This is the fourth in a series of four articles on historical and projected climate extremes in the United States. Here, we examine the results of historical and future climate model experiments from the phase 5 of the Coupled Model Intercomparison Project (CMIP5) based on work presented at the World Climate Research Programme (WCRP) Workshop on CMIP5 Climate Model Analyses held in March 2012. Our analyses assess the ability of CMIP5 models to capture observed trends, and we also evaluate the projected future changes in extreme events over the contiguous Unites States. Consistent with the previous articles, here we focus on model-simulated historical trends and projections for temperature extremes, heavy precipitation, large-scale drivers of precipitation variability and drought, and extratropical storms. Comparing new CMIP5 model results with earlier CMIP3 simulations shows that in general CMIP5 simulations give similar patterns and magnitudes of future temperature and precipitation extremes in the United States relative to the projections from the earlier phase 3 of the Coupled Model Intercomparison Project (CMIP3) models. Specifically, projections presented here show significant changes in hot and cold temperature extremes, heavy precipitation, droughts, atmospheric patterns such as the North American monsoon and the North Atlantic subtropical high that affect interannual precipitation, and in extratropical storms over the twenty-first century. Most of these trends are consistent with, although in some cases (such as heavy precipitation) underestimate, observed trends.
CMIP5 model simulations of historical and projected climate extremes in the United States are assessed.
This is the fourth in a series of BAMS articles on climate extremes in the United States (U.S.). These papers are based on workshops where leading scientists in the field came together to determine how best to assess the state of the science in understanding long-term climate variability and changes in various types of extreme events affecting the United States. The first workshop focused on severe local storms (Kunkel et al. 2013). The second workshop focused on the larger-scale phenomena of heat waves, cold waves, floods, and drought (Peterson et al. 2013). The third workshop examined the current understanding of coastal issues, including observed trends in winds, waves, and extratropical storms (Vose et al. 2014). One of the outcomes of those workshops and the resulting papers was the collective assessment of the state of knowledge regarding changes in various climate extremes (Fig. 1). The findings in these workshops also strongly correlate with the global findings in the recent Intergovernmental Panel on Climate Change (IPCC) special report on extreme events (Field et al. 2012; often termed the SREX report).
The previous three workshops focused on the state of current knowledge regarding observed trends in, and drivers of, extreme events. The fourth workshop, the World Climate Research Programme (WCRP) Workshop on phase 5 of the Coupled Model Intercomparison Project (CMIP5) Climate Model Analyses held in March 2012, focused on the ability of the latest generation of climate models to capture observed trends and features of the physical climate system. Our intent in this paper is to assess the latest scientific understanding of CMIP5 model ability to simulate observed and future trends in climate extremes; it is not our intention to provide a complete summary of the entire body of work presented at the WCRP workshop, nor are we able to summarize results from other climate model experiments external to the CMIP5 experiment. For consistency with the previous three workshops, this assessment is limited to those papers presented at that workshop that specifically focused on climate and extreme events relevant to the U.S. With its geographic focus on the U.S., this paper also contributes the ongoing U.S. National Climate Assessment (www.globalchange.gov/what-we-do/assessment).
With participation from over 20 modeling groups and more than 40 global models, CMIP5 represents the latest and most ambitious coordinated international climate model intercomparison exercise to date (Taylor et al. 2012). CMIP5 includes a wide range of experiments addressing cloud feedbacks, carbon cycle feedbacks, and paleoclimate. Here, we focus on simulations of the twentieth century based on natural and anthropogenic forcings and the twenty-first century (with extensions to 2300) based on four new scenarios called representative concentration pathways (RCPs; Meehl and Hibbard 2007; Hibbard et al. 2007; Moss et al. 2010; van Vuuren et al. 2011).
Phase 3 of the Coupled Model Intercomparison Project (CMIP3) was the first coordinated international set of climate model experiments to include twentieth- and twenty-first-century experiments (Meehl et al. 2005, 2007). The IPCC's third and fourth Assessment Reports (TAR and AR4) were largely based on CMIP3 simulations. Given the increases in spatial resolution and other improvements in climate modeling capabilities over the last decade since the CMIP3 simulations were completed, CMIP5 provides a unique opportunity to assess scientific understanding of climate variability and change over a range of historical and future conditions.
Despite increases in model resolution and complexity, projected patterns and magnitudes of future temperature and precipitation changes are not substantially different from CMIP3 to CMIP5, both globally and over North America, when differences in forcings are accounted for. Estimates of climate sensitivity (Andrews et al. 2012) and hence the range in future projections due to uncertainty in climate sensitivity is also largely unchanged (Knutti and Sedláček 2012).
To put the new RCP scenarios and CMIP5 models in context, Fig. 2 compares historical simulated and projected future changes in annual-mean surface air temperature averaged over the contiguous United States (CONUS) for the period 1900–2100 as simulated by the CMIP3 model ensemble using the Special Report on Emissions Scenarios (SRES) scenarios (Nakicenovic and Swart 2000) and CMIP5 model ensemble using the RCP scenarios. An ensemble average for each model based on all available realizations was calculated prior to the calculation of the equally weighted multimodel averages. At the lower end of the range, the stabilization scenario RCP2.6 reaches its peak change of 2°C above the 1901–60 average around the middle of this century. The higher scenario, RCP8.5, drives end-of-century temperature increases in excess of 6°C, significantly warmer than those projected by SRES A2. SRES B1 and RCP4.5 produce similar but not identical responses over the U.S. at the end of the century, as do SRES A1B and RCP6.0.
The CONUS mean change is similar to that projected for the global mean (land and ocean). An assessment of the total uncertainty for the CONUS projections would be almost certainly larger (Knutti and Sedláček 2012) but is not as straight-forward to estimate as previously done for global-mean temperature change in the IPCC AR4 using simple climate models, as there is no way to estimate regional climate sensitivities from those models.
Here, we focus on CMIP5-simulated historical and projected future trends in extreme temperature, heavy precipitation, drought, and extratropical cyclones. The CMIP5 models used in the various analyses are listed in Table S1 (i.e., more information can be found online at http://dx.doi.org/10.1175/BAMS-D-12-00172.2).
Observations dating back to 1900 show that the temperatures in the twenty-first century have the largest spatial extent of record breaking and much above normal mean monthly maximum and minimum temperatures (Karl et al. 2012). However, the frequency of intense short-duration hot spells is still second to the levels achieved during the hot and dry 1930s (Peterson et al. 2013; hot spells were defined as 4-day periods whose mean temperatures exceeded a threshold for a 1-in- 5-yr recurrence). There is also a highly significant decrease in record-breaking cold months including decreases in short-duration cold spells from a maximum in the 1980s to the lowest levels on record in the twenty-first century (Peterson et al. 2013; cold spells were defined as 4-day periods whose mean temperatures were below a threshold for a 1-in-5-yr recurrence). CMIP5-simulated changes in extreme high and low monthly temperatures (defined here as the single hottest and coldest months in a 30-yr period) show that these are expected to grow over time. Projected multimodel mean increases in the temperature of the hottest and coldest months of the year are large across the U.S. under the RCP8.5 scenario (see Fig. ES1 in the supplementary materials). For the contiguous U.S., cold spell temperature increases range from around 3°C in Florida to more than 8°C in the north-central U.S. for 2071–99 compared to 1971–2000. Hot spell temperature increases range from around 5°C in far southern areas and along the west coast to more than 7°C in parts of the Midwest and northern Rockies. Temperature increases in Alaska (Hawaii) are similar (slightly lower) for the hottest month and greater (lower) for Alaska (Hawaii) for the coldest month.
Using metrics for the combined temperature–humidity health effects [e.g., heat index (Steadman 1979), temperature–humidity index (HUMIDEX; Masterson and Richardson 1979), and wet bulb globe temperature (Sherwood and Huber 2010)], both CMIP5 and earlier model simulations consistently project increasing levels of heat stress across the U.S. (e.g., Delworth et al. 1999; Sherwood and Huber 2010; Willett and Sherwood 2012; Fischer et al. 2012). While the projected twenty-first-century changes for the temperature component of heat stress vary substantially across CMIP5 models, there is a clear joint behavior; models that show greater warming also show greater reductions in relative humidity over the continental U.S. This implies that projected increases in combined temperature–humidity measures are substantially more robust than from the two contributing variables independently (Fischer and Knutti 2013). Although most models project somewhat lower relative humidity on the hottest days, the combined effect of temperature and humidity changes is substantial increases in heat stress.
The 20-yr return value of the annual maximum or minimum daily temperature is one measure of changes in rare temperature extremes. In a changing climate, this metric is interpreted as a temperature that has a 5% chance of being exceeded by an annual extreme in any given year. Figure 3 (bottom) shows the projected change in the 20-yr return value of the annual maximum daily surface air temperature over North America at the end of this century (2081–2100) relative to the recent past (1986–2005) for the higher and lower emission RCP scenarios (Kharin et al. 2013). Under the lower RCP2.6 scenario, current annual maximum temperature extreme values are projected to occur between 4 and 10 times more frequently than at the beginning of the twenty-first century. Under the higher RCP8.5 scenario, current annual maximum extremes are projected to occur every year over the entire continent except for parts of Alaska. Figure 3 (top) shows the same for annual minimum daily surface air temperature, which is considerably larger than for the hot extreme temperatures. Under the RCP2.6 scenario, annual minimum temperature extreme values are projected to occur half as often in the southern states and about five times less often in the northern states. Under the RCP8.5 scenario, these minimum extreme values are not projected to recur over most of the continent.
Generally the bias in CMIP5 temperature extremes compared to observations follow similar errors to the corresponding seasonal mean. For warm extremes in Fig. 3, the CMIP5 models are 2°–5°C too high in the east half of the U.S. for return values calculated from 1986 to 2005 but lower than 2°C in the western half. For the cold extremes, the CMIP5 models are slightly more than 2°C colder than observed in the western half of the U.S. and less and 2°C colder than observed in the eastern half (Kharin et al. 2013). Multimodel differences in reproducing 1986–2005 observed temperature return values over land areas are slightly larger than the differences in reproducing observed mean seasonal temperatures with an average multimodel standard deviation of about 5°C (Kharin et al. 2013).
The frequency of record-breaking high or low monthly temperatures is another measure of extreme temperature change (Meehl et al. 2009). Figure 4 compares the frequency of high and low record monthly temperatures over a 50-yr period averaged over U.S. During the 1990s the high record frequency was about 0.5 month yr−1, roughly double that expected in an unchanging climate (top panel). An increase in mean temperature itself increases the chance of breaking a record high temperature by about 50%, as reflected by the red dashed curve (Wergen and Krug 2010). Similarly, minimum temperature records were overwhelmingly lower than would be expected in an unchanging climate (about 0.12), producing a high/low record-breaking temperature ratio of 4 during the last 10 yr—a ratio that should be equal to 1 in an unchanging climate. The historical runs capture the general decay trend in record frequency, but the high (low) temperature records recur slightly less (more) frequently than observed during mid-1980s to mid-1990s (middle panel).
Projections using the mid–low RCP4.5 scenario show that the high (low) monthly record temperatures would occur much more (less) frequently in the future (bottom panel) with respect to the 50-yr time frame starting in 2006. By the middle of the century under mid–low emissions (RCP4.5), record high temperatures are projected to be broken at a rate of 0.9 months yr−1 and record low temperatures at a rate of 0.07 months yr−1, which gives a high/low temperature record ratio greater than 10 (a value achieved for daily records in July 2012, during the worst U.S. drought in the past five decades). This large ratio cannot be entirely explained by the increase in mean temperature; rather, it suggests a change in the shape of the tails of the daily temperature distribution, consistent with other studies of extreme temperature (Wehner 2005).
The extreme precipitation index (EPI; Kunkel et al. 1999, 2003, 2007) has been previously used to provide strong evidence for an upward trend in the frequency and intensity of extreme precipitation events in the U.S. (e.g., Kunkel et al. 2013). Figure 5 (top) compares EPI decadal anomalies based on CMIP5 models to observations for 2-day duration 1-in-5-yr events over the CONUS. The EPI was calculated annually from 1901 to 2005, and then decadal averages were calculated for the period 1906–2005.
A positive trend in EPI anomalies is evident from observations over the past 4 decades. The multimodel median of CMIP5 simulations also shows an increasing trend in EPI anomalies over the same time period, albeit smaller than observed. The standard deviation between the models is extremely large, often greater than the signal, indicating that there are large differences between extreme precipitation events in the models (see Fig. ES2 in the supplementary materials, which compares the correlation coefficient of observed and modeled decadal average EPI values for the CONUS for each of the 26 CMIP5 models used). Many models have a correlation coefficient with observations greater than 0.50, with the Beijing Climate Center, Climate System Model, version 1.1 (BCC_CSM1.1), for example, approaching 1.00 for a 10-yr return. At the same time, however, seven of the models have a negative correlation, demonstrating the large spread in model ability to capture observed trends in extreme precipitation events. In terms of future projections, Fig. 5 (bottom) shows an increasing trend in EPI values under both the mid–low RCP4.5 and the higher RCP8.5 scenarios. For these projections, the multimodel spread is smaller than the signal, indicating strong agreement of an increase in the EPI across all models. Figure ES3 in the supplementary materials shows that there is a large variation between ensemble individual runs.
An alternate indicator of long-term trends in extreme precipitation is the fraction of the annual total precipitation that falls in the heaviest 1% of daily events. Figure 6 compares simulated historical changes in the top 1% of extreme CONUS precipitation over time with observed data, calculating the 99th percentile for the base period (1900–60), ignoring all days with less than 1 mm of precipitation at each grid point, and summing the data for days above that threshold. The models show an increase in the amount of precipitation falling in the largest 1% of events throughout the last century (1901–2000). CMIP5 historical changes in heavy precipitation are broadly consistent with changes in observed heavy precipitation from 1958 to 2007 (Karl et al. 2009). By the end of this century, a 50% increase in the annual fraction of precipitation falling in the heaviest events is projected for the mid–low scenario (RCP4.5), while a 90% increase is projected for the higher scenario (RCP8.5). In general, CMIP5 results suggest that a greater percentage of annual precipitation will fall in the top 1% of events over time and are consistent with the conclusions reached in similar analyses of CMIP3 models (Wehner 2005).
Long period return values represent much rarer extremes than the 99th percentile. Figure 7 (upper) shows that the CMIP5 projection of percent changes in the 20-yr return value of the annual maximum daily precipitation at the end of this century (2081–2100) relative to the recent past (1986–2005) under the higher and lower RCP scenarios increases everywhere in CONUS and Alaska (Kharin et al. 2013). Such rare precipitation events have been increasing (Kunkel et al. 2003; Min et al. 2011; Field et al. 2012) and are also projected to occur more frequently in the future (Fig. 7, lower) but not as often as for warm temperature events of the same current frequency. At the end of this century under the higher RCP8.5 scenario, the current 20-yr event is projected to occur about twice as often in the interior of the U.S., about 3–4 times more frequently along the coasts, and up to 7 times more frequently in parts of Alaska than it does now.
LARGE-SCALE DRIVERS OF PRECIPITATION VARIABILITY AND DROUGHT.
Drought has been a constant challenge for the U.S. Southwest and, in recent years, for the Southeast as well. A sizeable fraction of the precipitation in the arid Southwest derives from the North American monsoon. Past studies using CMIP3 models (e.g., Liang et al. 2008) have shown that climate models do not simulate all aspects of the circulation patterns associated with the monsoon well. The CMIP5 models' simulation of the seasonal cycle of precipitation (Cook and Seager 2013) appears improved over CMIP3 (Liang et al. 2008). Observations indicate a decrease in monsoon rainfall over the past two decades, which may partially result from anthropogenically forced warming (Arias et al. 2012). CMIP3 analyses of changes in the annual cycle of precipitation in the North American monsoon region suggested reductions during winter and early summer rainy seasons but indicate increased rainfall later in the rainy season (Seth et al. 2011). CMIP5 simulations (Fig. 8) show a similar response but a stronger reduction in precipitation in the winter and spring, possibly owing to the stronger radiative forcing in the RCP8.5 versus the SRES A2 scenario (also see Seth et al. 2013). A 1979–2005 historical comparison indicates CMIP5 models are drier (by 2–4 mm day−1) than observed through March–September in the core monsoon region, south of 20°N, and slightly wetter (0.5–2 mm day−1) than observed from May to September from 20° to 25°N (Sheffield et al. 2013).
In the Southeast, the seasonal cycle of precipitation is strongly influenced by the position of the western ridge of the North Atlantic subtropical high (NASH). Comparing historical simulated and observed year-to-year variations in summer [June–August (JJA)] precipitation in the Southeast U.S. identified a subgroup of the CMIP5 models that simulate the summer precipitation variability reasonably well owing to their proper representation of the link with the western ridge position. In this subgroup of models, future variability intensifies under the mid–low RCP4.5 scenario due to a pattern shift of the NASH western ridge. The NASH western ridge extends farther westward and leads to more frequent occurrences of both the northwestward and southwestward ridge patterns that are respectively related to the dry and wet Southeast U.S. summers—in other words, increasing interannual variability (Li et al. 2011, 2013).
At the global scale, previous evaluations of CMIP3 twenty-first-century projections (Sheffield and Wood 2008) indicated general decreases in soil moisture and a corresponding increase in drought frequency, duration, area, and severity with increasing temperature. CMIP5 models show similar twenty-first-century decreases in soil moisture in most global land areas in summer. There has been a recent increase in the frequency of severe to extreme drought in the western U.S., and the CMIP5 models simulate such an increase for the early twenty-first century (Fig. 9). There is consensus among the models for future summer soil moisture decreases throughout the U.S.. and for winter soil moisture decreases in most of the CONUS (Dirmeyer et al. 2013). Comparisons of CMIP3 and CMIP5 twentieth-century simulations against offline hydrological modeling estimates of global drought variability (Sheffield and Wood 2007) indicate that the models on average capture the regional variation in drought frequency, although there are large intermodel variations and a tendency to overestimate longer-term drought frequency (Fig. 9). The latter is related to differences in modeled variability at interannual to decadal time scales and differing land surface representations.
The south-central U.S. has been prone to drought and floods historically and experienced its worst single year drought in 2011. By the late twenty-first century (2073–99), the CMIP5 models ensemblemean projections suggest that the net surface water gain over this region, defined as the precipitation minus evapotranspiration (P – ET), will decrease significantly during winter, spring, and fall significantly (~0.2 mm day−1 or 20%) under the RCP8.5 scenario relative to that of 1979–2005. Such changes are mainly due to a stronger increase of ET during these seasons, which more than negates a small increase of rainfall during spring. Because soil moisture is recharged during winter and spring in the current climate (1979–2005), the projected reduction of net surface water gain in these seasons would reduce soil moisture and increase the risk of droughts.
Future changes in extratropical cyclones could affect the risk and severity of extreme precipitation over the CONUS, particularly along the eastern seaboard. Recent observational studies have documented a decrease in the frequency of warm season extratropical cyclones over the northeastern U.S. (Leibensperger et al. 2008), while other studies have shown a future decrease in cyclone frequency over the western Atlantic storm track using CMIP3 and other models (Lambert and Fyfe 2006; Bengtsson et al. 2006). Colle et al. (2013) present a more detailed summary of past studies investigating future cyclone changes for this region. Our analysis focuses on eastern North America during the cool season (November–March) using the Hodges (1994, 1995) cyclone tracking scheme to track the cyclones in 15 CMIP5 models (see supplement) using 6-h mean sea level pressure data. Colle et al. (2013) describes the tracking approach and some validation of the tracking procedure for the historical 1979–2004 period during the cool season. They also rank the models and show that six out of the seven top-performing models are the higher-resolution CMIP5 models.
Figure 10a shows the change in cyclone track density over eastern North America and much of the northern Atlantic between the 2039 and 2068 cool seasons and the historical (1979–2004) period, and dotted locations highlight where at least 73% (11 of the 15) of the models predict the same sign of the cyclone changes. Projected changes in cyclone tracks and cyclone deepening, or strengthening, vary substantially from one region to the next. For example, cyclone density is projected to decrease over the western Atlantic but change little or slightly increase over northern New England. Over the smaller U.S. East Coast region, relatively weak cyclones are projected to decrease while stronger cyclones (<980 hPa) are projected to increase (Fig. 10c); however, there is a relatively large standard deviation in the future change of deep cyclones, ranging from a near doubling to no change. Colle et al. (2013) highlights a statistically significant upward trend in the number of relatively strong cyclones along the U.S. East Coast through the mid-twenty-first century using the “best seven” CMIP5 models. In contrast, for the larger Atlantic domain there is a 3%–9% projected reduction in the number of relatively strong cyclones (Colle et al. 2013).
The rate at which cyclones strengthen, or deepen, is also projected to change. Over the Northeast, there is a 10%–30% mean increase in the number of CMIP5 cyclones deepening by more than 5 hPa in 6 h (Fig. 10d), with a relatively large spread from a 40% to 60% decrease to a 60% to 90% increase. Meanwhile, the mean CMIP5 weakening rates of more than 2 hPa in 6 h decrease by ~5%, but there is a relatively large uncertainty in this weakening. Just offshore of the U.S. East Coast deepening rates are projected to decrease by 10%–20% by the mid-twenty-first century (Fig. 10b). By the late twenty-first century, a widespread 10%–30% decrease in 5 hPa per 6-h deepening is projected over much of the western and northern Atlantic (not shown). Colle et al. (2013) provide some evidence to suggest this more rapid deepening is the result of increased latent heating. Overall, these results highlight the enormous complexity of projecting the impacts of global change on regional dynamics and storm systems. Additional research is needed using higher-resolution regional models, but overall these CMIP5 results suggest that increasing cyclone intensity may lead to more wind and heavy precipitation extremes along the U.S. East Coast.
This paper summarizes the results of a series of analyses based on the CMIP5 models examining changes in temperature extremes, precipitation extremes, droughts, and atmospheric patterns such as the North American monsoon and the North Atlantic subtropical high that affect extreme temperatures, interannual precipitation, and extratropical cyclones over the continental U.S.. Based on these analyses, including the comparison of the new CMIP5 model experiments with older CMIP3 projections and, where possible, with historical observed trends, we find the following:
Despite higher model resolution and increased complexity, the spatial patterns, direction of change, and overall magnitudes of projected changes in mean and extreme temperature and precipitation do not differ substantially from CMIP3 to CMIP5, particularly when differences in forcings are accounted for.
Historical observations, model simulations, and future projections consistently show increases in the frequency of high temperature extremes and decreases in low temperature extremes across different indicators that cover a broad range of return periods, quantiles, or record-breaking frequencies.
Observations, historical simulations, and future projections also agree on increases in heavy precipitation events consistent across a range of indicators. However, there are large differences between model simulations in the rate of heavy precipitation increase, with many tending to underestimate the historical observed trend. Models do project a further increasing trend in severe precipitation events in the future.
Projected changes in drought risk based on soil moisture show consistent increases in both summer and winter seasons across the U.S. as a whole. Model ability to simulate large-scale dynamical features such as the North American monsoon (for the Southwest) and the North Atlantic subtropical high (for the Southeast) is critical to simulating trends in long-term summer drought risk for those regions and CMIP5 models vary in the accuracy of their simulations of these features
Although extratropical cyclones may become weaker and less frequent over much of the western Atlantic storm track, they may become more intense and deepen more rapidly just inland of the U.S. East Coast, especially by the middle of the twenty-first century. The CMIP5 analyses suggest that increasing cyclone intensity may lead to more wind and heavy precipitation extremes along the U.S. East Coast.
The studies presented in this paper provide preliminary analyses of CMIP5 and the comparison with historical trends and with CMIP3 results for extreme events. We believe this is a useful first look at how our confidence in the patterns and direction of change for extreme events has solidified as better and higher-resolution models have become available, particularly as these new model simulations continue to paint the same broad-scale picture of increasing trends in high temperature and precipitation extremes found in earlier studies.
This work was supported in part by NASA Award NNX12AF32G (University of Illinois), from the Regional and Global Climate Modeling Program and the Earth System Modeling Program of the Office of Biological and Environmental Research in the Department of Energy Office of Science under Contract DE-AC02-05CH11231 (LBL), from NSF Grant AGS-1147608 (Duke), and by NOAA Award NA11OAR4310097 (Princeton University). We also acknowledge the support of NOAA Climate Program Office Modeling, Analysis, Predictions and Projections (MAPP) Program as part of the CMIP5 Task Force under Grant NA11OAR4310104 (Stony Brook) and NOAA NESDIS through the Cooperative Institute for Climate and Satellites—North Carolina under Cooperative Agreement NA09NES4400006.
A supplement to this article is available online (10.1175/BAMS-D-12-00172.2)