Outline SUPPLEMENTARY INFORMATION DOI: 10.1038/NCLIMATE2631 Future population exposure to US heat extremes Jones, O Neill, McDaniel, McGinnis, Mearns & Tebaldi This Supplementary Information contains additional discussion, tables, and figures on five topics: Supplementary Discussion 1: The SRES A2 Scenario Supplementary Discussion 2: The NARCCAP ensemble Supplementary Discussion 3: U.S. Census Divisions Supplementary Discussion 4: GCM-RCM results Supplementary Discussion 5. Model uncertainty Supplementary Discussion 6. Projected exposure under alternative scenarios Supplementary Discussion 1. The SRES A2 Scenario The Special Report on Emissions Scenarios (SRES), published by the IPCC in 2000, included a set of emissions scenarios created to serve as a foundation for global change research [1]. Each scenario consists of a qualitative storyline describing broad-scale patterns of development and change over the 100 year period 2000-2100. These narratives, designed to represent consistent demographic, social, economic, technological, and environmental changes, are coupled with quantitative projections of important drivers such as population and GDP. Emissions models use these projections to drive future projections of emissions consistent with each scenario. NARCCAP was designed in part to focus on the uncertainty across different GCMs and RCMs, and thus uses only one emissions scenario, the SRES A2. The A2 was chosen since it was a common high emission scenario used to drive the CMIP3 global models. Since one of NARCCAP s goals was to provide climate scenarios for use by the impacts and adaptation communities, a high end scenario was chosen under the assumption that if one could adapt to a high scenario one could likely adapt to less extreme scenarios. As such we used the A2 scenario to guide the population projection in our analysis as well, including aggregate national-level population change, national urbanization rate, and local-patterns of spatial population change driven by parameter estimates in the population downscaling model [2]. The A2 storyline emphasizes heterogeneity. The primary themes are self-reliance and preservation of local identities. Fertility patterns across world regions converge slowly, resulting in relatively high population growth. Economic development is regionally oriented and per capita economic growth and technological change more fragmented and slower than other storylines. The A2 scenario is a high emissions scenario, and in general is associated with a larger increase in global surface temperatures. For a complete version of the A2 scenarios see Nakicenovic et al (2000). Supplementary Discussion 2. The NARCCAP Ensemble NARCCAP includes four GCMs that provide boundary conditions for six RCMs (see Supplementary Table 1). Funding limitations precluded filling the full 4 by 6 matrix of simulations, but a balanced fractional factorial statistical design was used such that each RCM used boundary conditions from two different GCMs, and each GCM provided boundary conditions to three different RCMs [3]. Only one set of the 12 sets of simulations (ECP2/HadCM3) was not completed. Thus, the ensemble in this work includes eleven different GCM-RCM combinations, illustrated in Supplementary Table 2. NATURE CLIMATE CHANGE www.nature.com/natureclimatechange 1
Supplementary Table 1. General circulation and regional climate models included in NARCCAP. Supplementary Table 2. NARCCAP GCM-RCM ensemble. Regional Climate Model CRCM ECP2 HRM3 MM5I RCM3 WRFG General Circulation Model CCSM X X X CGCM3 X X X GFDL X X X HadCM3 X X Supplementary Discussion 3. U.S. Census Divisions In this work we aggregate climate and population data from the national-level to that of the U.S. Census Division in order to carry out part of our analysis. We highlight divisional patterns in the shifting distributions of extreme heat, population, and exposure and illustrate how projected changes in climate and population contribute to each. Supplementary Figure 1 illustrates the nine US Census Divisions. S2
Supplementary Figure 1: The nine United States Census Divisions The following acronyms used to denote Census Divisions in Figure 2 and Figure 5: NE: New England MA: Mid-Atlantic ENC: East North Central WNC: West North Central SA: South Atlantic ESC: East South Central WSC West South Central MTN: Mountain PAC: Pacific Supplementary Discussion 4. GCM-RCM Results For each of the eleven GCM-RCM combinations we calculated the spatial distribution of the projected average annual number of days above 35 C for the period 2041-2070, the projected change in number of days above 35 C, the projected exposure in person days for the period 2041-2070, and the projected change in exposure. The model data was bias-corrected via kernel density distribution mapping [8] against the Maurer, et al. 2002 observational data set [9]. Here we present results for each member of the NARCCAP ensemble (Supplementary Figures 2 12). S3
Supplementary Figure 2. CRCM-CCSM results. a. b. Supplementary Figure 3. CRCM-CGCM3 results. S4
Supplementary Figure 4. ECP2-GFDL results. a. b. Supplementary Figure 5. HRM3-GFDL results. S5
Supplementary Figure 6. HRM3-HADCM3 results. a. b. Supplementary Figure 7. MM5I-CCSM results. S6
Supplementary Figure 8. MM5I-HADCM3 results. a. b. Supplementary Figure 9. RCM3-CGCM3 results. S7
Supplementary Figure 10. RCM3-GFDL results. a. b. Supplementary Figure 11. WRFG-CCSM results. S8
Supplementary Figure 12. WRFG-CGCM3 results. Supplementary Discussion 5. Model Uncertainty Between-model variation in climate outcomes produces distinct geographic patterns of model uncertainty in projected exposure. To assess these patterns we considered the cell-specific standard deviation and coefficient of variation. The former is illustrative of geographic variation in the absolute level of variation in exposure, while the latter is representative of the degree to which exposure varies relative to the ensemble mean. Supplementary Figure 13 depicts the standard deviation and coefficient of variation for both the projected change in days above 35 C and exposure side-by-side. We find that variation in the projected number of days above 35 C is largest in Central Texas, Florida, and Southern Georgia. Not surprisingly these are also areas projected to experience a large number of extreme heat days. Portions of the Central Plains, Desert Southwest (particularly Arizona), and Deep South (Louisiana) exhibit smaller standard deviations than surrounding areas despite a significant number of warm days, suggesting more agreement across models. By comparison, if we assess between-model variation in days above 35 C relative to the ensemble mean (the coefficient of variation) we find the most uncertainty in the Rocky Mountains, Pacific Northwest, and Appalachian Mountains, all areas in which the projected number of extreme heat days are relatively low. Variation in the projected change in exposure is, not surprisingly, heavily influenced by projected population change, and therefore tends to be highest in urban areas. However, variation in projected exposure relative to the ensemble mean follows a pattern similar to that of projected change in extreme heat days. Uncertainty is greatest in areas of the Rocky Mountains and Pacific Northwest, as well as parts of the Northern Plains. S9
Supplementary Figure 13.Standard deviation (a) and coefficient of variation (b) in projected change in days above 35 C, and standard deviation (c) and coefficient of variation (d) in exposure. Supplementary Discussion 6. Projected exposure under alternative scenarios Projected change in exposure from our constant population and constant climate scenarios are illustrated in Supplementary Figure 14. In the constant population scenario change in exposure results entirely from the projected increase in days above 35 C, while in the constant climate scenario change in exposure results from population change. In the case of the former the largest increases occur in the most densely populated cities, particularly those in the South which are projected to experience the largest number of extreme heat days. Significant increases are also projected in rural regions of the South, where substantial warming drives exposure up even in areas with smaller populations. In the latter scenarios significant population growth in cities of the Southeast, Texas, California, and the Desert Southwest drive increasing exposure. Stable population coupled with a small number of extreme heat days yields a smaller projected change in the large Northeastern cities. In rural areas of the Northeast, Midwest, Great Plains, and Deep South the NCAR A2 scenario projects population loss, leading to a decrease in exposure. S10
Supplementary Figure 14. Projected change in exposure under the constant population (a) and constant climate (b) scenarios. Figure 4 (main text) illustrates the relative contribution of the climate, population, and interaction effects at the national-level. At the census division-level (Supplementary Figure 15) the relative importance of the climate and population effect exhibits substantial variation. In the South Atlantic Division the climate effect is responsible for a larger portion of the change in exposure, whereas in Western Divisions the population effect is a stronger driver of change, particularly in the Mountain Division. In the divisions projected to experience minimal population growth (e.g., New England, Mid-Atlantic) the climate effect is a substantially more significant driver of change in exposure. Supplementary Figure 13 Decomposition of aggregate division-level change in exposure (ensemble mean). S11
To further decompose the population effect we ran two additional constant climate scenarios in which the base-period spatial distribution of the population was also held constant. In the first the population was scaled (at the grid-cell level) by projected national-level population change, and in the second population was scaled according to projected population change at the level of census division (Supplementary Figures 16a and 16b, respectively). In the nationally-scaled scenario change in exposure is greatest in large urban areas in those regions already experiencing a large number of extreme heat days such as Los Angeles, Phoenix, Dallas, and Houston, as well as the Central Valley of California and the Southern Great Plains. Because the spatial distribution of the population is held constant in both scenarios the divisionally-scaled scenario exhibits a pattern identical to the nationally-scaled scenario. However, the intensity of projected change is larger in those divisions projected proportionally more population growth such as the Pacific (Los Angeles), Mountain (Phoenix), West South Central (Dallas, Houston, San Antonio), and South Atlantic (Atlanta, Washington DC, Orlando) Divisions. a. b. Supplementary Figure 16. Projected change in exposure under the constant climate nationally-scaled (a) and divisionally-scaled (b) scenarios. S12
References 1. Nakicenovic, N. et al. Special report on emissions scenarios. A special report of working group III of the intergovernmental panel on climate change. (Cambridge University Press, 2010). 2. Jones, B. & O Neill, B. C. Historically grounded spatial population projections for the continental United States, Environ. Res. Lett. 8, 044021 (2013). 3. Mearns, L. O. et al. Climate change projections of the North American Regional Climate Change Assessment Program (NARCCAP). Clim. Change 120, 965-975 (2013). 4. Collins, W. D. et al. The Community Climate System Model: CCSM3. J. Clim. 19, 2122-2143 (2006). 5. Flato, G. M. The third generation coupled global climate model (CGCM3). Available at: http://www.ec.gc.ca/ccmac-cccma/default.asp?n=1299529f-1 (2005). 6. Gordon, C. et al. The simulation of SST, sea ice extents and ocean heat transports in a version of the Hadley Centre coupled model without flux adjustments. Clim. Dyn. 16, 147-168 (2000). 7. Anderson, J. et al. The new GFDL global atmospheric and land model AM2-LM2: Evaluation with prescribed SST simulations. J. Clim. 17, 4641-4673 (2004). 8. McGinnis, S., Nychka, D., and Mearns, L.O. A new distribution mapping technique for bias correction of climate model output. In: Machine Learning and Data Mining Approaches to Climate Science: Proceedings of the Fourth International Workshop on Climate Informatics, Lakshmanan, V., Gilleland, E., McGovern, A., & Tingley, M., eds. (Springer, 2015, in press) 9. Maurer, E. P., Wood, A.W., Adam, J.C., Lettenmaier, D.P. & Nijssen B. A long-term hydrologically-based data set of land surface fluxes and states for the conterminous United States, J. Clim. 15, 3237-3251 (2002). S13