Regional Climate Change Effects: Useful Information for Transportation Agencies
2 Methodology Overview
The projections of climate change effects presented in this report were developed through a systematic process initiated by FHWA in the summer of 2009. This section describes the key elements of FHWA's methodology; Appendix A provides more details on the approach, which provides climate change information by U.S. region, by time horizon, and by climate variable. The regions are identical to those used in the U.S. Global Change Research Program (USGCRP) climate impact analyses (USGCRP 2000, 2009). Three time horizons were chosen for each region: near-term (2010-2040), mid-century (2040-2070), and end-of-century (2070-2100). Projected changes in climate are reported for temperature, precipitation, storm events, and sea-level rise.
As described below, initial research efforts attempted to capture regional or sub-regional projections from all publicly available, peer-reviewed studies for these climate effects. During the course of this research, FHWA consulted with a range of nationally recognized climate scientists for their insights and recommendations regarding the most credible regional projections for use by State DOTs and local transportation agencies. As a result of these conversations, FHWA gained access to key data sets not previously published in their entirety, including data compiled from the CMIP3 database that underlies the USGCRP's Global Climate Change Impacts in the United States (2009) report. The tables of regional climate changes provided in this report were derived from an analysis of that data set. That information is complemented with the results of an extensive literature search and review, which is summarized in Appendix C. The search was conducted using a variety of databases of journal articles, government reports, and other peer-reviewed publications encompassing a variety of spatial scale information from regional to city-scale.
The conclusions and data sets revealed through these efforts were further evaluated and scrutinized, and subsequently vetted with regional climate experts. A methodology was developed that identified which studies in the Climate Change Effects Typology Matrix correlated with optimum model characteristics, and these studies were then included in the regional narrative in the main report.
It should be noted that each study cited in this report has a unique set of model characteristics and associated uncertainty6, which makes comparing results across studies challenging. Although all of the differences in assumptions and approaches among these studies are not explicitly described in this report, care was taken to ensure that only logically comparable aspects of the studies are presented.
FHWA's methodology for identifying, categorizing, reviewing, and presenting projections of climate change effects in this report involved the following steps:
- Selection of relevant climate effects: The climate change effects determined to most affect highways and highway networks and discussed in this report include changes in average and extreme temperature, changes in average and extreme precipitation, and sea-level rise. Several recent reports have highlighted the importance of these effects with respect to the U.S. transportation system and highways in particular (NRC 2008; USGCRP 2009; CSIRO 2006; CIG 2007). Extreme precipitation is associated with storm activity such as convective storms, extratropical storms, or tropical storms. There is limited information available pertaining to storm activity projections that describes changes in storm intensity, frequency, and duration. Additional variables such as relative humidity, solar radiation, and extreme cold events are also relevant, but regional information for these variables was not available for inclusion in this report.
- Literature review: The compilation of regional projections began with a literature search using relevant, carefully composed search terms across relevant databases of publications on the environment, energy, technology, and government.
- Screening literature review findings: The search was refined to include articles, government reports, and peer-reviewed publications with a published date post-2003 and available by June 2009. Effort was made to include seminal reports that became available after this date. This approach was taken to ensure that the versions of the climate models used were less likely to draw from out-of-date parameterizations, and that the emission scenario projections used were more likely to be based on the IPCC SRES scenarios used in the Fourth Assessment Report.
- Populating the Climate Change Effects Typology Matrix: The literature was organized into the Climate Change Effects Typology Matrix (located in Appendix C) by U.S. region based on a recent panel-reviewed report (USGCRP 2009); by time horizon where near-term represents 2010-2040, mid-century represents 2040-2070, and end-of-century represents 2070-2100; by climate effects (listed in the order of temperature, precipitation, storm activity, and sea-level rise); and by spatial coverage.
- Consultation with federal climate experts: A group of national federal climate experts from organizations including NOAA's National Climatic Data Center, USGS, DOE and others provided guidance on the criteria used in this report for determining whether a study would be included in the Section 3 regional narratives. There was a strong consensus for providing a plausible range of projections (tied to the low B1 and moderately high A2 IPCC greenhouse gas emissions scenarios) as opposed to a single point value for the mid-century and end-of-century projections. Projections of seasonal and annual temperature and seasonal precipitation compiled from CMIP3 database for use in the 2009 USGCRP report were used together with the results of the literature review (see Appendix A for more information on the emissions scenarios).
- Additional data analysis and refinements to regional climate change data: This report includes new analyses of the CMIP3 database of climate model integrations compiled by Michael Wehner of the Lawrence Berkeley National Laboratory and used in the USGCRP (2009) report7
- . Regional values are determined by using the corresponding grid cells of each climate model that fall within each region (Figure 6, for example, was developed using this collection of projected data). Then for each region and each of the three time frames, the following statistics have been computed for temperature and precipitation: "mean," "likely," and "very likely" (see below for a definition of terms). The results are included in regional tables in Chapter 3, the regional maps in Appendix B, and the Climate Change Effects Typology Matrix (Appendix C). These results provide mean conditions (as opposed to variability) at the regional scale.
- Inclusion of downscaled data: High-resolution temperature and precipitation projections for the continental United States developed through statistical downscaling of the results of 16 climate models of the CMIP3 database were provided to FHWA (Liang et al. 1994; Maurer et al. 2002)8
- . The projections are provided for a low (B1) and moderately high (A2) emission scenarios for three future projections including near-term (2010-2039), mid-century (2040-2069), and end-of-century (2070-2099) relative to a 1971-2000 baseline. Figures of the temperature projections are provided in Appendix B, while figures of thresholds such as extreme temperature are provided within this report (Maurer et al. 2002; Maurer et al. 2008). Downscaled data may be preferred when projections are required for an area finer than the spatial resolution provided by the climate models; particularly if the location is not well-represented by the larger-scale averages. In addition, downscaling data provides a mechanism for translating larger-temporally scaled climate model projections to finer-temporally scaled climate variability.
- Consultation with regional climate experts: The Climate Change Effects Typology Matrix was vetted with regional climate experts to discuss studies included and identify any missing studies. Based on this review, some studies were removed from the matrix or placed into the national section. The consultation also included discussions of particular regional problematic climate effects. There was a general consensus that it is important to not "cherry pick" climate models for calculating the mean and ranges from the USGCRP data. Alaska is the exception where the general consensus was to draw results from the five top performers identified by the Walsh et al. (2008) study.
- Treatment of uncertainty: There is always some degree of uncertainty associated with model projections. The Climate Change Effects Typology Matrix includes this information, when included in the source study, in a column labeled "certainty." In general, the model projections are more uncertain the further they are into the future. A small range of uncertainty tends to exist in the near-term time horizon, while a larger range of plausible values exists for the long-term. The temperature and precipitation information from the USGCRP data set that are presented for each region were quantitatively analyzed in this study to characterize plausible future climate conditions and the associated uncertainty. Each scenario/model combination produces a single data point. There are 15 models run for the A2 emission scenario to 19 models run for the B1 emission scenario, producing 15 to 19 mean results for each variable in each time frame. The following information is provided for precipitation and temperature for each region (see Figure 2):
- "Mean" - The mean range is the average of all of the simulations in the lower emission scenario (B1) and the average of all of the simulations in the higher emission scenario (A2). It is a simple measure of the central tendency of the projections and the uncertainty associated with future greenhouse gas (GHG) emission rates.
- "Likely" - The likely range is computed by first determining the standard deviation above and below the mean for each scenario.9 Then, the minimum and maximum of these four values (i.e., two from each scenario) are defined as the likely range. The range is a measure of the differences (and uncertainty) associated with the models that were used, as well as the uncertainty of future GHG emission rates.
- "Very Likely" - The very likely range is computed in the same way as the likely range, except that two standard deviations are used instead of one.10
Figure 2: This figure describes the process of creating the mean, likely, and very likely ranges from the USGCRP data and reported in Section 4. A curve represents an idealized version of connecting the points of results obtained from a variety of climate models. The blue curve and associated labels represent the mean, likely, and very likely values for a given emission scenario; likewise for the green curve assuming a Gaussian distribution. Each curve describes model uncertainty for the given emission scenario. The ranges provide the uncertainty associated with both the emission scenario and the climate models.
- Developing Regional Narratives for Section 3: A methodology was developed to synthesize the array of projections provided by the Climate Change Effects Typology Matrix into a regional narrative discussion that can assist in informing future analysis of climate impacts on the highway system. The criteria used in this report for determining whether a study would be included in the regional narratives of Section 3 were guided by discussions with climate experts and are illustrated in Figure 3.
Figure 3: Criteria used to assess whether a study is included in the narrative regional discussions.
6As described by Hawkins and Sutton (2009), model uncertainty, natural variability uncertainty, and scenario (i.e., GHG emissions scenario) uncertainty contribute to the total uncertainty associated with each projection. The structure of a modeling study helps define the associated uncertainty. For example, the magnitude of the uncertainty related to each of the three factors varies according to the time horizon of the projections. Hawkins and Sutton (2009) find that natural variability represents a large portion of the total uncertainty in applying climate projections in the near-term, dropping off significantly by mid-century. Model uncertainty is also a significant contributor to total near-term uncertainty and tends to stay relatively similar in magnitude through the projected century. For mid-century, the scenario and model uncertainties are somewhat similar in magnitude. For end-of-century, the scenario uncertainty contributes the greatest degree of uncertainty to the total. These uncertainties also change relative to each other when projections are provided at a smaller spatial scale (i.e., global to regional). For finer scale analysis, natural variability, in particular, significantly affects total uncertainty of a projection across all future time periods. back
7These values are averaged for each region from the corresponding grid cells of each climate model in the CMIP3 database. This is appropriate as the information provides mean conditions of temperature and precipitation at the regional scale. Statistical downscaling of the CMIP3 database for the continental United States is used in this report for the figures of extremes (such as the days the maximum temperature reaches or surpasses 90°F); the downscaling results are not available for locations outside the continental United States. See Appendix A for more information. back
8Bias-corrected and spatially downscaled climate projections derived from CMIP3 data, described by Maurer et al (2007). We acknowledge the Program for Climate Model Diagnosis and Intercomparison (PCMDI) and the WCRP's Working Group on Coupled Modelling (WGCM) for their roles in making available the WCRP CMIP3 multi-model dataset. Support of this dataset is provided by the Office of Science, U.S. Department of Energy. back
9Assuming the data are well represented by a Gaussian distribution, the likely range represents about 68% of the values extending from the 15th percentile to the 85th percentile. back
10Assuming the data are well represented by a Gaussian distribution, the very likely range represents about 95% of the values extending from the 2.2th percentile to the 97.8th percentile. back