vvEPA
United States
Environmental Protection
Agency
EPA/600/R-20/309 August 2020 www.epa.gov/ord
w
A Systematic Approach
for Selecting Climate
Projections to Inform
Regional Impact
Assessments
Office of
Research and Development
Center for Public Health &
Environmental Assessment
Health & Environmental
Effects Assessment Division
-------
EP A/600/R-20/3 09
August 2020
FINAL
A Systematic Approach for Selecting Climate Projections to
Inform Regional Impact Assessments
Center for Public Health and Environmental Assessment
Office of Research and Development
U.S. Environmental Protection Agency
Washington, DC
-------
DISCLAIMER
This document has been reviewed by the U.S. Environmental Protection Agency, Office
of Research and Development, and approved for publication. Any mention of trade names,
products, or services does not imply an endorsement by the U.S. Government or the U.S.
Environmental Protection Agency. The EPA does not endorse any commercial products,
services, or enterprises.
ABSTRACT
The increasing volume of climate model output creates challenges for those seeking to
understand or use those projections in a way that is scientifically sound, but also efficient with
respect to time and resources. A common approach to resolving these competing goals is to
identify a subset of the available climate projections that still describes the relevant
characteristics of the entire suite. This report synthesizes and describes alternate approaches for
systematically identifying a set of climate projections that are best suited for a user-defined
research question or objective. The advantages and disadvantages of these approaches are
highlighted, and generally depend on a tradeoff between including more (lower risk tolerance) or
fewer (less information to process) climate projections. This report provides information in the
context of a new Web-based tool: Locating And Selecting Scenarios Online (LASSO). The
LASSO tool automates much of the selection process by guiding users through a step-by-step
procedure of first building a scatterplot visual representation of a suite of climate projections,
then assisting the user in identifying the projections that most closely align with their specific
concerns or questions. The report presents four approaches for sub-setting climate projections
that are generally suitable for a variety of applications and includes exemplars for each of the
EPA Regions. The tool includes a much larger suite of pre-computed scatterplots, maps, and
spatial data that describe climate projections by EPA Region, state (contiguous), two scenarios,
and both annual and seasonal summaries.
Preferred citation:
U.S. EPA. A Systematic Approach for Selecting Climate Projections to Inform Regional Impact
Assessments. U.S. Environmental Protection Agency, Washington, D.C., EPA/600/R-20/309, 2020.
2
-------
CONTENTS
DISCLAIMER 2
ABSTRACT 2
LIST OF TABLES 4
LIST OF FIGURES 4
LIST 01 ABBREVIATIONS AM) ACRONYMS 5
AL I I IORS AM) REVIEWERS 5
AUTHORS 5
REVIEWERS 6
QUALITY ASSURANCE 6
ACKNOWLEDGEMENTS 6
1. EXECUTIVE SUMMARY 7
2. INTRODUCTION AM) BACKGROUND 10
2.1 PURPOSE AM) SCOPE 01 REPORT 10
2.1.1 Context 10
2.1.2 Scope 11
2.2 NEED AND RATIONALE FOR SCENARIO SELECTION 12
2.3 IDENTIFYING CLIMATE INFORMATION NEEDS 13
2.3.1 Climate Information in Decision Making 13
2.3.2 Identifying Information Relevant to Decision Context 14
3. EPA'S LASSO TOOL: A PRACTICAL APPROACH TO SELECTING CLIMATE
PROJECTIONS 19
3.1 PURPOSE OI EPA'S LASSO TOOL 19
3.2 SELECTING CLIMATE PROJECTIONS 19
3.3 SELECTION STRATEGIES IN THE LASSO TOOL 20
3.3.1 Lasso 22
3.3.2 Four Corners 23
3.3.3 Middle Corners 24
3.3.4 Double Median 25
3.4 OUTPUTS IROYl THE I.ASSO TOOL 25
3.5 CURRENT LIMITATIONS OF THE PROCESS AND TOOL 26
4. BACKGROUND ON SCENARIOS, MODELS, AND SELECTION CRITERIA 27
4.1 CLIMATE SCENARIOS 27
4.2 DOWN SCALED CLIMATE INFORMATION 27
4.3 MODEL INTERDEPENDENCE AND PERFORMANCE 28
4.4 STRATEGIES FOR SELECTING PROJECTIONS 30
3
-------
4.4.1 Overview of Selection Criteria 30
4.4.2 Finding the "Best" Model(s) 31
5. LASSO SCATTERPLOTS FOR EPA REGIONS 33
5.1 REGION 1 33
5.2 REGION 2 34
5.3 REGION 3 35
5.4 REGION 4 36
5.5 REGION 5 37
5.6 REGION 6 38
5.7 REGION 7 39
5.8 REGION 8 40
5.9 REGION 9 41
5.10 REGION 10 42
6. CONCLUSION 43
REFERENCES 44
LIST OF TABLES
Table 1. Summary of scenario selection strategies 21
LIST OF FIGURES
Figure 1. Overview of approach for selecting climate projections to inform regional impact
assessments 12
Figure 2. Decision contexts and associated time horizons 16
Figure 3. Users of the LASSO tool have the option to customize results using an interactive
scatterplot to select climate projections 20
Figure 4. Illustrating the Lasso selection strategy. Black dots denote individual climate
projections. Red circles indicate models selected by the Lasso strategy 22
Figure 5. Illustrating the Four Corners selection strategy. Black dots denote individual climate
projections. Red circles indicate models selected by the Four Corners strategy. Dotted lines are
drawn at the minimum and maximum projected values for each axis 23
Figure 6. Illustrating the Middle Corners selection strategy. Black dots denote individual climate
projections. Red circles indicate models selected by the Middle Corners strategy. Dotted lines
are drawn at the 25th and 75th percentile projected values for each axis 24
Figure 7. Illustrating the Double Median selection strategy. Black dots denote individual climate
projections. Red circles indicate the model selected by the Double Median strategy. Dashed lines
are drawn at the median projected values for each axis 25
Figure 8. Effective number of models from CMIP3 based on statistical analysis of effective
degrees of freedom or effective sample size from a given dataset. The correlation values are
calculated based on model error structure 30
4
-------
Figure 9. LASSO scatterplots for EPA Region 1, excluding Puerto Rico and the Virgin Islands.
33
Figure 10. LASSO scatterplots for EPA Region 2 34
Figure 11. LASSO scatterplots for EPA Region 3 35
Figure 12. LASSO scatterplots for EPA Region 4 36
Figure 13. LASSO scatterplots for EPA Region 5 37
Figure 14. LASSO scatterplots for EPA Region 6 38
Figure 15. LASSO scatterplots for EPA Region 7 39
Figure 16. LASSO scatterplots for EPA Region 8 40
Figure 17. LASSO scatterplots for EPA Region 9, excluding Hawaii and island territories 41
Figure 18. LASSO scatterplots for EPA Region 10, excluding Alaska 42
LIST OF ABBREVIATIONS AND ACRONYMS
Coupled Model Intercomparison Project Phase 3
Coupled Model Intercomparison Project Phase 5
Environmental Protection Agency
General Circulation Model
Intergovernmental Panel on Climate Change
Locating And Selecting Scenarios Online
Regional Climate Model
AUTHORS AND REVIEWERS
The Integrated Environmental Assessment Branch within the Center for Public Health
and Environmental Assessment in the Office of Research and Development of the U.S. EPA is
responsible for publishing this report. This document was prepared by ICF International under
Contract No. EP-C-14-001, WA 3-93. Philip Morefield served as Technical Project Officer and
provided overall guidance and technical direction.
AUTHORS
U.S. EPA
Philip Morefield
ICF International, Washington, D.C.
Judsen Bruzgul, Brad Hurley, and Hannah Wagner
CMIP3
CMIP5
EPA
GCM
IPCC
LASSO
RCM
5
-------
REVIEWERS
U.S. EPA Reviewers
Christopher Weaver (ORD/CPHEA), Alexandra Dichter (Region 1/OARM)
External Peer Reviewers
Timothy Kittel (Univ. of Colorado), Amy Snover (Univ. of Washington), Claudia Tebaldi
(NCAR)
QUALITY ASSURANCE
This report has been reviewed and adheres to EPA QA and peer review policy
requirements. The work for LASSO was conducted under an approved EPA ORD Quality
Assurance (QA) Project Plan, Improving data availability andfunctionality of the LASSO Tool,
and has undergone internal technical review and external peer review. It is of known and
acceptable quality to support its intended use.
ACKNOWLEDGEMENTS
The authors would like to thank Thomas Johnson, Jeremy Martinich and the numerous
other EPA employees who provided early feedback on the LASSO tool and approach. The
thoughtful comments from reviewers substantially improved this report.
6
-------
1. EXECUTIVE SUMMARY
Advances in both the scientific understanding of the climate system and the capability to
produce climate simulations have led to a growing number of new and updated projections of
future climate suitable for use in impacts, vulnerability, and risk assessments in the United
States. The volume and complexity of this information present immense challenges for non-
specialists attempting to identify relevant projections for their specific analytic or assessment
needs. This report presents a practical approach for selecting climate projections, which are
simulated responses of the climate system based on a set of assumptions about changes in natural
and anthropogenic forcings. Specifically, it provides guidance to help users answer the question,
which climate projection(s) should I use? It accomplishes this by helping users visualize
strategies for capturing the key uncertainties represented by all the available model projections,
using a manageable number of representative projections that can serve as input into their
analysis or assessment.
The report's primary audience is U.S. Environmental Protection Agency (EPA) Regional
staff who may have limited familiarity with outputs from General Circulation Models (GCMs) or
underlying scenarios, but still seek a robust approach for integrating future climate information
into analyses or assessments that will support decision making at a regional level. Additionally,
the information in this report may be useful to anyone who needs U.S. climate projections for
analyzing impacts from long-term changes in temperature or precipitation patterns.
The report presents a practical approach for selecting climate projections with the aid of
an EPA online tool called LASSO (Locating And Selecting Scenarios Online). It also provides a
set of figures describing climate projections for each EPA Region, generated from LASSO
results.
Readers of this report will come away with the following:
1. An understanding of some key considerations for identifying climate information
that can be used to address decision needs.
2. A practical approach for selecting climate projections for meeting these needs,
including the function of the LASSO tool and strategies for using it to identify a
manageable but representative subset of projections.
3. Additional basic reference information about climate projections and available
climate models (GCMs), considerations related to selecting relevant model
projections, and the benefits of the LASSO approach.
7
-------
Need and Rationale for Scenario Selection
In seeking climate projections for an analysis or assessment, analysts confront the
contradictory goals of: (i) addressing deep uncertainties by describing a broad range of potential
future climates and (ii) minimizing the number of climate models included in the analysis in the
interest of practical time and resource constraints. Maximizing the range of futures helps bound
uncertainties that span future scenarios as well as the range of unique climate model projections
under those scenarios. Minimizing the number of climate models is necessary to ensure the
assessment remains manageable within time and resource constraints, and that a coherent
conclusion can be reached and communicated. The LASSO tool design helps the user
systematically, transparently, and efficiently balance these tradeoffs.
Identifying Climate Information Needs
The climate information used to support impact and vulnerability analysis should reflect
the decision-making context. While each analysis will have unique objectives and constraints,
analysts can consider several key elements when identifying climate information needs:
Temporal resolution and time horizon
Spatial resolution of the affected area
Risk or uncertainty tolerance of the decision maker
Relevant climate variables
This report provides an overview of each of these elements and describes how they
influence climate information needs.
EPA's LASSO Tool: A Practical Approach to Selecting Climate Projections
EPA's LASSO tool aims to streamline the process of selecting appropriate data for an
analysis, while at the same time reducing the overall volume of data that the analyst will need to
work with. The tool generates scatterplots of model projections for a specific EPA Region,
timeframe, and scenario, with selected climate parameters (such as precipitation and
temperature) on each axis. The user can then employ one of the projection selection strategies
described in this report to quickly and easily identify a manageable subset of projections that
bound the range of a larger group of projections in two dimensions simultaneously. The report
describes four strategies for identifying subsets of projections (Lasso, Four Corners, Middle
Corners, and Double Median) and discusses their advantages and disadvantages. Future versions
of LASSO may include additional features and functionality.
8
-------
Background on Scenarios, Models, and Selection Criteria
Climate models cannot simulate future changes without relying on assumptions about the
future variability of environmental factors that affect the global climate system. Because the
future can never be predicted with absolute certainty, climate scientists have developed a range
of scenarios that depict different possible pathways, which in turn lead to different combinations
of future climate forcings. Climate models use these scenarios to project (rather than predict)
how the climate would change under each hypothetical scenario. Climate research centers around
the world have developed many models of global and regional climate, each of which generates
somewhat different projections of future climate under the same scenario.
Under such currently irreducible uncertainty, an optimal approach to selecting climate
projections would incorporate as many climate projections as possible, perhaps including all
models and scenarios available. However, given the large number of potential combinations of
climate scenarios and models, attempting to select all data can be time-consuming, resource-
intensive, and a strain on data processing capabilities. Spending time at the outset to clearly
identify what climate information is needed to support the analysis can help an analyst identify
selection criteria to reduce the number of projections while still providing enough information to
be useful for decision making. The report describes potential selection criteria to consider,
including consistency with global projections, physical plausibility, applicability in impact
assessments, representativeness, accessibility, vintage, resolution, and validity. This report also
provides an overview of the need to consider interrelationships among models, some of which
share some of the same code base, in order to properly interpret projections from a range of
models.
Conclusion
The practical approach presented here and operationalized in the LASSO tool can assist
analysts and others with the task of selecting specific climate projections from a range of climate
models and scenarios to inform analyses of potential impacts and vulnerabilities. By considering
the climate information needed to support decision making, users can employ the approach best
suited to the decision context and analytical constraints. The example scatterplots provided
within this report provide a readily accessible starting point for those working in EPA Regions
across the country.
9
-------
2. INTRODUCTION AND BACKGROUND
Users must balance selecting more projections to better capture uncertainty and/or reduce
risk with selecting fewer projections due to practical constraints
Requirements and objectives of a specific decision should guide selection of relevant
climate information
To ensure climate information will be useful to the needs of decision makers, consider
temporal resolution and the decision time horizon, spatial resolution, tolerance for risk,
and relevant climate variables.
2.1 PURPOSE AND SCOPE OF REPORT
2.1.1 Context
As the study of future climate
impacts and vulnerability continues to
evolve, new and updated climate model
outputs are being produced that
incorporate the latest and best
understanding of the global climate
system. At the same time, new methods
and approaches are continually being
developed to regionalize or "downscale"
these global climate model outputs to
finer spatial resolutions more useful in
studies of potential future impacts. While
this new information can play a critical
role in the assessment of future climate risks, both the increasing volume and complexity of
information can make locating and identifying the most relevant and useful projections of future
climate for a particular application a confusing, laborious, and technically challenging task
(Moss et al., 2014). This is particularly true because, despite the many significant advances in
our understanding of the climate system, important aspects of future climate remain impossible
to predict, and climate projections are thus subject to intractable uncertainties that cannot be fully
accounted for in future climate studies.
Consequently, analysts or decision makers who need regional projections of future
climate to support a specific need or address a specific question are presented with an
overwhelming number of data sources and unique model projections from which to choose. They
can thus benefit from simple tools designed to efficiently scan and concisely summarize large
volumes of climate model output, while working within a framework of straightforward,
practical strategies for selecting the most relevant outputs for their purposes.
Key Terms
Climate Model: A numerical representation of the
climate system based on the physical, chemical, and
biological properties of its components, their
interactions and feedback processes, and accounting
for some of its known properties. In order to develop
projections, climate models rely on a set of
assumptions about air pollution and land use change
(scenarios or pathways) (IPCC, 2014).
Climate Projection or Simulation: The simulated
response of the climate system to a scenario or
pathway of future concentrations of air pollutants,
natural climate forcings, and other factors derived
using climate models (IPCC, 2014).
10
-------
2.1.2 Scope
In recognition of these challenges, this document provides information on practical,
systematic approaches for selecting a relevant and useful subset of climate projections from
existing sources of climate model output, to readily inform analyses of change impacts and
vulnerabilities.
The report is intended to be most useful for those with at least a modest understanding of
climate scenarios and projections whose aim is to integrate future climate information into
assessments to inform decision making at a regional level. The information in this report should
also be relevant to analysts conducting climate-related studies at national or local scales, and
decision makers seeking additional background on the range of potential future climate for a
given area. The practical approach for selecting climate projections is discussed in the context of
a new EPA online tool called LASSO (Locating And Selecting Scenarios Online), which assists
in the implementation of projection selection strategies.
This report is a resource that helps answer the question, which climate projections should
I use? It accomplishes this by helping users visualize strategies for capturing key uncertainties
represented by dozens of available model projections, and then identifying a manageable subset
of representative projections that serve as input into a given analysis or assessment. The report
presents principles that can be used to select climate information for analysis based on factors
such as time horizon, spatial resolution, risk tolerance, and climate variables of interest. It
focuses on the process of selecting and obtaining raw data sets from online climate information
sources, rather than obtaining climate impact analysis and synthesis products. While this report
does not provide technical guidance on the use of raw climate data in vulnerability and impact
assessments, it includes references to sources of information that may be useful to that end.
Figure 1 shows the overall steps in the scenario selection process.
11
-------
Identify Climate Information Needs
Confirm Climate Projection Needs (Section 2.3.1)
Align Climate Information with Decision Needs (Section 2.3.2)
Select Spatial Resolution
Select Scenarios
Select Climate Variables
Determine Temporal Resolution and Time Horizon
Select Projections using Selection Strategies (Section 3.3)
Obtain Data
Figure 1. Overview of approach for selecting climate projections to inform regional impact
assessments
This report is organized to allow the user to efficiently obtain the information needed to
complete the process for selecting climate projections. Section 2 describes the context, scope,
and need for the scenario selection process. This section also provides context for identifying
climate information needs. Section 3 outlines the steps and decisions involved in using the
LASSO tool. Sections 2 and 3 provide enough information for experienced analysts to
understand the approach and use the LASSO tool to help them select climate projections. Section
4 provides detailed background and additional reference information that may be useful for
readers seeking more technical information on climate projections or who want to dive deeper
into particular topics. The report indicates areas where readers may refer to Section 4 to further
their understanding and knowledge.
2.2 NEED AND RATIONALE FOR SCENARIO SELECTION
It has been demonstrated that identifying or ranking the "best" climate models based on
an ability to replicate historical climate does not result in a more confident prediction about
12
-------
future climate (Kunti et al., 2010; Pierce et al, 2009; Santer et al., 2009; Brekke et al., 2008;
Coquard, 2004). Instead, when using climate information for an analysis or assessment, an ideal
approach is to use information from all available climate models. Using all possible projections
allows one to consider the full range of plausible futures, which is consistent with approaches
seeking robust decisions in the face of uncertainty about the future environment (Weaver et al.,
2013). We lack the ability to confidently identify the model or models that provide the most
accurate climate projections decades into the future. By considering the largest possible number
of climate projections, analysts and decision makers can be better poised to identify "no regrets"
approaches to adaptation, and less susceptible to unexpected manifestations of a changing
climate ("no regrets" approaches in this context refer to strategies that will be effective in all
possible climate futures). See Section 4.3 and Section 4.4.2 for more information on relationships
among models and model evaluation.
In an ideal situation users would incorporate all available climate projections into their
work. However, they are likely to be quickly confronted by practical constraints that limit the
number of usable models for a given analysis or assessment. The most recent generation of
downscaled climate projections are stored in several hundred individual files and require dozens
of Terabytes (TB) of storage, in addition to the hundreds of hours needed to download the full set
of projections. Time and resources needed for other related tasks such as data processing and
communicating results also increase to likely infeasible levels if all available climate projections
are considered. However, the impracticality of using all climate projections does not reduce the
critical importance of considering as many projections as possible given a set of operational
constraints. Users will need to strike an appropriate balance between:
selecting more projections to better capture uncertainty and/or risk
selecting fewer projections in the face of practical resource and logistics constraints
The LASSO tool helps users resolve this dilemma by providing a systematic, transparent, and
logical process for selecting and, if desired, acquiring a subset of climate projections.
2.3 IDENTIFYING CLIMATE INFORMATION NEEDS
2.3.1 Climate Information in Decision Making
Analyses involving climate information are typically performed within a larger context of
a decision-making process or are intended to inform future decision-making. Thus, the
requirements and objectives of a specific management or policy decision should narrowly guide
the process of identifying and selecting relevant climate information (Johnson and Weaver,
2009). Keeping this principle in mind can help analysts ensure they are gathering appropriate
13
-------
data while also helping them focus on obtaining only those data that will be relevant to the
analysis.
The process of identifying and then incorporating climate information into an analysis
requires several steps. The analyst should begin by considering the context of the decision: what
are the underlying problems that the decision maker aims to address? The next step is to consider
the role of future climate conditions within that context, allowing the analyst to begin
considering what kind of information might be required (National Research Council, 2009). For
some analyses, quantitative information about the future may not be necessary. For other
applications, detailed high-resolution climate projections may be useful, and can serve as input to
other models (such as hydrological models) that can simulate impacts such as changes in flood
risk or infrastructure vulnerabilities (Kotamarthi et al., 2016; Moss et al., 2014).
Once an analyst understands the decision context, the next step in the process is to align
climate information and decision needs. By considering the key elements that must be
established in order to effectively inform a decision the analyst can ensure that inputs of climate
data and subsequent outcomes from the analysis are relevant to the decision. This alignment is
the focus of Section 2.3.2.
It is only when these initial steps of (i) establishing the decision context and (ii)
identifying the climate information that is most relevant within that decision context are
completed that one should move to the next step of obtaining climate information. Identifying
decision needs is a critical step before selecting climate information, so an analyst should be
prepared with this information before using the LASSO tool. This requires both identifying
sources of information and determining what specific subset of those information sources to
choose (Moss et al., 2014). This topic is the focus of Section 3, which outlines practical
strategies for selecting climate information using the LASSO tool.
2.3.2 Identifying Information Relevant to Decision Context
To ensure climate information will be useful to the needs of decision makers, analysts
will need to consider elements such as temporal resolution and the decision time horizon, spatial
resolution, the decision maker's tolerance for risk, and relevant climate variables. This section
provides an overview of each of these key elements and describes how they relate to the needs of
decision makers.
Climate variables: Identifying relevant climate variables is an important step in acquiring
climate information for decision making.
The information required for climate variables, such as precipitation and temperature,
depends strongly on the goals of the analysis and the assets or populations being studied. For
example, decisions on design specifications for some types of infrastructure may be vulnerable to
14
-------
changes in mean seasonal precipitation, while other types of infrastructure (e.g., culverts) may be
influenced by changes in extreme precipitation (Stainforth et al., 2007).
Many analyses require an understanding of future changes in the frequency, intensity, or
duration of extreme events such as heat waves or extreme rainfall events. Higher model
resolution does enable improved simulation of extreme events, although the accuracy of a
model's projections does not necessarily increase linearly with increases in resolution (Flato et
al., 2013).
The LASSO tool allows the user to select from a variety of precipitation and temperature
variables and time frames. Some analyses will require information derived from the temperature
and precipitation variables, such as streamflow or number of heatwaves; calculating derived
variables is beyond the scope of the LASSO tool.
Temporal resolution and time horizon: Information from climate models is available at
different time intervals (e.g., daily, monthly) andfor a variety of time horizons (e.g., mid-century,
end-of-century). The chosen temporal resolution and time horizon of the climate information
used in the analysis should be matched to the decision context and duration of influence of the
decision.
Information from climate models is often available at different temporal resolutions, from
hourly to daily, monthly, seasonal, annual, or decadal. Hourly or daily information may be
important for determining projections of extreme events (e.g., number of days above the
historical 95% maximum temperature). Given natural variability in the climate system that can
obscure climate trends when looking over short timescales, 30-year windows of climate model
outputs may be combined to provide long-term averages (e.g., average temperature change for
the period 2070-2099) as a best practice. In addition, to determine potential future change in
climate conditions, users may compare modeled future conditions with modeled conditions that
correspond to a historical baseline period (1981-2010 is a common historical baseline period to
capture recent historical change). The decision context, including potential impacts due to
thresholds in extreme events, should guide the choice of the temporal resolution and baseline
historical period of climate information included in an analysis.
The duration of a decision's influence (e.g., the useful lifetime of planned infrastructure)
and the frequency with which decisions need to be made are important factors to consider when
choosing climate information for analysis (SERDP, 2016; National Research Council, 2009). For
example, long-term infrastructure investments might require projections of relevant conditions
spanning several decades or longer, while farmers, fisheries managers, and emergency managers
would benefit from information on seasonal or inter-annual conditions (National Research
Council, 2009; IPCC-TGICA, 2007). Figure 2 illustrates the varying time horizons for a range of
15
-------
activities; EPA's LASSO tool allows users to select data by season and for a variety of time
horizons.
1 / / /
Jr> ..p it ^ /
// ////,
1 1
4? if J*
/ / / / / / / f
** / / / / / / / /
* at 4 r $ J? J;
i i i i r i i i
0 10 20
30 40 50 60 70 80 90 100
Time horizon (years from present)
Figure 2. Decision contexts and associated time horizons
Box indicates the time horizons most applicable to the use of climate model information. Source: Lu, 2011
Spatial resolution: Climate information is available at varying resolutions, but higher
resolution does not necessarily mean more accurate information.
Analysts working on local- or regional-scale assessments face the challenge of matching
climate information to the spatial scale of the decision (National Research Council, 2009;
National Research Council, 2010). Although climate projections typically focus on global or
continental scales, most decision contexts require information on local areas.
General Circulation Models (GCMs) are mathematical models that simulate the physics,
chemistry, and biology that influence the climate system (Walsh et al., 2014; Flato et al., 2013).
These models approximate processes at the spatial scale that the model can resolve based on a
combination of observations and scientific understanding (Walsh et al., 2014). Most GCMs
divide the world into grid cells of about 60 to 100 miles per side and cannot simulate fine-scale
changes at the regional level (Walsh et al., 2014), such as terrain (e.g., highly mountainous
regions) and coastal environments that can influence climate features at a small scale
(Kotamarthi et al., 2016). Because of this, a procedure known as downscaling is applied to
translate GCM projections into higher-resolution information that can be used as input to local or
regional impact analyses (Walsh et al., 2014); see Section 4.2 for more information on
downscaling.
EPA's LASSO tool allows users to select climate information by EPA Region and source
of statistically downscaled precipitation and temperature information. Several different sources
of statistically downscaled data exist, such as Bias Corrected Spatially Downscaled (BCSD),
Localized Constructed Analogs (LOCA), and Multivariate Adaptive Constructed Analogs
(MACA). Each of these datasets are downscaled information from the Coupled Model
Intercomparison Project (CMIP) 5 GCMs, and vary in the statistical techniques used, geographic
16
-------
coverage, and available variables. See Section 4 for additional information regarding statistical
downscaling.
Risk tolerance: A decision maker's risk tolerance may help determine which scenarios to
include and what level of uncertainty is acceptable.
Analysis to support effective climate-related decisions requires an assessment and
understanding of risk and risk tolerance (Moss et al., 2014; Snover et al. 2013). Analysis can
support risk management by using a range of climate scenarios and projections, informed by an
understanding of their inherent uncertainties and limitations (National Research Council, 2009).
As stated above, despite significant advances in our understanding of the climate system,
important aspects of future climate remain difficult or even impossible to predict, and projections
are thus subject to substantial uncertainties that must be accounted for in analyses. These include
the inherent unpredictability about future climate forcings; necessitating assumptions about
future conditions; variability within the climate system, on timescales of seasons to decades; and
imperfect understanding of the response of the climate system to future forcing, meaning that no
single climate model is able to provide "the answer" about future conditions (Hawkins and
Sutton, 2009; see Section 4 for a more in-depth discussion of these and related issues).
In practical terms, this means that decision makers will need to deal with a range of
future outcomes, as represented by the different projections from individual climate models. This
range may, in turn, imply a broad range in the severity of the specific impacts most relevant for a
given decision context, which the user must account for in the analysis, noting that even a range
of GCM projections may not capture the full range of potential future conditions (Snover et al.
2013). For decision makers who are risk-averse, information on low-probability but high-
consequence events, as well as scenarios of future climate that include "worst case" conditions,
may be essential to include in the analysis. Where decision makers have a greater tolerance for
risk, they may instead focus more attention on more moderate cases with a higher probability of
occurrence or, alternatively, a range of future climate scenarios that include both "best case" and
"worst case" outcomes to facilitate consideration of a wider range of potential solutions (SERDP,
2016). Selecting an even number of scenarios helps to avoid a common tendency to choose a
middle scenario under the false assumption that it is the most likely scenario (Snover et al. 2016).
Transparency and open acknowledgement of uncertainty are key to informed climate-
related decision making (National Research Council, 2009). While all projections of future
conditions have an inherent degree of uncertainty, non-specialists may not fully understand the
nature of uncertainty in climate projections. This may lead them to misperceive useful
information as too unreliable to support actionor conversely to place too much confidence in
projections (National Research Council, 2009). Uncertainty is inherent in nearly all decision
making and should not preclude analysis of potential climate impacts that can inform action
17
-------
(SERDP, 2016). The LASSO tool allows for a variety of selection strategies to accommodate a
range of tolerance to risk and uncertainty. It allows users to more easily harness the collective
wisdom of the existing suite of state-of-the-art climate models for their region, system, and
decision context.
18
-------
3. EPA'S LASSO TOOL: A PRACTICAL APPROACH TO SELECTING CLIMATE
PROJECTIONS
Scatterplots are helpful visual devices that aid in the scenario selection process
The LASSO tool can be used to apply selection strategies to scatterplots and obtain a
subset of raw climate data for use in analyses
3.1 PURPOSE OF EPA'S LASSO TOOL
As described above, the LASSO tool helps users select climate projections from groups
of models for use in decision making, impact analyses, and vulnerability assessments, given a
user's specific needs and decision context. LASSO represents an approach designed to be
sufficient for immediate needs and goals while not guaranteed to be optimal or perfect. The
output of the tool is a subset of raw climate projection data and figures based on the user-
specified study area, scenarios, and climate variables, as well as the chosen selection strategy.
LASSO addresses the deceivingly complex question "which climate projections should I use?"
by disaggregating the problem into discrete, logical steps.
3.2 SELECTING CLIMATE PROJECTIONS
The LASSO tool guides users through a process of six steps:
1. Define the study area - pick one of 10 EPA Regions or one of the lower 48
states or the District of Columbia
2. Select a data source - both the BCSD and LOCA datasets are available
3. Select a pathway - RCP 4.5 and RCP 8.5 appear most frequently in impact
studies
4. Climate variables - Two combinations of variable (temperature, precipitation),
season (annual, winter, spring, summer, fall), and time period (2021-2050, 2041-
2070, 2070-2099) must be selected to form the axes of the LASSO scatterplot
5. Selection strategies - One or more approaches for identifying a subset of climate
projections
6. Download your results - Users have the option to immediately download spatial
data, maps, and scatterplot graphics, or use an interactive scatterplot widget to
explore other data sources, scenarios, etc.
Viewing climate model projections in two dimensions provides an effective way to
evaluate their range and variety. Two-dimensional scatterplots are familiar to a broad,
interdisciplinary audience, are generally easy to interpret, and provide a simple, concise visual
reference as compared to other diagrams or figures representing three or more dimensions of
19
-------
information. Furthermore, downscaled climate projections frequently include only temperature
and precipitation measures1, in which case two-dimensional summaries are an appropriate
choice. Scatterplots lend themselves to quadrant-based groupings or typologies, such as "cool
and wet" or "hot and dry". These descriptive typologies are useful in that they not only capture
the hydrological gradient and distinct sets of impacts, but they are also intuitive and easily
communicated to a broad audience.
Selection Strategies
Save Interactive View
Selected Sceriuiio Files (/)
ฉf
ฉ'
ฉ0
o
Not Selected for Download
Emmissions Pathway: RCP 8.5
Study Arซซ: Region 3 (DC. Ufc. ML). PA, VA. WV|
Onto Source: |LOCA)
O
a*
0
y
c
0 5 10 15
Change in Mean Annual Temperature 2070-2099
Figure 3. Users of the LASSO tool have the option to customize results using
an interactive scatterplot to select climate projections
In this approach, the tool presents climate projections, based on the x- and y-axis
parameters chosen by the user, on a scatterplot. The user can then select a subset of these
projections for use (see Section 2.2) using one of the selection strategies discussed in Section 3.3.
3.3 SELECTION STRATEGIES IN THE LASSO TOOL
In developing a climate analysis for a location or project, an analyst may be confronted
with a large universe of future climate projections. For example, selecting the four RCPs and the
more than 40 downscaled CMIP5 climate models can result more than 150 unique projections.
The selection strategies discussed below help the user quickly and easily identify a subset
of projections, shown in scatterplots generated by the LASSO tool, that bound the range of a
larger group of climate projections in two dimensions (such as air temperature and precipitation)
simultaneously. At a high level, the selection strategies work by calculating and plotting change
statistics of all the climate models in a two-dimensional space, then selecting specific projections
based on their geometric position in the resulting scatterplot. This general approach allows for
1 There has been a recent trend of downscaling efforts that include several additional climate variables, such as
relative humidity and wind speed.
20
-------
the efficient selection of representative ensemble members that describes a range of possible
future change in a systematic way that is also logically desirable to the user.
Table 1 lists and compares the selection strategies (described in the sections below) that
can be used in the LASSO tool.
Table 1. Summary of scenario selection strategies
Strategy
Advantage
Disadvantage
Lasso
Captures the full envelope of potential
change described by the climate models
Typically results in -8-10 models;
time and resource requirements
potentially very high
Four Corners
Fewer models than the Lasso strategy
means less time and fewer resources; still
captures a broad range of potential futures
Could miss the minimum or
maximum of each axis
Middle Corners
Ignores projections that might be
considered outliers; captures a range of
values without a perceived focus on
extreme outcomes
Disregarding extreme projections may
confer some risk; gives the impression
that selected projections represent
more likely futures
Double Median
Lowest relative need of time and resources;
central projections of change may be useful
in some contexts
Easily misinterpreted as the "most
likely" or "best" scenario; information
about unexpected or extreme changes
is completely absent
21
-------
3.3.1 Lasso
The lasso strategy works by identifying the set of points in the LASSO scatterplot that
make up an imaginary "envelope" around all other points. This boundary polygon is also referred
to as the convex hull and, in the context of LASSO, can be used to capture the full range of
changes projected by a group of climate models. Similar approaches have been suggested or
even used by others to identify a subset of climate projections for impact studies (Cannon, 2015;
Salathe, 2007; Stainforth, 2007). This selection method will necessarily capture the minimum
and maximum values for each variable but will also include other projections that provide
additional information about potential combinations of change. Of the selection approaches
presented in this report, the Lasso strategy corresponds to the lowest risk tolerance, i.e., the
largest amount of information is included. A disadvantage to this strategy is that incorporating
the resulting information into an analysis may require a larger amount of time and resources. See
Figure 4. Black dots denote individual climate projections and red circles indicate those models
selected by the Lasso strategy.
Some temperature
More temperature
increase
increase
Figure 4. Illustrating the Lasso selection strategy. Black dots denote individual
climate projections. Red circles indicate models selected by the Lasso strategy.
22
-------
3.3.2 Four Corners
The Four Corners strategy captures a broad range of potential climate futures by
choosing a representative projection from each of four hypothetical quadrants. An imaginary
bounding box can be drawn around the scatterplot values and selecting the model that is closest
(in Euclidean distance) to each of the four corners of this box yields a subset that maximizes
differences among four projections. This approach has been used widely in climate studies to
identify a useful subset of climate change projections (e.g., Hosseinizadeh et al., 2015). This
strategy captures a limited number of projections compared to the Lasso approach, likely
reducing the amount of time and resources needed to process, analyze, or summarize the range of
information. However, this technique risks missing the minimum or maximum projections on
either axis, as demonstrated by the unselected black dots lying on the dotted line in Figure 5.
Some temperature
More temperature
increase
increase
Figure 5. Illustrating the Four Corners selection strategy. Black dots denote
individual climate projections. Red circles indicate models selected by the
Four Corners strategy. Dotted lines are drawn at the minimum and maximum
projected values for each axis.
23
-------
3.3.3 Middle Corners
The Middle Corners selection strategy is similar to the Four Corners approach in that the
goal is to identify a projection from each of four quadrants. However, the Middle Corners
strategy uses the 25th and 75th percentiles of each axis to identify the corners of an imaginary
box. The Bureau of Reclamation (2015) used a nearly identical approach to identify climate
projections for an assessment of risk in the western U.S. This strategy is less likely to include
model results that might be considered outliers, however this strategy will also disregard the
most extreme projections of change where exposure and vulnerability may reach levels of
concern. Note the relatively tight grouping of selected projections in Figure 6.
*.**
Some temperature
More temperature
increase
increase
Figure 6. Illustrating the Middle Corners selection strategy. Black dots
denote individual climate projections. Red circles indicate models selected
by the Middle Corners strategy. Dotted lines are drawn at the 25th and 75th
percentile projected values for each axis.
24
-------
3.3.4 Double Median
The Double Median strategy identifies a single projection by minimizing the Euclidean
distance from a point at the intersection of the median value of each axis. This approach is useful
when a central projection is needed, for example, to avoid the perception that results are only
representative of extremes. This central estimate may also be combined with other selection
strategies, such as Four Corners (Hosseinizadeh et al., 2015). Including a central estimate of
change may facilitate a path toward consensus or provide a useful benchmark when comparing
impacts under multiple scenarios of climate change. However, great care must be taken to avoid
suggestions that the Double Median strategy is the "best" projection or "most likely" outcome.
Note the wide range of potential future change that is not at all captured by the single, central
projection selected in Figure 7.
C
ฃ ซ,
ซ $
Q. ฃ
ฃ
median
I
%>*
I
Some temperature
More temperature
increase
increase
Figure 7. Illustrating the Double Median selection strategy. Black dots
denote individual climate projections. Red circles indicate the model
selected by the Double Median strategy. Dashed lines are drawn at the
median projected values for each axis.
3.4 OUTPUTS FROM THE LASSO TOOL
After users have selected climate scenario information, the LASSO tool provides the
ability to download raw climate projection data based on the user-specified study area, scenarios,
and climate variables. In addition, users can download static map images of the climate scenario
information for the specified study area. Refer to Section 5 for examples of LASSO outputs.
Interested users can find additional details regarding downloadable data within the tool;
download functions may change over time. As noted above, users should be aware that there are
uncertainties with any climate projections that cannot be fully accounted for in future climate
25
-------
impact studies. Approaches for dealing with these intractable uncertainties are beyond the scope
of this report.
3.5 CURRENT LIMITATIONS OF I II I PROCESS AND TOOL
The LASSO tool and the approaches to using it described above include some limitations
and caveats. Using the tool requires some familiarity with best practices in matching climate
information to decision needs, as well as an understanding of how to use climate information
accurately and appropriately. For example, a user might select a too-limited subset of data using
the LASSO tool, unknowingly introducing large uncertainty.
Examples of Related Climate-Data Tools
Several available tools provide access to climate data, although they may be difficult to apply in
decision-making. For example, they cannot be used to download geographic information system-
ready data, provide only limited guidance on what data to use, or present only spatially and
temporally constrained summaries.
U.S. Global Change Research Program Climate Explorer: Offers graphs, maps, and data of
observed and projected temperature, precipitation, and related climate variables for every county in
the contiguous United States. Web page: https://crt-climate-explorer.nemac.ore/.
U.S. Geological Survey National Climate Change Viewer: Includes the historical and future
climate projections from 30 of the downscaled models for two of the RCP emission scenarios (4.5 and
8.5). Allows users to visualize projected changes in climate (maximum and minimum air temperature
and precipitation) and the water balance (snow water equivalent, runoff, soil water storage and
evaporative deficit) for any state, county and United States Geological Survey Hydro logic Units. Web
page: https://www2.usgs.eov/landresoiirces/lcs/nccv.asp.
NCAR/GIS Program Climate Change Scenarios Data Portal: offers shapefiles, text files,
and images of climate change projections. Many 2D variables from modeled projected climate are
available for the atmosphere and land sector. Web page: http://gisclimatechan.ge.ucar.edu/.
US. Geological Survey Geo Data Portal: provides access to numerous datasets, including
gridded data for climate and land use. Web page: https://cida.usgs.gov/gdp/.
The LASSO tool provides basic functionality for selecting projections. The tool currently
uses a historical baseline of 1981-2010. Future versions of the tool may add additional
functionality and data, such as the ability of the user to directly select a custom group of
projections, access to additional downscaled climate model data archives, or additional climate
variables.
26
-------
4. BACKGROUND ON SCENARIOS, MODELS, AND SELECTION CRITERIA
Climate scenarios are used to explore a range of potential future climate conditions and
levels of impact
Downscaling techniques are usually applied in order to generate higher-resolution
information, which may be appropriate as an input to local or regional analyses
Decisions can more completely capture the range of possibilities by using results of
multiple models running multiple scenarios
Because of the large number of sources and types of climate projections available, using
selection criteria to narrow the number of projections can be helpful in simplifying the
selection process
This section contains additional information on scenarios, models, and selection criteria and
is intended for readers interested in more technical information and context.
4.1 CLIMATE SCENARIOS
Climate modelers use GCMs to project the Earth's future climate under a range of
scenarios. In most cases these scenarios were adopted by the Intergovernmental Panel on Climate
Change (IPCC) for the Fifth Assessment Report (2013), or an earlier set of scenarios from the
IPCC's Fourth Assessment Report (2007).
Scenarios represent a significant, but necessary, source of uncertainty and risk in climate
projections. If only a worst-case scenario is considered, there is a risk of incurring unnecessary
costs (e.g., through over-engineering). In contrast, assuming only an optimistic scenario runs the
risk of costly damages if future conditions turn out to be far less favorable. When looking at
relatively long-term climate conditions (i.e., end of 21st century), model selection and the choice
of scenario are the key sources of uncertainty (Hawkins and Sutton, 2009) and should reflect the
risk tolerance of the decision maker (SERDP, 2016).
4.2 DOWNSCALED CLIMATE INFORMATION
As discussed in Section 2.3, climate projections are generally produced at a relatively
coarse spatial resolution, whereas most decision contexts require highly localized climate
information. Downscaling can be applied to translate GCM projections into higher-resolution
information, which may be appropriate as an input to local or regional impact analyses (Walsh et
al., 2014). There are two types of models commonly used for downscaling: dynamical and
statistical. Both rely on inputs from GCMs.
Dynamical downscaling models, often referred to as regional climate models (RCMs)
(Walsh et al., 2014), use outputs from a GCM as boundary conditions to drive a separate higher-
resolution model over a limited spatial domain that better represents local or regional physical
processes (Kotamarthi et al., 2016; Flato et al., 2013). These models are very computationally
intensive.
27
-------
Statistical downscaling models use observed relationships between large-scale weather
features and local climate to statistically translate projections from GCMs down to a finer scale
(Walsh et al., 2014; Flato et al., 2013; Lu, 2011). Statistical downscaling models are best suited
for analyses requiring a range of future projections that reflect the uncertainty in scenarios and
climate sensitivity, at the scale of observations that may already be used for planning purposes
(Walsh etal., 2014).
Climate models do not perfectly simulate historical conditions and raw model output may
have systematic differences, or biases, between a simulated climate statistic and the
corresponding real-world climate statistic (Maraun 2016). For example, some models may have a
general bias toward warmer conditions than recorded in the historical observed record; other
models may generally show wetter conditions when compared with the historical data. To
address these biases, it is now a standard practice to "bias-correct" downscaled datasets for
impact modeling using techniques such as multiple linear regression, quantile mapping, or the
delta change approach (Maraun 2016) to better align with the observed conditions.
RCMs can directly simulate the response of regional climate processes to global change
and are not reliant on the statistical patterns from the past holding in the future, while statistical
models can better remove any biases in simulations relative to observations. Ideally, climate
impact studies could use both statistical and dynamical downscaling methods, but this coupled
approach is very resource-intensive (Walsh et al., 2014).
4.3 MODEL INTERDEPENDENCE AND PERFORMANCE
In addition to considering multiple scenarios, decisions can more completely capture the
range of possibilities by using results of multiple models running multiple scenarios. Using a
single model run is not considered scientifically rigorous because different GCMs often produce
different results, and there is no consensus that any one model is comprehensively better or more
accurate than others.
Models introduce additional sources of uncertainty (in addition to the uncertainties
related to future climate forcings), such as scientific uncertainty about the climate system and its
sensitivity (Kotamarthi et al., 2016). Natural climate variability, including the climate system's
inherent randomness, plays an especially important role in uncertainty over short timescales (10-
20 years) (SERDP, 2016; Walsh et al., 2014; Flato et al., 2013). For longer timescales (mid to
late century), using information from multiple climate models and scenarios can capture the
range of possible outcomes and uncertainties (SERDP, 2016; Flato et al., 2013). Considering the
full range of outputs from models, rather than their average or median values, provides a more
accurate representation of uncertainty, although low-probability, high-consequence future
conditions may still fall outside of the full model set.
28
-------
GCMs are frequently updated or otherwise revised using code that worked well in
previous iterations. Additionally, code for modules and routines is frequently shared within the
modeling community, allowing revisions to be completed with more community participation
while requiring less time for development and implementation. Specifically, many of the GCMs
used in CMIP5 are models that have been revised or updated from previous versions over the last
two decades. Many of these come from the same modeling center or share some of the same
underlying code. The strengths and weaknesses of particular models can thus be passed on to
newer model versions and to other models through the exchange of code and ideas.
The relationships among models is often hard to distinguish when models are titled
differently or appear from unrelated modeling centers. The use of related models can lead to an
unrealistically small spread in projections, resulting in a bias toward an artificial consensus in
model predictions. While excluding some models from an ensemble may be necessary given
technical or computational challenges, special attention should be paid to down-weighting
models that have very similar controls or a shared lineage in order to avoid this kind of
convergence bias.
Pennell and Reichler (2011) developed a statistical analysis to evaluate interdependences
among ensemble members in CMIP3. In Figure 8, the grey column indicates the effective
number of models based on the correlation of the error structure determined for each model.
Pennell and Reichler (2011) argue that this demonstrates that the CMIP3 group is "not a very
diverse ensemble" given the low number of independent models, even though the large number
of models in CMIP3 gives the opposite impression. Analysts should keep these
interdependencies in mind when selecting projections: a too-limited set of projections from
related models may introduce unexpected biases.
29
-------
GISSA
GISSR
GISSH
IPSL4
MRICM
ECHOG
INGV4
ECHM5
PCM11
CCSM3
FGOAL
C3T63
C3T47
CNRM3
BCM20
MIROM
MIROH
HADGM
HADCM
GFD20
GFD21-B
GFD21-A
INM30
CSR35
CSR30
0.1 0.3 0.5 0.7 0.8 0.9
Correlation
Figure 8. Effective number of models from CMIP3 based on statistical analysis of
effective degrees of freedom or effective sample size from a given dataset. The
correlation values are calculated based on model error structure.
Source: Pennell and Reichler, 2011
4.4 STRATEGIES FOR SELECTING PROJECTIONS
4.4.1 Overview of Selection Criteria
Because of the large number of sources and types of climate projections available, using
selection criteria to narrow the number of projections can be helpful in the selection process.
Climate model research groups and the CMIP process commonly address some of the criteria,
including:
Consistency with global projections: Projections should be consistent with a broad
range of global warming projections based on alternative scenarios of climate forcing
(IPCC-TGICA, 2007). The IPCC serves as a source of reference for the range of
global warming projections.
Physical plausibility: Projections should be physically plausible within the basic laws
of physics, and the combination of changes in different variables should be physically
consistent (IPCC-TGICA, 2007).
A variety of additional criteria are important to selecting projections, which are relevant to the
information needed by the user to implement the LASSO tool and supported by the LASSO
tool's practical approach. Potential additional selection criteria to consider include the following:
^ 1
Lr
HZ
-1
1
\
U,
1
H
1
"~l 1
1
30
-------
Applicability in impact assessments: Projections should describe changes in a
sufficient number of variables on a spatial and temporal scale that allows for impact
assessment. For example, impact models may require input data on variables such as
precipitation, humidity, and wind speed at spatial scales ranging from global to site-
specific and at temporal scales ranging from annual means to daily or hourly values
(IPCC-TGICA, 2007). Currently, the LASSO tool provides information only on
changes in mean annual temperature and changes in mean annual precipitation for the
United States. Users can view and download information at the EPA Region level.
Representativeness: Projections should be representative of the potential range of
future regional change in order to estimate a realistic range of possible impacts
(IPCC-TGICA, 2007). Applying the results of more than one GCM in an impact
assessment provides a range of representative results. GCMs can differ widely from
each other in their estimates of regional changes, especially for variables such as
precipitation, where some models may project wetter conditions in a region while
others project drier conditions. The LASSO tool facilitates selecting a practical set of
results from more than one GCM.
Accessibility: Scenarios used in projections should be straightforward to obtain,
interpret, and apply (IPCC-TGICA, 2007); the LASSO tool helps to directly support
accessibility.
Vintage: Recent model simulations are likely to be more reliable than those of an
earlier vintage. They are based on recent knowledge, incorporate more processes and
feedbacks, and usually have a higher spatial resolution than earlier models (IPCC,
2001). LASSO relies on recent model simulations (i.e., CMIP5).
Resolution: As climate models have evolved and computing power has grown, their
resolution has tended to increase. Some of the early GCMs operated on a horizontal
resolution of some 1,000 km with between 2 and 10 levels in the vertical. More recent
models run closer to a spatial resolution of 250 km with approximately 20 vertical
levels. Although higher-resolution models contain more spatial detail, this does not
necessarily guarantee superior performance (IPCC, 2001). The LASSO tool provides
access to statistically downscaled data at approximately 4-6 km grid cell resolution.
Validity: Analysists should use data from GCMs that simulate the present-day climate
most faithfully, on the premise that these GCMs should also yield the most reliable
representation of future climate. This approach involves comparing GCM simulations
that represent present-day conditions with the observed climate (IPCC, 2001). The
LASSO tool does not directly compare simulations with observed climate, but many
downscaled climate data sources are corrected for systematic biases.
4.4.2 Finding the "Best" Model(s)
Evaluations of climate model projections are based on the ability to faithfully simulate
the past; future skill is unknowable. There is thus no objective standard for what constitutes
31
-------
"good" or "good enough," because there is no way to know in advance how accurately any
model simulates future conditions. Furthermore, because the most appropriate climate inputs for
any given application depend on the nature of both risk tolerance and associated vulnerabilities,
there is no prescriptive guidance for the "best" climate models, methods, or projections to use in
any situation (Kotamarthi et al., 2016). Any determination of the best or most appropriate
model(s) would be specific to a time period, geographic area, and numerous other qualifications.
The process for selecting, obtaining, and incorporating climate data into decisions can be time-
consuming, and high levels of technical skill and knowledge are required. In the end, however,
conclusions drawn from climate projections are still subjective, reflecting the scope of the
information included in the analysis or assessment.
Mendilk and Gobiet (2015) describe a statistical, quantitative methodology to enable
selection of a representative set of models from the ensemble while still maintaining the essential
characteristics of the ensemble. First, an analysis is completed among the climate variables to
establish patterns of change within the multi-model ensemble. Second, a cluster analysis,
focusing on models with similar simulations, is performed to isolate these multivariate patterns.
Third, a sampling method is introduced to gather a single representative model from each cluster.
Mendilk and Gobiet's (2015) analysis was able to reduce an ensemble from 25 models to 5 for a
given example. However, such a methodology is time-intensive to implement and perform,
potentially making it impractical for a time- or resource-constrained analysis.
Together, these facts suggest that a heuristic (i.e., imperfect but "good enough") approach
may be preferable and certainly more practical than trying to identify a single or small group of
ideal models for the particular analysis at hand. LASSO offers this type of approach to problem-
solving: its outputs are sufficient for immediate needs and goals while not guaranteed to be
"optimal" or "perfect."
32
-------
5. LASSO SCATTERPLOTS FOR EPA REGIONS
The following scatterplots describe a range of potential climate change for EPA regions across a small, but generally
informative subset of possible climate variables and time horizons.
These scatterplots describe the average annual change in temperature on the horizonal axes (Fahrenheit) and change in
precipitation (%) on the vertical axes.
Values shown are the difference between the 2070-2099 future period and 1981-2010 historical period.
Model realizations that correspond to the 'Lasso' selection strategy are identified; other projections are shown as gray dots.
REGION 1
Annual
Summer (JJA)
Winter (DJF)
40
io
2 20
O
tr
40
o-
A
. ?o6
27/' 1
20 -
0 -
D
B
21
.. ?5,
26,
19,
?51
ฆ32i
18i :4,
17,
-I 1 1 1"
2!
271
38*1
ฆ26,
251
5*1
c
\206
26,
18,
i i
1V
27,
1
32,
i i i i i i i
F
CM
o
CM
. ' 25,
' .
19, * * *
27, ' -321
1 = ACCESS1-0
2 = ACCESS1-3
3 = bcc-csm1-1
4 = bcc-csm1-1-m
5 = CanESM2
6 = CCSM4
7 = CESM1-BGC
8 = CESM1-CAM5
9 = CMCC-CM
10 = CMCC-CMS
11 = CNRM-CM5
13 = EC-EARTH '
14 = FGOALSq2
17 = GFDL-CM3
18 = GFDL-ESM2G
19 = GFDL-ESM2M
20 = GISS-E2-H
22 = GISS-E2-R
24 = HadGEM2-AO
25 = HadGEM2-CC
26 = HadGEM2-ES
27 = inmcm4
28 = IPSL-CM5A-LR
29 = IPSL-CM5A-MR
31 = MIROC-ESM
32 = MIROC-ESM-CHEM
33 = MIROC5
34 = MPI-ESM-LR
35 = MPI-ESM-MR
36 = MRI-CGCM3
37 = NorESM1-M
10 12 14 16
10 12 14 16
6 8 10 12 14 16
Figure 9. LASSO scatterplots for EPA Region 1, excluding Puerto Rico and the Virgin Islands.
33
-------
REGION 2
Annual
Summer (JJA)
Winter (DJF)
40 H
30
20 -
CL
<-> in
Cฃ '0
0 -
-10 -
A
40 -
30 -
20 -
oo
CL
m J
Cฃ 10 -
0-
-10 -
19
;r
206
--26.
I 1 I-
B
D
2i
1^2 _* 1?^51
27," 28'j34,p2i
T 1 1-
10 12 14 16
33.
1ฎi ***,
17i
26
*4l 10,
i 1 i i i i r~
c
32 2Si
36,
ฆ> 206
34,
.
19,''
*
32,
/LI 1
-i 1 1 1 r-
E
ฆ
F 202
2,
V* #
*
19, t32,
....... j7,
^9,
2 3T
1 32, . 25i
-
1 = ACCESS1-0
2 = ACCESS1-3
3 = bcc-csm1-1
4 = bcc-csm1-1-m
6 = CCSM4
9 = CMCC-CM
10 = CMCC-CMS
17 = GFDL-CM3
18 = GFDL-ESM2G
19 = GFDL-ESM2M
20 = GISS-E2-H
22 = GISS-E2-R
24 = HadGEM2-AO
25 = HadGEM2-CC
26 = HadGEM2-ES
28 = IPSL-CM5A-LR
29 = IPSL-CM5A-MR
31 = MIROC-ESM
32 = MIROC-ESM-CHEM
33 = MIROC5
34 = MPI-ESM-LR
35 = MPI-ESM-MR
36 = MRI-CGCM3
37 = NorESM1-M
10 12 14 16
10 12 14 16
Figure 10. LASSO scatterplots for EPA Region 2.
34
-------
REGION 3
40 -
IT)
"tf
CL
O
Cd
LO
CO
CL
-20 ฆ
A
20 -
19,*t ,, .
,
2714;
Annual
17,
-20 -
26i
I 1 1 1-
40- Q
20 -
36, 35J ' 171
i it
25,
84,
Summer (JJA)
Winter (DJF)
B
-
c
17,
237,
1ง26 : 17,
36V*. '*
'4, 25,
-
36, 26,
34, * 6
19, S 8,
f7 *14
1 9,
E
F
36,. 25,
*
.17i
19a .14i
Vi *
27, .*
29,
24, *25i
-
10 12 14 16
10 12 14 16 0
Figure 11. LASSO scatterplots for EPA Region 3.
1 = ACCESS1-0
2 = ACCESS 1-3
3 = bcc-csm1-1
4 = bcc-csm1-1-m
6 = CCSM4
7 = CESM1-BGC
8 = CESM1-CAM5
9 = CMCC-CM
10 = CMCC-CMS
17 = GFDL-CM3
18 = GFDL-ESM2G
19 = GFDL-ESM2M
20 = GISS-E2-H
22 = GISS-E2-R
24 = HadGEM2-AO
25 = HadGEM2-CC
26 = HadGEM2-ES
28 = IPSL-CM5A-LR
29 = IPSL-CM5A-MR
31 = MIROC-ESM
32 = MIROC-ESM-CHEM
33 = MIROC5
34 = MPI-ESM-LR
35 = MPI-ESM-MR
36 = MRI-CGCM3
37 = NorESM1-M
10 12 14 16
35
-------
REGION 4
Annual
30
20 -
10
0-
Summer (JJA)
LO
"fr
CL
O
* -10
-20
-30 H
A
1 17,
v* *24
4, . 1
~l 1 1 1 T T l~
30
CL
O
-10
B
17,
22 s
V-
\ -1
i i 1 1 1 1 i
10 12 14
Winter (DJF)
34,'
s.,7?
18,-
r,
8-
29-
~1 1 1 I 1 I I
^ 35,
_
r^~
LU
36 f -a \171
-
222 *
27/,'.. 25,
-
0"7
"
19,/ 32,
-
_
3&r- 25,
3ป2
222 : 25,
36*
18*1
29,
10 12 14
Figure 12. LASSO scatterplots for EPA Region 4.
1 = accessi-0
2 = ACCESS 1-3
3 = bcc-csm1-1
4 = bcc-csm1-1-m
6 = CCSM4
7 = CESM1-BGC
8 = CESM1-CAM5
9 = CMCC-CM
10 = CMCC-CMS
17 = GFDL-CM3
18 = GFDL-ESM2G
19 = GFDL-ESM2M
20 = GISS-E2-H
22 = GISS-E2-R
24 = HadGEM2-AO
25 = HadGEM2-CC
26 = HadGEM2-ES
28 = IPSL-CM5A-LR
29 = IPSL-CM5A-MR
31 = MIROC-ESM
32 = MIROC-ESM-CHEM
33 = MIROC5
34 = MPI-ESM-LR
35 = MPI-ESM-MR
36 = MRI-CGCM3
37 = NorESM1-M
10 12 14 16
36
-------
5.5 REGION 5
Annual
A
40 ฆ
in 20 -
0.
a OH
-20 ฆ
-40
20 K
22g
19J/%, .32,
*1,24,
I 1 1 1 1 1 1 1-
D
40 -
in 20 H
00
Q_
O
0C 0 -
-20 -
-40 ฆ
22, Jฐ1
29,
19,
2^17,
32,
25,
-1 1 1 1 1 1 1 1-
Summer (JJA)
-i 1 1 1 1 1 1 r
Winter (DJF)
B
22 6
17,
25,
1i
c
17 25i
1 8
36,
#V1
19,
' -321
6e
29,
-| 1 1 1 1 1 1 r
i 1 1 1 1 1 1 |
~i 1 1 1 1 1 1 r-
E
F
-
36, . .. .*
9i
.
is? . 32i
2?? 17i
. *
29,
1, 25,
1 = ACCESS1-0
2 = ACCESS 1-3
3 = bcc-csm1-1
4 = bcc-csm1-1-m
6 = CCSM4
7 = CESM1-BGC
8 = CESM1-CAM5
9 = CMCC-CM
10 = CMCC-CMS
17 = GFDL-CM3
18 = GFDL-ESM2G
19 = GFDL-ESM2M
20 = GISS-E2-H
22 = GISS-E2-R
24 = HadGEM2-AO
25 = HadGEM2-CC
26 = HadGEM2-ES
28
29
34 =
35 =
IPSL-CM5A-LR
IPSL-CM5A-MR
MIROC-ESM
MIROC-ESM-CH
MIROC5
MPI-ESM-LR
MPI-ESM-MR
36 = MRI-CGCM3
37 = NorESM1-M
10 12 14 16 18
10 12 14 16 18 0 2 4
10 12 14 16 18
Figure 13. LASSO scatterplots for EPA Region 5.
37
-------
REGION 6
Annual
20 -
o-
in
ti-
ll.
O
a: -20 H
-40 -
A
21
ai 1
20 6
Oy
LO
CO
Q_
O
2^
i i 1 1 i r~
Summer (JJA)
B
2,33
31,
171
22 s *
24 -i
Winter (DJF)
~i t 1 1 i r~
320*
&$121
26,
r
181 ' ***241
281 29,
-i 1 1 1 1 1
20 ฆ
D
"
E
F 35,
0-
;,s
2V-
132 8,
; * .<
321
"
22 2
. .25,
-20 ฆ
"
y
* OJ
CN
CNJ
17-i
is;
-40 ฆ
29,
3,
291
"
28 29,
1 = ACCESS1-0
2 = ACCESS1-3
3 = bcc-csm1-1
4 = bcc-csm1-1-m
6 = CCSM4
9 = CMCC-CM
10 = CMCC-CMS
17 = GFDL-CM3
18 = GFDL-ESM2G
19 = GFDL-ESM2M
20 = GISS-E2-H
22 = GISS-E2-R
24 = HadGEM2-AO
25 = HadGEM2-CC
26 = HadGEM2-ES
28 = IPSL-CM5A-LR
29 = IPSL-CM5A-MR
31 = MIROC-ESM
32 = MIROC-ESM-CHE
34 = MPI-ESM-LR
35 = MPI-ESM-MR
36 = MRI-CGCM3
37 = NorESM1-M
10 12
14
10 12
14
10 12
14
Figure 14. LASSO scatterplots for EPA Region 6.
38
-------
REGION 7
Annual
60 -1
40 -
20 -
0 -
-20 -
-40 -
60
40 -
lo 20 "
CO
Q.
ฐ 0-1
ce u i
-20
-40
A
n, 20 6
'Z"%
6 1, '
D
r 1 i i r T"
36, ,10|7
32,
4 * ;.251
291
I 1 1 1 1 1 1-
Summer (JJA)
Winter (DJF)
B
c
19, fn. 171
'24,
-
17 A 25,
3V' 241
6*6
1i
.
E
F 35,
36,
' 17,
222. i
' S1,
4,
24i25,
-
36, . '.t ' 24,
22^
202
3i'2&;,
10 12 14 16 0
I 1 1 1 1 1 1-
i 1 1 1 1 1 r~
10 12 14 16 0
Figure 15. LASSO scatterplots for EPA Region 7.
1 = ACCESS1-0
2 = ACCESS 1-3
3 = bcc-csm1-1
4 = bcc-csm1-1-m
6 = CCSM4
7 = CESM1-BGC
8 = CESM1-CAM5
9 = CMCC-CM
10 = CMCC-CMS
17 = GFDL-CM3
18 = GFDL-ESM2G
19 = GFDL-ESM2M
20 = GISS-E2-H
22 = GISS-E2-R
24 = HadGEM2-AO
25 = HadGEM2-CC
26 = HadGEM2-ES
27 = inmcm4
28 = IPSL-CM5A-LR
29 = IPSL-CM5A-MR
31 = MIROC-ESM
32 = MIROC-ESM-CHEM
33 = MIROC5
34 = MPI-ESM-LR
35 = MPI-ESM4V1R
36 = MRI-CGCM3
37 = NorESM1-M
10 12 14 16
39
-------
5.8 REGION 8
40 -
20
Annual
A
^ 0 -
"t
CL
-40 -
40 ฆ
20 ฆ
^ 0 -
CO
CL
o
^ -20 ฆ
-40 -
D
si"
4;" *. 24,
t 1 1 1 r t~
35,; :
22-
: ,32,
. 25,
'24,
Summer (JJA)
B
19v !:.1 17^
0 2
10 12 14 16 0
24,
I I T T 1 I 1
10,
19i ฆ; . 17i
*' 32,
22
2 . *
24,
Winter (DJF)
36,
22
10 12 14 16 0
I,
~i r 1 1 1 1 1
36,
22,
3^,
-33,
Figure 16. LASSO scatterplots for EPA Region 8.
1 = accessi-0
2 = ACCESS 1-3
3 = bcc-csm1-1
4 = bcc-csm1-1-m
6 = CCSM4
7 = CESM1-BGC
8 = CESM1-CAM5
9 = CMCC-CM
10 = CMCC-CMS
17 = GFDL-CM3
18 = GFDL-ESM2G
19 = GFDL-ESM2M
20 = GISS-E2-H
22 = GISS-E2-R
24 = HadGEM2-AO
25 = HadGEM2-CC
26 = HadGEM2-ES
28
29
34 =
35 =
= IPSL-CM5A-LR
= IPSL-CM5A-MR
= MIROC-ESM
= MIROC-ESM-CHEM
= MIROC5
MPI-ESM-LR
MPI-ESM-MR
36 = MRI-CGCM3
37 = NorESM1-M
10 12 14 16
40
-------
5.9 REGION 9
Annual
160
120 -
80
in
O 40-|
DC
0-
-40 -
Summer (JJA)
Winter (DJF)
A
51
160
120
80
iq
CO
O 40 -|
cc
0 -
-40 -
D
i 1 i
36, .
22|a*i
B
3^1
32,
31,
181
i.
24,17,
22r
10 12 14
~i 1 1 1 1 1 r
10 12 14
36, 29,
ปs.24
*1%33^"
31,
-i 1 1 1 1 1 r-
E
37,
: 32,
36,*. .
^0^9,' 17,
LL
36,
. -29,
22tQ2
X * *
' % 3??,
1 = ACCESS1-0
2 = ACCESS 1-3
3 = bcc-csm1-1
4 = bcc-csm1-1-m
6 = CCSM4
7 = CESM1-BGC
8 = CESM1-CAM5
9 = CMCC-CM
10 = CMCC-CMS
17 = GFDL-CM3
18 = GFDL-ESM2G
19 = GFDL-ESM2M
20 = GISS-E2-H
22 = GISS-E2-R
24 = HadGEM2-AO
25 = HadGEM2-CC
26 = HadGEM2-ES
28
29
34 =
35 =
IPSL-CM5A-LR
IPSL-CM5A-MR
MIROC-ESM
MIROC-ESM-Ch
MIROC5
MPI-ESM-LR
MPI-ESM-MR
36 = MRI-CGCM3
37 = NorESM1-M
EM
10 12 14
Figure 17. LASSO scatterplots for EPA Region 9, excluding Hawaii and island territories.
41
-------
5.10 REGION 10
Annual
Summer (JJA)
Winter (DJF)
^r
Q_
$ -20 H
-40 -
LO
CO
Q_
g -20 H
-40 -
-60
1 = ACCESS1-0
2 = ACCESS1-3
3 = bcc-csm1-1
4 = bcc-csm1-1-m
6 = CCSM4
9 = CMCC-CM
10 = CMCC-CMS
17 = GFDL-CM3
18 = GFDL-ESM2G
19 = GFDL-ESM2M
20 = GISS-E2-H
22 = GISS-E2-R
24 = HadGEM2-AO
25 = HadGEM2-CC
26 = HadGEM2-ES
28 = IPSL-CM5A-LR
29 = IPSL-CM5A-MR
31 - MIROC-ESM
32 = MIROC-ESM-CHEM
33 = MIROC5
34 = MPI-ESM-LR
35 = MPI-ESM-MR
36 = MRI-CGCM3
37 = NorESM1-M
2 4 6 8 10 12 14 16 0 2 4 6 8 10 12 14 16 0 2 4 6 8 10 12 14 16
Figure 18. LASSO scatterplots for EPA Region 10, excluding Alaska.
42
-------
6. CONCLUSION
Locating and identifying useful projections of future climate can be a laborious and
technically challenging task. Users must seek to balance using the maximum number of models
possible to capture inherent uncertainty in projections of future climate against the practical
constraints of the analysis environment. The scatterplot scenario selection process, which uses
heuristic approaches such as the Lasso, Four Corners, Middle Corners, and Double Median
facilitated by the LASSO tool, streamlines the challenging process of selecting a subset of
relevant data for decision needs (Bureau of Reclamation, 2015), although it does have tradeoffs.
For example, a user might select a too-limited subset of data, unknowingly introducing large
uncertainties. Best practices in matching climate information to decision needs and appropriate
use of climate information are key to effective use of the LASSO tool.
Setting out to select climate information, users should consider the climate information
needed to support decision-making, and employ the approach best suited to the decision context
and analytical constraints. The example maps provided within this report provide a readily
accessible starting point for those working in EPA Regions across the country.
43
-------
REFERENCES
Brekke, L; Dettinger, M; Maurer, E; Anderson, M (2008) Significance of model credibility in estimating climate
projection distributions for regional hydroclimatological risk assessments. Climatic Change 89(3-4): 371-394.doi:
10.1007/sl0584-007-9388-3.
Bureau of Reclamation (2015) West-Wide Climate Risk Assessments: Irrigation Demand and Reservoir Evaporation
Projections. Technical Memorandum No. 68-68210-2014-01. Bureau of Reclamation Technical Service Center,
Water Resources Planning and Operations Support Group, Denver, Colorado Desert Research Institute, Division of
Hydrologic Sciences, Reno, Nevada
Cannon, A. 2015. Selecting GCM Scenarios that Span the Range of Changes in a Multimodel Ensemble:
Application to CMIP5 Climate Extremes Indices. Journal of Climate, 28(3), 1260-1267.
Coquard J; Duffy PB; Taylor KE; Iorio JP (2004) Present and future surface climate in the western USA as
simulated by 15 global climate models. Climate Dynamics 23 455-72.
Intergovernmental Panel on Climate Change (IPCC) (2001) Climate Change 2001: Impacts, Adaptation &
Vulnerability.
IPCC (2014) Climate Change 2014: Synthesis Report. Contribution of Working Groups I, II and III to the Fifth
Assessment Report of the Intergovernmental Panel on Climate Change. IPCC, Geneva, Switzerland.
IPCC-TGICA (Intergovernmental Panel on Climate Change-Task Group on Scenarios for Climate and Impact
Assessment) (2007) General Guidelines on the Use of Scenario Data for Climate Impact and Adaptation
Assessment. Version 2. Prepared by T.R. Carter on behalf of the Intergovernmental Panel on Climate Change, Task
Group on Data and Scenario Support for Impact and Climate Assessment.
Flato, G; Marotzke, J; Abiodun, B; Braconnot, P; Chou, SC; Collins, W; Cox, P; Driouech, F; Emori, S; Eyring, V;
Forest, C; Gleckler, P; Guilyardi, E; Jakob, C; Kattsov, V; Reason, C; Rummukainen, M (2013) Evaluation of
Climate Models. In: Climate Change 2013: The Physical Science Basis. Contribution of Working Group I to the
Fifth Assessment Report of the Intergovernmental Panel on Climate Change. Cambridge University Press,
Cambridge, United Kingdom and New York, NY, USA.
Hawkins E; Sutton, R (2009) The potential to narrow uncertainty in regional climate predictions. Bull Am Meteorol
Soc, 90:1095-1107.
Hosseinzadeh, A; SeyedKaboli, H; Zareie, H; Akhondali, A; Faijad, B (2015) Impact of climate change on the
severity, duration, and frequency of drought in a semi-arid agricultural basin. Geoenvironmental Disasters 2:23.
Johnson, T; Weaver, C (2009) A Framework for Assessing Climate Change Impacts on Water and Watershed
Systems. Environmental Management 43:118-134.
Knutti, R; Furrer, R; Tebaldi, C; Cemak, J; Meehl, G (2010) Challenges in combining projections from multiple
climate models. Journal of Climate, 23(10): 2739-2758. doi: 10.1175/2009JCLI3361.1
Kotamarthi, R; Mearns, L; Hayhoe, K; Castro, CL; Wuebble, D (2016) Use of Climate Information for Decision-
Making and Impacts Research: State of Our Understanding. Prepared for the Department of Defense, Strategic
Environmental Research and Development Program.
Lu, X (2011) Applying Climate Information for Adaptation and Decision-Making: A Guidance and Resource
Document. National Communications Support Programme.
Moss, R; Scarlett, PL; Kenney, MA; Kunreuther, H; Lempert, R; Manning, J; Williams, BK; Boyd, JW; Cloyd, ET;
Kaatz, L; Patton, L (2014) Ch. 26: Decision Support: Connecting Science, Risk Perception, and Decisions. Climate
44
-------
Change Impacts in the United States: The Third National Climate Assessment, Melillo, JM; Richmond, TC; Yohe,
GW, Eds., U.S. Global Change Research Program, 620-647. doi:10.7930/J0H12ZXG.
National Research Council (2009) Informing Decisions in a Changing Climate. Panel on Strategies and Methods for
Climate-Related Decision Support, Committee on the Human Dimensions of Global Change. Division of Behavioral
and Social Sciences and Education.
National Research Council (2010) Informing an Effective Response to Climate Change. America's Climate Choices:
Panel on Informing Effective Decisions and Actions Related to Climate Change, Board on Atmospheric Sciences
and Climate. Division on Earth and Life Studies.
Pennell, C; Reichler, T (2011) On the Effective Number of Climate Models. Journal of Climate, 24(9), 2358-2367.
doi: 10.1175/2010jcli3814.1
Pierce, D; Barnett, T; Santer, B; Gleckler, P (2009) Selecting global climate models for regional climate change
studies. Proceedings of the National Academy of Sciences 106(21): 8441-8446.
Salathe, E.P., P.W. Mote, and M.W. Wiley. 2007. Review of scenario selection and downscaling methods for the
assessment of climate change impacts on hydrology in the United States pacific northwest. International Journal of
Climatology, 27(12), 1611-1621.
Santer, B; Taylor, K; Gleckler, P; Bonfils, C; Barnett, T; Pierce, D; Wigley, T; Mears, C; Wentz, F; Bruggemann,
W; Gillet, N; Klein, S; Soloman, S; Stott, P; Wehner, M (2009) Incorporating model quality information in climate
change detection and attribution studies. Proceedings of the National Academy of Sciences 106(35): 14778-14783.
Snover, A.K., Mantua, N.J., Littell, J.S., Alexander, M.A., McClure, M., Nye, J. 2013. Choosing and using climate
change scenarios for ecological impacts assessments and conservation decisions. Conservation Biology 27(6): 1147-
1157, doi: 10.1111/cobi.l2163Stainforth, DA; Downing, TE; Washington, R; Lopez, A; New, M (2007) Issues in
the interpretation of climate model ensembles to inform decisions. Phil. Trans. R. Soc. doi: 10.1098/rsta.2007.2073
Strategic Environmental Research and Development Program (SERDP) (2016) Climate-Sensitive Decision-Making
in the Department of Defense: Synthesis of Ongoing Research and Current Recommendations. US Department of
Defense.
Walsh, J; Wuebbles, D; Hayhoe, K; Kossin, J; Kunkel, K; Stephens, G; Thorne, P; Vose, R; Wehner, M; Willis, J;
Anderson, D; Kharin, V; Knutson, T; Landerer, F; Lenton, T; Kennedy, J; Somerville, R (2014) Appendix 3:
Climate Science Supplement. Climate Change Impacts in the United States: The Third National Climate
Assessment, Melillo, JM; Richmond, TC; Yohe, GW; Eds., U.S. Global Change Research Program, 735-789.
doi: 10.7930/J0KS6PHH.
Weaver, C. P., R. J. Lempert, C. Brown, J. A. Hall, D. Revell, and D. Sarewitz, 2013. Improving the contribution of
climate model information to decision making: The value and demands of robust decisions frameworks. Wiley
Interdisciplinary Reviews: Climate Change 2013 4:39-60.
45
-------
A Systematic Approach for Selecting Climate Projections
to Inform Regional Impact Assessments
------- |