large sample. I expect that the model
coefficients would need to be updated over time as more data became
available, yet another justification for not omitting known drivers of
LUCC. For example, what about employment?
We acknowledge that
employment is an important
driver of land use change, but
chose to omit it due to the
difficulty of projecting county
employment throughout the
U.S. into the future. We chose
to focus on more predictable
demographic processes. While
this does ignore a driver of
LUCC, the scenario-based
approach is intended to explore
a range of possible futures.
Dawn Modeling growth as a function of previous growth means essentially the
Parker model is a reduced-form temporally autoregressive model. Yet, it is not
clear that the authors test or correct for temporal autocorrelation. This
also means that counties that grew in the past will be projected to
continue doing so, and counties that were shrinking will continue to do
so. Such highly inductive, pattern-driven models, in my opinion, are
unlikely to be adequate to project land-use change over long time scales.
This approach also severely limits prospects for sensitivity analysis with
respect to, for example, changes in employment or costs of living over
time. There is also the question of future resource constraints.
Temperature and sunlight explain a lot of recent migration because water
has been available and energy prices have been low. Both factors are
changing and are likely to continue to change in the future. These
changes could reverse current trends towards Western and Southern
migration.
We acknowledge that.
All models face this criticism -
- and it comes back to
understanding the assumptions
of the model. Again, these are
projections that reflect specific
assumptions that are described
in the SRES scenarios. And,
other reviewers recognize that
this may be a necessary
tradeoff.
Dawn p. 14, 31-34: Do absolute cost distances between locations really explain
Parker migration? Or rather, is there a threshold at which a move from New
York to Denver is really not so different than a move from New York to
San Francisco? And, wouldn't the distance from New York to
Washington have a different influence on decision making than the
distance from New York to New Jersey? Again, maybe some of these
Our analysis found that
population exerts a stronger
pull than distance, so while we
did find an inverse relationship
between migration and
distance, the gravitational pull
18
-------
Reviewer
Reviewer Comments
EPA Response
problems could be solved through a two-stage migration model. The
travel cost model is very detailed (likely a reflection of the strengths of
the team), but it may be too detailed given the generality of the other
model components.
of large population centers
outweighs relatively small
distance in distance when
considering multiple potential
long distance moves. A two-
stage migration model may
improve intraregional
migration estimates; future
work may take this option
under consideration.
Dawn p. 19-24: It would be very helpful to see the model's projections
Parker evaluated against some real-world data. Evaluation against other
projections is not sufficient, especially given that the methods used to
create those other projections were not carefully examined. These other
projections were also made for fairly high growth urban areas. It is
difficult to know how to evaluate the models' projections. If current
trends continue, they might be accurate, but over a 100 year time frame,
trends established over a 20 year time frame are not likely to continue. It
is a concern that "the ICLUS model is not able to predict population
growth due to migration in small rural counties with high natural
amenities" (p. 21, 15-16), given that ex-urban development is a major
concern.
It was obviously not possible
to check the demographic
model's behavior against real-
world data, given that only
other set projections are
available for comparison. Tests
might be possible if we began
the model in the past and ran it
through the present for
comparison, though sufficient
starting population data were
not available. Therefore, we
decided that a scenario-based
approach intended to explore a
range of possible futures would
provide value despite high
uncertainty about the
projections. Some text
regarding validation of
SERGoM was also added.
Dawn SERGoM model: A strength of this model is that it forecasts housing
Parker density, not simply residential location. The extensive non-developable
lands layers that the model incorporates are also a strength. Model
performance has also been formally validated to some degree (p. 27, 24-
27). However, again, the model is highly inductive and potentially over-
fitted to the data. Even a statistical model that contains a larger range of
drivers of location (for example, (Irwin and Bockstael 2002; Verburg et
al. 2002)) might be more robust for out-out-sample model prediction.
Clearly such a model would have to be run on a fairly coarse scale, given
data limitations. The model appears to take road networks and
groundwater availability as given; clearly these will change overtime.
This model also very much assumes that historical growth patterns will
continue, but not does model the drivers of growth (p. 27,1 8-12; 28-29).
Like nearly all other land use
models, there is an important
distinction between what the
model allows, and how it is
actually parameterized and run.
SERGoM does allow
parameters such as the road
network to change over time ~
yet there is simply no data
available (nationally) to do
this. However, the travel time
from urban areas does change
dynamically as a function of
the emergence of new urban
areas, something that
SERGoM shares with other
-------
Reviewer
Reviewer Comments
EPA Response
Cellular Automata inspired
models (such as Engelen et al.
2007).
Dawn
Parker
p. 26, lines 11-20: This method and explanation are not at all clear.
This section was improved.
Dawn Finally, as the authors point out, some important feedbacks are not
Parker implemented in the model, such as traffic congestion and feedback from
climate change. These are important needed extensions, however the
underlying methodology may evolve.
No response necessary
Dawn It is difficult to evaluate the model projections. They very much
Parker represent current trends. However, land-use planning and zoning is quite
different in the NE than in the south and desert SW, and these difference
do not appear in model output. Just one example where better data inputs
(zoning constraints, for example) might improve model performance.
Added text under Options for
Future Study.
David As mentioned in Question 1 the goals are not as clearly laid out as they
Skole should be, so it is not entirely possible to answer this question. The
strongest merit of this approach is to provide insight on the future
demographic distributions given current structures of the population and
settlements with current trends. It is not possible to use these models to
make accurate forecasts (predictions) because the drivers of land use
change are more complex than they are represented here. As mentioned
in the response for Question 1 this approach can be useful for some
goals: for instance to lay a foundation for estimating transportation
mobile source emissions, or household energy demand and location and
its associated emissions.
Revised introduction to clarify
goals.
David But the modeling approach is rather simple and lacks processes. There is
Skole no opportunity to look at the complex relationship between land use and
climate, with climate feedbacks on land use - in spite of some strong
overstatements about integrated assessments in the text.
Revised introduction.
David There is a growing literature on types of land use modeling and it would
Skole have been useful for the report. It may be necessary to state what options
for methods the team had and why other methods are not in fact used. For
instance urban growth dynamics - ergo sprawl - have been modeled in
several ways, some of which are more dynamic than this approach. There
is a well developed literature from economic geography on location
theory and some interesting spatial models based on Ricardo-Von
Thunen rent theory. There are also a suite of regression models built
around economic growth models such as REMI. These economic growth
models incorporate income parameters and other economic factors in
addition to demography. Historical studies of urban-suburban growth
(sprawl) show it is strongly tied to economic conditions - a rapidly
growing economy yields rapid urban development in the outlying areas.
These economic projections are thus necessary for making the
projections of land use change. There are also a number of spatial
association models, which use co-location of built up land with other
Thanks ~ please refer to
response to Ql. We added
additional citations and
reviews to compare to some of
these models ~ but also cite
the Theobald 2001; 2003; and
2005 papers which have cited
much of the work that is cited
in this comment
20
-------
Reviewer
Reviewer Comments
EPA Response
factors to operate the spatial allocation rules. This report's spatial
allocation is largely driven by a simple weighting function derived from
roads-distance.
David The strength of the modeling lies in its early attempt to perform a simple
Skole spatial map for the entire US. While I think the modeling is simple and
probably does not capture most of the necessary dynamics of economics
and land use, focusing too strongly on housing density alone, I think
there this is an important study. It has great value as a starting point for
further modifications and elaborations.
No response necessary
David One of the most difficult aspects of this study for this reviewer to
Skole understand is the argument for using the SRES story lines. The
suggestion made in the text is that the SRES was chosen because it is
widely accepted. I found this rationale lacking merit in many ways. First,
only the basic so-called story lines were used rather than more elaborate
data parameters established from the story lines in the full SRES.
Moreover, this report only relies on the demographic storylines when the
full SRES had other domains. Second, by the time the down scaling
exercises were done to get story lines for the US case, they no longer
well matched those of the global or regional IPCC SRES. This then begs
the question why to use them in the first place.
We have added some text to
the introduction about why the
SRES storylines were chosen.
David The weakness of the modeling method is that it cannot capture some of
Skole the more important attributes of land use change and land competition
that will likely confront the US landscape in the future. As well, as noted
earlier, the models cannot readily account for bio-physical processes
associated with land use change - water, biogeochemistry, and energy
balance.
Discussion of model
limitations added to
introduction.
-------
(4) The endpoints of housing density and impervious surface cover were chosen to provide data for
further analyses on environmental impacts. What other endpoints may be relevant to calculate to
support the goals of this report?
Reviewer
Reviewer Comments
EPA Response
Daniel This is really kept wide open in the report's stated objectives. The
Brown authors want to enable assessments, but they don't specify what kinds.
So, the list of possible endpoints is quite long. One could start with other
types of land covers. Probably the most important would be tree cover,
as it has implications for carbon storage on the landscape. Agriculture
might also be important; as noted in the results, a significant amount of
the new housing development in these scenarios would come from
agriculture. Because the estimates are driven by demographic change
only, evaluations of these other land-use sectors would be nearly
impossible within the current structure of the model (clearly another
weakness that could be named in answer to Question 3). Clearly the loss
of farmland to housing can be represented, but not the creation of new
farmland to make up for this loss and because of incentives for biofuels,
or the abandonment of marginal farmlands, nor the afforestation of large
residential lots in the east.
These are good comments; the
authors have modified section
5.4 to include some of these
suggestions.
Daniel If there is an interest in linking to climate models, the outcomes would
Brown need to be translated into terms that can be used to represent latent and
sensible heat fluxes (LAI, surface roughness, NPP). These can also be
important in understand hydrological impacts, through integration with
eco-hydrological models. The report suggests that it would be possible to
calculate changes in vehicle miles traveled (VMT), as a result of
changing settlement patterns, which could then go into emissions
estimates. The data could be combined with variables that relate to
climate sensitivity, like water availability, temperature extremes, air
conditioning availability, etc, and used in assessments of human and
community vulnerability under alternative climate scenarios.
These are good comments; the
authors have modified section
5.4 to include some of these
suggestions.
Steven Endpoints. The endpoints of housing density and impervious surface
Manson cover are useful endpoints given the goal of the model. There are others
that would be helpful in future studies, as described below, such as non-
urban land uses like agriculture or a more explicit focus on
transportation. This said, the land allocation schema can be used to assess
impacts on all land types and it incorporates transportation networks and
commute times, which in turn could be used to ascertain transportation-
related impacts (e.g., commuting times and pollutant emissions).
These are good comments; the
authors have modified section
5.4 to include some of these
suggestions.
Steven Imperviousness. A key advantage of imperviousness is that it can be tied
Manson to the rapidly expanding literature on linkages between impervious
surfaces and a range of environmental impacts. Overall, imperviousness
is one useful proxy for environmental impact (complementing the land
cover impacts of residential density) and the report authors are clear to
note that this is just one step towards a full understanding of land
use/climate interactions.
No response necessary
-------
Reviewer
Reviewer Comments
EPA Response
Steven Residential density. The advantages of imperviousness hold for
Manson residential density. The chief difficulty faced in any kind of modeling,
but especially with future land use and ecosystem services modeling, is
trying to incorporate changes in the economic or technological basis for
impact estimation. One advantage of the demographic focus of this
model is that it can be tied to different economic or technological
dynamics, especially as it relates to housing density (e.g., changes in
housing technologies) and other aspects of the human system.
No response necessary
Dawn These two endpoints are very important for water quality analysis. Many
Parker other endpoints are also as important, including changes in forest cover,
the carbon sequestration profile of converted landscapes, and calculations
of vehicle miles traveled and congestion of road networks under different
scenarios.
These are good comments; the
authors have modified section
5.4 to include some of these
suggestions.
David This has been addressed above as well. Clearly any use of the models to
Skole estimate mobile source emissions could be quite valuable.
No response necessary.
David As well, the modeling approach could take advantage of scenarios and
Skole parameters that take account of land competition. For instance, one could
estimate a rate of penetration of bio-fuels into the fuel mix and estimate
the land area needed - first for grain and then for cellulose - to constrain
the modeled built area expansion. This could be a first order estimate of
the effect of biofuels on the geographic distribution of land use change. It
would have the effect of constraining the spatialization of housing
density - perhaps in much of the same way as does the Commercial and
Industrial Land Use (see page 26 line 26). To this could be added a
transport cost for biofuel - i.e., the production and processing being done
in low housing density areas (rural) and the consumption being done in
the predicted high density regions (east and west coast urban). It could
frame the start of an analysis of the bio-fuel infrastructure requirements,
and also the emissions from production to consumption locations.
These are good comments; the
authors have modified section
5.4 to include some of these
suggestions.
David Another endpoint related to bio-fuels could be to build a scenario in
Skole which new land expansion is a function of biofuel requirements rather
than housing. Instead of driving the model with population, use an
estimate of biofuel land demand and the SERGoM model, to estimate the
spatialization of grain and/or cellulose expansion.
A quick estimation of the amount of land needed to meet all our liquid
fuel demand using grain alcohol has been attempted. The current land
base supporting grain wheat and bean production in the US is
approximately 250 M acres. Grains comprise approx half of this, or about
100 M acres. Of this amount, approximately 23% is now devoted to
ethanol fuel production - about 23 M acres ~ and this amount produces
3% of US fuel. The global average is closer to 5%. Using the global
value, we would need to increase ethanol production by 20-fold over
current levels to meet 100% fuel needs from grain ethanol. In the US that
would mean increasing the acreage from 23 to 400 M acres. This would
exceed the total available cropland by 2 fold and would increase the grain
No response necessary
24
-------
Reviewer
Reviewer Comments
producing regions by 4-fold by 2100.
EPA Response
-------
(5) What model modifications, additional analyses, or additional endpoints would you recommend to
include in a future study?
Reviewer
Reviewer Comments
EPA Response
Daniel
Brown
Depending on computational resources available for this
approach, develop an interactive interface that allows users to
interact with the scenarios and incorporate stochasticity into the
estimation process so that users can see and evaluate the
consequences of the range of possible outcomes given
uncertainty and variability in the inputs.
The availability of the tool is
now discussed in the preface.
In the Options for Future
Study, we now indicate that
adding stochastic variation will
be considered as a possible
improvement.
Daniel
Brown
Consider ways to move towards including other land use
changes, including in the agricultural and forest sectors.
Mentioned in options for future
work.
Daniel
Brown
Also, consider representing variability in the impacts residential
development both across the density categories and regionally. I
would think that the impacts of a given level of imperviousness
vary by ecosystem type and that this variability renders simple
categorizations like that used here relatively flawed. Consider
just reporting percent impervious by watershed, rather than
categories of stress level.
Thanks ~ we agree that it
would be interesting to
examine how IS changes as a
function of ecosystem ~ and
have added this as a suggested
future analysis. We have used
a legend that applies categories
of stress level that is based on
past literature and we believe
that it generally holds up well.
Of course the raw %IS are
provided in the datasets and so
those could be used as well if
the categorical legend is not
desired.
Daniel
Brown
I think the density categories should be dynamic, but it seems
that they probably are not. This may not be important, since the
relationship between density and impervious surface is
continuous and not based on the categories. However, I think the
allocation still is based on the categories.
Allocation of housing units is
not based on IS classes, rather
IS is an output or function of
housing units. This section was
clarified in the text.
Daniel
Brown
Run additional scenarios that try to bracket better the high and
low impact outcomes (i.e., explore the space for the best and
worst outcomes on some measure) to identify desirable and
undesirable conditions and the conditions under which they
Mentioned in options for future
work.
occur.
Daniel
Brown
Explain why change in imperviousness is at such a high rate in
the plains under scenario B2. I understand that it's based on a
small denominator, but why the increase - is this an artifact of
not allowing people to move out from small counties?
We included maps showing
absolute IS and relative change
in IS due to artifacts caused by
small denominators (see
Figures 5-25 and 5-26, for
example). Those counties are
all very small, with populations
ranging from a few hundred to
-------
Reviewer
Reviewer Comments
EPA Response
about 1,000 people. You are
correct that this is an artifact of
small counties' exemption
from the migration model. In
general, our approach has some
drawbacks when modeling the
smallest counties.
Daniel
Brown
Follow up on the suggestion to analyze the effects of alternative
scenarios on vehicle miles traveled (VMT) so that the settlement
pattern scenarios can be fed back into the emissions estimation
process.
Mentioned in options for future
work.
Steven Endpoints. Land use/land cover more broadly conceived is probably the
Manson most likely candidate for a new endpoint if the model were to be
expanded. Given that the model examines changes in residential density
as a result of conversion, another promising direction is examining the
balance among non-residential uses such as agriculture or forestry. The
way in which the scenario results are linked to NLCD is a step in this
direction (e.g., the examination of wetland impacts), which leaves room
for a complementary, explicit agriculture submodel, for instance.
Mentioned in options for future
work.
Steven A greater focus on transportation (especially linkages among vehicular
Manson traffic, infrastructure development, and urban growth) would also be
helpful to better specify commuting effects or better illustrate feedbacks
between land use development and transportation. The chief difficulty
with dealing with transportation/land use linkages is that there are few
truly integrated land use/transportation models that can operate at the
regional scale in a way that would work with SERGoM. This is an area
of future research more broadly in civil engineering and social science.
Mentioned in options for future
work.
Steven Scenarios. One area of additional potential explication is further
Manson emphasizing that the global scenarios (especially A1/B1) do not
necessarily account for the actual 'story line' of the relationship between
demographics and economic development, given the complex
interactions subsumed in this relationship. This said, the report is careful
to examine how the scenarios are open to interpretation (e.g., page 28).
Overall, the qualitative interpretation of the scenarios is plausible (page
7).
No response necessary
Steven PUMA interpolation. One potentially useful extension would be to
Manson investigate the effect of apportioning PUMA data spatially amongst
counties; relatedly, the basis for this apportionment could be more clear
(page 13). Allocating population via an areal interpolation mention that
accounts for settlement locations or some other secondary variable may
be a useful model extension, especially given the attention to using
settlement location in deriving the distance matrix. In terms of
verification and validation, internal validation of the model may be a
helpful approach (e.g., holding back some data from the calibration
phase) versus just assessing model fit and sensitivity (Appendix B), but
Thank you for these
suggestions. We have updated
the text in section 3.5.1 to
better describe how PUMAs
were aggregated and
disaggregated. When migration
records for PUMAs were
disaggregated among two or
more counties, data were
disaggregated to counties
28
-------
Reviewer
Reviewer Comments
EPA Response
comparison to state estimates is still the most important step of external
validation (page 22).
based on total county
population. Although the
distance matrix takes
settlement location into
account, population
distribution within counties
does not affect the
demographic projections in any
other way. Future work may
allow us to test the effect of
our chosen method of PUMA
disaggregation, or better data
(such as IRS records) may
allow us to take an alternative
approach.
Steven Land types. In future applications, it would be good to conduct
Manson sensitivity testing on how the model deals with the balance between
commercial/industrial land use vs. infill/brownfield development (page
26). This may have particular relevance for the 50+ population group
given their role in reverse migration (e.g., their influence on downtown
condominium development). This lack does not call into the model
projections given other sources of variability (per Appendix B), but it is
an increasingly important factor in the United States, given the graying of
the population.
Mentioned in options for future
work.
Dawn The importance of this modeling task cannot be underestimated. Land-
Parker use change has been estimated to account for up to around 25% of
anthropogenic carbon contributions, and global land-use change models
require robust land-use and land-cover change estimates (Parker, Hessl,
and Davis 2008). Modelers in other part of the globe, where resources
and data are better than we have in the US, are probably 20 years ahead
of us in terms of the development of regional and national land-use
change models. National level carbon policies for the US are likely to be
developed in the near future. Yet, the modeling community is not yet
able to provide policy makers with robust, validated national level land-
use change models that are based on cutting-edge science. Given that
context, the modeling effort described in this report represents a
significant and important investment by EPA.
No response necessary
Dawn I suggest an adaptive, exploratory modeling strategy where several
Parker alternative models are developed, and model projections are formally
compared using standard verification and validation tools. This model
and its future modifications could be a part of that effort. However, the
outputs of this model should be compared to real-world data, and
especially to projections from other related models on a regional and
statewide basis where possible. Ideally an alternative model should be
developed that is more structural and process-based (including models
that feed back across time and space and more drivers of LUCC).
Investments should be made to facilitate sharing of information about
No response necessary.
-------
Reviewer
Reviewer Comments
EPA Response
models and results, so that all existing relevant LUCC models and
examples of LUCC models coupled with water quality, transport, and
carbon models can be accessed and compared. There will not be a single,
static answer to the question of which modeling tool will best project
LUCC into the next century. Again, we need an aggressive national
program to support adaptive, scientifically grounded LUCC modeling
efforts. I strongly believe that we cannot build effective water quality, air
quality, and carbon policies based on sub-adequate land-use and land-
cover change models, and given my extensive interactions with other
researchers through conferences, expert workshops, and scientific
publication, I believe that other LUCC researchers share this view. For
instance, the completed LUCC project and the new Global Land Project
(http://www.globallandproject.org/), for which I serve on the scientific
steering committee, place a high priority on development of land-use and
land-cover data.
There is a desperate need to improve the quality, quantity, and
availability of data inputs for regional and national level land-use change
models. Better coordination is also needed between government agencies
related to land-use and land-cover data generation, documentation,
archiving, and sharing. Many data resources exist that are simply not
available to researchers and/or are not available across agencies.
Dawn
Parker
Examples of data limitations for this model:
p. 123.5.1: The lack of county-to-county migration data is a major
concern. The lack of overlap between counties and PUMAs is another
concern. Both have caused down-scaling in this study that is potentially
problematic. I suggest Monte Carlo simulation (see, for example, (Lewis
and Plantinga 2007)) to evaluate the sensitivity of results to down-scaling
algorithms. It would also be helpful to have data on household, rather
than individual, migration, and model migration and location choices at
the household scale.
p. 18 19-21: The authors note additional data limitation related to
demographic factors.
IRS records provide one
potential source of household
migration that we may
investigate in the future.
However, the PUMA-to-
county transition is not
necessarily as problematic as it
sounds. All large population
centers involved the grouping
of PUMAs (where no error is
introduced) rather than the
apportionment of PUMAs.
This covered over 70% of the
population. Admittedly, our
methods would have greater
uncertainty with smaller
counties. We added text
elaborating on our methods
and acknowledging both of
these concerns.
Dawn In general: Data on housing density are needed at a finer scale than
Parker census units to validate this model. Such data are available only
sporadically at a national level, and access and costs to data, when they
exist, are uneven.
No response necessary
David The strength of this approach is the use of a spatial allocation model.
Skole However, it would be worth exploring additional ways to spatialize rather
These are helpful concerns. We
have added some of these
30
-------
Reviewer
Reviewer Comments
EPA Response
than simply on population and weighting a distance function. For
instance, economic growth is a key factor well known to influence
expansion of the built area. Incorporating REMI type models in a spatial
context would be an important future modification. Additionally an
improved capability to look at land competition and trade-offs would be
useful. Lastly, building more feedbacks into the model so that climate
affects on land use could be taken into account would be useful.
points into the discussion of
future steps. Adding a REMI-
type model would be useful,
but was not feasible in this
round, hence the focus on
developing scenarios to
explore a range of possible
outcomes.
David I would suggest the team review carefully the modeling work being
Skole developed at the Joint Global Change Research Institute at the University
of Maryland. Their EPIC model could greatly enhance the agriculture
modeling of the EPA effort (NB I have no affiliation at all with the UMd
team).
Thanks, this would be useful to
explore to incorporate an
agricultural land.
David A diversity of land-change models exists that explain, predict, and project
Skole the kind and location of change in land covers and land uses. Below I list
a number of references that could be useful in considering different
approaches to modeling, some of which would help the team build more
process-level capabilities into their approach.
A variety of modeling approaches are used to improve our understanding
of land change and to encode that understanding for these purposes of
projection and prediction. These approaches include stochastic,
optimization, supply and demand, dynamic, process-based simulation,
cellular automata, agent-based, and a variety of statistical-empirical
models. Coupling land-change models with models of biogeochemical,
water, and ecological processes faces a number of challenges but could
be part of the EPA future efforts. The spatial and temporal scales of
land-change models need to be compatible with both the driving
processes of land change and process models of environmental systems,
and the land change and environmental models must share specific
semantic, onotological, and technical specifications in order to allow
inter-model communication and coupling. Thus, although there has been
much research that contributes to our understanding of land-use and land-
cover change, from an observational or empirical basis, there remains a
need to develop models of land-use and land-cover changes at spatial
scales from local to global, and time scales from short (<5 years) to long
(> 50 years), that are compatible with environmental models relevant for
the CCSP and other agencies and programs needs.
Thanks, these are useful
thoughts and a number of
citations to other modeling
approaches have been added,
including adding an item to
future steps.
David Land change and the reciprocal interactions with environmental and
Skole socio-economic systems have direct and indirect impacts on the health
and sustainability of society and of ecosystems yet these are poorly
developed in the EPA approach. A synthetic understanding of land-
change modeling approaches is needed so that these reciprocal relations
can be both studied, in the case of explanatory models, and projected
through computer-based tools that encode the best scientific
understanding and allows the wide-ranging application benefits agency
programs to be realized. Importantly, the study will provide guidance to
a wide range of science- and application-based model users on the
No response necessary.
-------
Reviewer
Reviewer Comments
EPA Response
strengths and weakness of the various approaches. Such guidance is not
currently widely available.
David Another fruitful area of future enhancements would be in coupling land-
Skole change models with models or biogeochemical, water, and ecological
processes faces a number of challenges. The spatial and temporal scales
of land-change models need to be compatible with both the driving
processes of land change and process models of environmental systems,
and the land change and environmental models must share specific
semantic, onotological, and technical specifications in order to allow
inter-model communication and coupling.
We have added some of these
suggestions to the discussion
of future steps. We have also
improved the introduction to
better describe what this
approach is and isn't suitable
for.
David
Skole
Addition Material: a brief survey of models:
Stochastic
Brown, D. G., Pijanowski, B. C. and Duh, J. D. (2000). Modeling the
relationships between land use and land cover on private lands in the
Upper Midwest, USA. Journal of Environmental Management 59(4):
247-263.
Butcher, J. B. (1999). Forecasting future land use for watershed
assessment. Journal of the American Water Resources Association 35(3):
555-565.
Muller, M. R. and Middleton, J. (1994). A Markov Model of Land-Use
Change Dynamics in the Niagara Region, Ontario, Canada. Landscape
Ecology 9(2): 151-157.
Thornton, P. K. and Jones, P. G. (1998). A conceptual approach to
dynamic agricultural land-use modelling. Agricultural Systems 57(4):
505-521.
Optimization
Riebsame, W. E., Meyer, W. B. and Turner, B. L. (1994). Modeling
Land-Use and Cover as Part of Global Environmental- Change. Climatic
Change 28(1-2): 45-64.
Supply and demand
Waddell, P. (2000). A behavioral simulation model for metropolitan
policy analysis and planning: residential location and housing market
components of UrbanSim. Environment and Planning B-Planning &
Design 27'(2): 247-263.
Dynamic, process-based simulation
Landis, J. and Zhang, M. (1998). The second generation of the California
urban futures model. Part 1: Model logic and theory.
Environment and Planning B-Planning & Design 25(5): 657-
666.
Landis, J. D. (1994). The California Urban Future Model: a new
generation of metropolitan simulation models. Environment and
Planning B-Planning & Design 21: 399-421.
Thank you.
32
-------
Reviewer
Reviewer Comments
EPA Response
Stephenne, N. and Lambin, E. F. (2001). A dynamic simulation model of
land-use changes in Sudano- sahelian countries of Africa
(SALU). Agriculture Ecosystems & Environment 85(1-3): 145-
161.
Cellular automata
Clarke, K. C. and Gaydos, L. J. (1998). Loose-coupling a cellular
automaton model and GIS: long-term urban growth prediction
for San Francisco and Washington/Baltimore. International
Journal of Geographical Information Science 12(7): 699-714.
Clarke, K. C., Brass, J. A. and Riggan, P. J. (1994). A Cellular-
Automaton Model of Wildfire Propagation and Extinction.
Photogrammetric Engineering and Remote Sensing 60(11):
1355-1367.
Jenerette, G. D. and Wu, J. G. (2001). Analysis and simulation of land-
use change in the central Arizona-Phoenix region, USA.
Landscape Ecology 16(7): 611-626.
Messina, J. P. and Walsh, S. J. (2001). 2.5D Morphogenesis: modeling
landuse and landcover dynamics in the Ecuadorian Amazon.
Plant Ecology 156(1): 75-88.
van der Veen, A. and Otter, H. S. (2001). Land use changes in regional
economic theory. Environmental Modeling & Assessment 6(2):
145-150.
White, R., Engelen, D. and Uljee, I. (1997). The use of contrained
cellular automata for high resolution modelling of urban land use
dynamics. Environment and Planning B 24(3): 323-343.
White, R., Engelen, D. and Uljee, I. (2000). Modelling land use change
with linked cellular automata and socio-economic models: a tool
for exploring the impact of climate change on the island of St
Lucia. Spatial Information for Land Use Management. Hill, M. J.
and Aspinall, R. J. Reading, Gordon and Breach: 189-204.
Agent-based
Ligtenberg, A., Bregt, A. K. and van Lammeren, R. (2001). Multi-actor-
based land use modelling: spatial planning using agents.
Landscape and Urban Planning 56(1-2): 21-33.
Otter, H. S., van der Veen, A. and de Vriend, H. J. (2001). ABLOoM:
Location behaviour, spatial patterns, and agent-based modelling.
Jasss-the Journal of Artificial Societies and Social Simulation
4(4): U28-U54.
-------
(6) Please comment on the public comments submitted for this draft report. Specifically, which
comments should or should not be addressed in the final draft?
Reviewer
Reviewer Comments
EPA Response
Daniel I think the statistical methods are fine. Clearly multicollinearity
Brown problems need to be dealt with and stepwise processes are a reasonably
standard way to deal with them. Clearly missing variables add little to the
predictive power of the model (given the contribution of those included),
which is the key measure of importance in this case. The use of
Classification and Regression Trees (CART) is appropriate in this case.
The authors of the report mistakenly refer to the technique as categorical
regression trees, but in fact the method (named correctly in the previous
sentence) can deal with continuous measures in the form of regression
trees. It's true that it produces discrete estimates, but it is an appropriate
method for continuous measures that does not, in fact, throw out detail in
the data.
CART was corrected in the
text.
Daniel I understand the emphasis on impervious surfaces; though also recognize
Brown the importance of other land changes and mitigation activities by
developers, farmers and other land users. This point about mitigation
also goes to the heterogeneity of the impacts of impervious surfaces and
the critique would be mitigated if the authors backed off on the absolute
categorization of all impervious levels into levels of ecosystem stress.
There is variability in the relationship between housing density and
imperviousness and a stochastic modeling approach could be used to
introduce that variability. I don't believe that it would have a huge
impact at the national level, but it might also address some of this
concern.
Thanks, we reworded the text
to place less emphasis on the
legend classed (e.g., "stressed")
and more on the quantitaive
value. We also provided a
caveat not to interpret the
relative designations too
strongly, as these are general
indicators of condition only.
However, we also reiterate that
one of the strongest indicators
of watershed health,
substantiated by numerous
studies, is % of impervious
surface, which is why this is an
important indicator, and why it
is important to help interpret
what the general numbers
mean in a qualitative way for
the general public.
Daniel The suggestion of looking at the effects individual variables separately in
Brown the scenarios is a reasonable one, if the goal is to tease out these
individual effects. I don't actually get the sense that the goal is to test
what is the more important factor, as implied by this critique, but rather
to project plausible scenarios. For that goal, the bundled nature of the
scenarios presented is reasonable.
We added some clarifying text
in Section 2.2.
Daniel While I agree that evaluations of Smart Growth alternatives would need
Brown to be carefully defined before they can be implemented for scenario
development, I don't see any implication in the report to the contrary.
Nor do I see any conclusions drawn with respect to Smart Growth that
could be regarded as at all controversial (as there are none).
No response necessary
-------
Reviewer
Reviewer Comments
EPA Response
Steven
Manson
There was one attached comment, from the National Association of
Home Builders (NAHB). Overall, the NAHB comments have merit and
should be taken into account as they relate to four general issues: 1) the
choice of statistical techniques; 2) the emphasis on impervious surface
cover; 3) the scenarios used to assess the impact of land development
patterns; and 4) references to Smart Growth.
a) There is room to modify or better explain the statistical
techniques. The caveats that the NAHB raises about stepwise
regression are valid but this approach is a common social science
method. As with most statistical methods, the analysis and
degree of expertise applied to the analysis is usually more
important than the potential foibles of the method. Concerns
about multicollinearity could be addressed in the data preparation
or model specification steps (e.g., via pairwise comparisons) but
including the full model specification after removing
multicollinear variables would nonetheless be useful. Otherwise,
the report could better explain the rationale and process for using
CART to derive imperviousness (page 39, Appendix C).
More explanation was added
for CART.
Steven
Manson
b) The emphasis on imperviousness is an issue in that there exist
other aspects of land cover that can be considered, as noted
above under question 5. Nonetheless, imperviousness is an
important variable and a useful one when tied to residential
density.
No response necessary
Steven
Manson
c) The scenario-land use linkages could use more explication in the
report, but overall, the way the scenarios are employed here are a
useful and accepted way of understanding issues we may
encounter in the future. Per comments under question 5 above
about scenarios, more could be specified under scenario
development, but the report is clear in most respects.
No response necessary
Steven
Manson
d) The report could be clearer in how it refers to Smart Growth
(SG). There is a growing body of empirical research linking SG
to a range of impacts. While these impacts tend to be negative in
many respects, there are counter examples and areas of ongoing
research that should be recognized (e.g., Handy, S. 2005. Smart
Growth and the Transportation-Land Use Connection: What
Does the Research Tell Us? International Regional Science
Review 28 (2): 146-167.) More broadly, however, it appears that
the report is not speaking to the pros/cons of smart growth per se,
but instead to the impacts of 'environmental' perspectives
towards land use planning. If regional and urban planners believe
that compact growth patterns are environmentally sensitive
(leaving aside whether they are or not) then they will likely
implement policies to produce these patterns. This seems to be
the tack adopted by the report (page 51), but it could be more
clear on this point.
Language referring to smart
growth is clarified so that it is
clear we are referring to low
impact development with the
term.
36
-------
Reviewer
Reviewer Comments
EPA Response
Dawn The NAHB comments should be addressed. It is important to note that
Parker home builders have some incentive to protect the natural resource base,
since the amenity and non-use values within and in the local
neighborhood of their developments are captured in the sales prices of
their homes. They may provide local public goods through these
incentives. However, since they are not able to capture the benefits from
the global public good aspects of open space (such as climate regulation),
there is still an important role for agencies such as the EPA for protecting
open space and the ecosystem functions that it supports.
We have improved our
discussion of CART and IS,
and changed how we discuss
Smart Growth. Please see
responses to other comments
above.
Dawn I share their concern regarding the stepwise regression. While my
Parker knowledge of categorical regression is limited, their comments are
logical from a statistical perspective. I also agree (as stated above) that
vehicle miles traveled are important to examine. Their suggestions for
alternative scenarios are also a potential next step that deserves
consideration. While I don't share their concern regarding the current
references to smart growth, the more detailed investigations between
smart growth policies and environmental impacts would be of broad
interest for future work.
We have improved our
discussion of CART. We have
added VMT as a possible area
for future work.
David This reviewer received only one public comment, from the National
Skole Association of Home Builders. They address the following comments
and I have remarks associated with each of them.
Choice of particular statistical techniques: The comments are valid but do
not appear to be strong enough to be further addressed in any significant
way. As I have commented before, there is a general tendency in the text
not to be explicit about the choices made in methods. I think the authors
owe the reader an explanation of alternatives and why the methods
selected for this study were chosen. Again, the lack of sufficient
validation exercises leaves the report open to these criticisms.
We added some discussion of
the use of statistical methods in
the impervious surface
analysis.
David Emphasis on percent of impervious surface cover: I generally agree with
Skole this comment by the NAHB, and have raised that issue above. Unlike the
NAHB I can see some linkages between housing density and emissions,
but the link to impervious surface is less strong. One could develop an
urban heat island model, or perhaps develop a runoff model that would
be influenced by storm intensity, but these are a stretch. I must agree with
the NAHB that this emphasis on IS needs considerable justification.
We added more discussion
about why we chose to look at
IS.
David Scenarios used to assess the impact of land development patterns: I agree
Skole with this concern of the NAHB. There is a strong disconnect in logic with
the selection of the Story Lines and the prediction of IS. To remedy this, I
suggest the authors strengthen the analysis and discussion of outright
land use change - i.e., from agriculture or forest to built and then
consider the direct emissions issues associated with these changes.
Reduce the level of discussion and emphasis on IS. Generally speaking
the IS discussions in section 5.3 (page 39) do not logically fit in this
analysis.
We added more discussion
about why we chose to look at
IS. In the Options for Future
Work section, we added that
future improvements may
involve a stronger focus on
other land use changes beyond
housing density.
-------
Reviewer
Reviewer Comments
EPA Response
David References to Smart Growth: The comments of the NAHB are baseless
Skole and should not be considered by the EPA. Smart Growth is an
unfortunate use of terms in the EPA study, and perhaps a different term
could be used. I would recommend some references to the work of the
Urban Policy Center of the Brookings Institution for references to the
urban decentralization problem and a discussion.
Language referring to smart
growth is clarified so that it is
clear we are referring to low
impact development with the
term
38
-------
Additional Reviewer Comments
Reviewer
Reviewer Comments
EPA
Response
Daniel Although the specific goals of the document are not well articulated in the body of the
Brown report, the report does suggest (p. 1) that its results will "(1) enable us, our partners, and
our clients to conduct assessments of both climate and land use change effects across the
United States: (2) provide consistent benchmarks for local and regional land use change
studies; and (3) identify areas where climate-land use interactions may exacerbate impacts
or create adaptation opportunities." These goals are important and there is clearly a need
within the scientific community to bridge the modeling of land-use and climate change,
assess their interactions, and evaluate the possibility for interacting impacts. The executive
summary (p. x) indicates that "This report describes the modeling methodology for the
EPA-ICLUS project and some initial analyses using the outputs." This is more a
description of its content than its goals, but it does make clear that the document is a first
step, rather than a complete assessment.
We have
revised the
Introduction
and Executive
Summary to
better describe
the study's
goals.
Dawn
Parker
References:
Clark, W. A. V., and F. J. Van Lierop. 1987. Residential mobility ans household location
modeling. In P. Nijkamp, ed. Handbook of Regional and Urban Economics.
Elsevier Science Publishers, Amsterdam
Engelen, G., R. White, and A. C. M. de Nijs. 2003. Environment Explorer: a Spatial Policy
Support Framework for the Integrated Assessment of Socio-Economic and
Environmental Policies in the Netherlands. Integrated Assessment 4:97-105.
Irwin, E., and Bockstael. 2002. Interacting agents, spatial externalities, and the evolution
of residential land use patterns. Journal of Economic Geography 2:31-54.
Jantz, C.A., S. J. Goetz, and M. K. Shelley. 2003. Using the SLEUTH Urban Growth
Model to Simulate the Impacts of Future Policy Scenarios on Urban Land Use in
the Baltimore-Washington Metropolitan Area. Environment and Planning B
30:251-271.
http://scholar.google.com/url?sa=U&q=http://www.whrc.org/resources/
Published literature/pdf/JantzEnvPlanB .03 .pdf
Landis, J., and M. Zhang. 1998. The second generation of the California Urban Futures
Model: Part 1, Model logic and theory. Environment and Planning B: Planning and
Design 25:657-666.
http://www.pion.co .uk/ep/epb/abstracts/b25/b25 065 7 .html
Lewis, D., A.J. Plantinga. 2007. Policies for Habitat Fragmentation: Combining
Econometrics with GIS-Based Landscape Simulations. Land Economics 83:109-
127.
Parker, D., A. Hessl, and S.C. Davis. 2008. Complexity, Land-Use Modeling, and the
Human Dimension: Fundamental Challenges for Mapping Unknown Outcome
Spaces. Geoforum 39:789-804.
http://dx.doi.0rg/10.1016/i.geoforum.2007.05.005
Thank you for
these
additional
references.
-------
Reviewer
Reviewer Comments
EPA Response
Dawn Verburg, P., K. Kok, R.G. Pontius, A. Veldkamp, A. Angelsen, B. Eickhout, T.
Parker Kram, S.J. Walsh, D.C. Parker, K. Clarke, D. Brown, K.P. Overmars,
and F. Bousquet. 2006. Modeling land-use and land-cover change. In E.
Lambin and H. Geist, eds. Land-use and Land-cover Change: Local
Processes, Global Impacts. Springer Berlin Heidelberg, New York
Verburg, P., M. Rounsevell, and A. Velkamp. 2006. Scenario-based studies of
future land use in Europe. Agriculture, Ecosystems & Environment 14:
1-6.
http://dx.doi.0rg/10.1016/i.agee.2005.ll.023
Verburg, P.H., W. Soepboer, A. Veldkamp, R. Limpiada, V. Espaldon, and
S.S.A. Mastura. 2002. Modeling the spatial adynamics of regional land
use: The Clue-S model. Environmental Management 30:391-405
Waddell P. 2002. UrbanSim: Modeling Urban Development for Land Use,
Transportation and Environmental Planning. Journal of the American
Planning Association 68:297-314.
38
------- |