Transient regional climate change: Analysis of the summer climate response in a high-resolution, century-scale ensemble experiment over the continental United States

Authors

  • Noah S. Diffenbaugh,

    1. Woods Institute for the Environment and Department of Environmental Earth System Science, Stanford University, Stanford, California, USA
    2. Purdue Climate Change Research Center and Department of Earth and Atmospheric Sciences, Purdue University, West Lafayette, Indiana, USA
    Search for more papers by this author
  • Moetasim Ashfaq,

    1. Woods Institute for the Environment and Department of Environmental Earth System Science, Stanford University, Stanford, California, USA
    2. Purdue Climate Change Research Center and Department of Earth and Atmospheric Sciences, Purdue University, West Lafayette, Indiana, USA
    3. Oak Ridge National Laboratory, Oak Ridge, Tennessee, USA
    Search for more papers by this author
  • Martin Scherer

    1. Woods Institute for the Environment and Department of Environmental Earth System Science, Stanford University, Stanford, California, USA
    Search for more papers by this author

Abstract

[1] Integrating the potential for climate change impacts on policy and planning decisions requires quantification of the emergence of subregional climate changes that could occur in response to transient changes in global radiative forcing. Here we report results from a high-resolution, century-scale ensemble simulation of climate in the United States, forced by atmospheric constituent concentrations from the Special Report on Emissions Scenarios A1B scenario. We find that 21st century summer warming permanently emerges beyond the baseline decadal-scale variability prior to 2020 over most areas of the continental United States. Permanent emergence beyond the baseline annual-scale variability shows much greater spatial heterogeneity, with emergence occurring prior to 2030 over areas of the southwestern United States but not prior to the end of the 21st century over much of the south central and southeastern United States. The pattern of emergence of robust summer warming contrasts with the pattern of summer warming magnitude, which is greatest over the central United States and smallest over the western United States. In addition to stronger warming, the central United States also exhibits stronger coupling of changes in surface air temperature, precipitation, and moisture and energy fluxes, along with changes in atmospheric circulation toward increased anticylonic anomalies in the midtroposphere and a poleward shift in the midlatitude jet aloft. However, as a fraction of the baseline variability, the transient warming over the central United States is smaller than the warming over the southwestern or northeastern United States, delaying the emergence of the warming signal over the central United States. Our comparisons with observations and the Coupled Model Intercomparison Project Phase 3 ensemble of global climate model experiments suggest that near-term global warming is likely to cause robust subregional-scale warming over areas that exhibit relatively little baseline variability. In contrast, where there is greater variability in the baseline climate dynamics there can be greater variability in the response to elevated greenhouse forcing, decreasing the robustness of the transient warming signal.

1. Introduction

[2] Although the influence of anthropogenic increases in greenhouse gas concentrations on global and continental-scale temperatures is now well understood [Intergovernmental Panel on Climate Change, 2007], there remain a number of important outstanding unknowns about the response of the climate system to changes in global radiative forcing. One of the most prominent is the response of processes that govern climate at subregional spatial scales [e.g., Giorgi et al., 2008]. This response is particularly important for understanding possible climate change impacts, as fine-scale processes can create substantial subregional heterogeneity in changes to impacts-critical climate variables [e.g.,Ashfaq et al., 2010a; Diffenbaugh et al., 2005, 2007; Fischer and Schär, 2010; Gao et al., 2008, 2006; Hirschi et al., 2011; Rauscher et al., 2008; White et al., 2006].

[3] Integrating the potential for climate change impacts into policy and planning decisions requires quantification of the emergence of subregional climate changes that could occur in response to transient changes in global radiative forcing [e.g., Carter et al., 2007]. However, this transient response is likely to be strongly influenced by the internal variability of the climate system [e.g., Diffenbaugh and Scherer, 2011], particularly in the relatively narrow range of forcing expected over the near-term decades [e.g.,Cane, 2010; Hawkins and Sutton, 2009; Hurrell et al., 2009]. Indeed, although some regions show robust climate change in response to expected near-term increases in global radiative forcing [e.g.,Anderson, 2011a, 2011b; Diffenbaugh and Ashfaq, 2010; Diffenbaugh and Scherer, 2011], internal climate system variability exerts a stronger and more persistent influence on regional scales than at the global scale, dominating the near-term spread in global climate model projections over many regions [Hawkins and Sutton, 2009, 2011].

[4] In the current study, we focus on the time of emergence of summer temperature change beyond the variability that currently exists in different areas of the continental United States. The time of emergence of climate change beyond the baseline variability is an indicator of the statistical robustness of the projected change. In addition, systems experiencing change that substantially and permanently exceeds the baseline variability will need to adapt to a new climate that exhibits little overlap with the present temperature range. Systems experiencing earlier permanent emergence face a more rapid transition to this new climate, while systems experiencing later emergence face a more gradual transition. Therefore, in addition to being a measure of statistical robustness, the time of emergence is also an important measure of the pace with which projected global warming is likely to move the climate of a given area outside of the envelope of baseline variations to which systems are currently accustomed. Given the importance of a range of time scales of climate variability, we explore the time of emergence of climate change beyond both the annual-scale and decadal-scale baseline variability.

[5] Our emphasis on the emergence of summer temperature change is motivated by the importance of summer temperature for a variety of natural and human systems, including impacts on air quality [Jacob and Winner, 2009], human health [Kalkstein and Greene, 1997; Kunkel et al., 1996; Poumadère et al., 2005], energy demand [Colombo et al., 1999], water demand [Worthington and Hoffman, 2008], crop yield and quality [Jones et al., 2010; Lobell et al., 2007; Schlenker and Roberts, 2009], food security [Battisti and Naylor, 2009], ecosystem productivity [Ciais et al., 2005], forest mortality [Breshears et al., 2005], wildfire activity [Westerling et al., 2006] and coastal fog [Johnstone and Dawson, 2010]. Given these important impacts, we focus our time of emergence quantification on summer temperature. However, given the clear importance of other climate variables, including the potential for coupling between temperature and surface energy and moisture fluxes [e.g., Chang and Wallace, 1987; Hong and Kalnay, 2000; Koster et al., 2004; Manabe et al., 1981; Schubert et al., 2004b; Wetherald and Manabe, 1995], we analyze a suite of atmospheric and surface variables in addition to temperature, including relationships between antecedent spring conditions and subsequent summer temperature.

[6] The challenge of understanding the transient emergence of climate change at subregional scales is enhanced by the fact that high-resolution, multidecadal, multimember ensemble climate model experiments are rare in the literature. High-resolution experiments are needed to resolve the fine-scale climate processes that can regulate the magnitude and spatial variability of climate change [e.g.,Christensen and Christensen, 2003; Diffenbaugh et al., 2005; Duffy et al., 2003; Fischer et al., 2007b; Rauscher et al., 2008; White et al., 2006]. Likewise, transient experiments are needed to quantify the magnitude and timing of the climate response as radiative forcing is progressively enhanced [Christensen et al., 2007; Hawkins and Sutton, 2009, 2011]. Further, multimember ensemble experiments are needed to capture the effects of internal climate system variability on the evolution of climate in response to transient changes in radiative forcing, particularly on near-term decadal time scales [e.g.,Hawkins and Sutton, 2009, 2011; Meehl et al., 2010].

[7] Diffenbaugh and Ashfaq [2010]report results of the first high-resolution multidecadal ensemble simulation over the continental United States, simulating the period from 1950 to 2039 in the Intergovernmental Panel on Climate Change (IPCC) Special Report on Emissions Scenarios (SRES) A1B scenario [Intergovernmental Panel on Climate Change, 2000]. They find that this high-resolution ensemble simulation improves the representation of extremely hot conditions in the continental United States relative to the Coupled Model Intercomparison Project Phase 3 (CMIP3) global atmosphere-ocean general circulation model (AOGCM) ensemble. They also find substantial intensification of hot extremes in the United States over the next three decades as greenhouse gas concentrations increase in the A1B scenario. This intensification is associated with a shift toward anomalously dry summer conditions at the surface and anomalously anticyclonic circulation in the midtroposphere. However, the near-term warm season response is far more robust over the western United States than over the central and eastern United States. Given the spatial contrast in robustness over the near-term decades, we extend the simulations ofDiffenbaugh and Ashfaq [2010]through the end of the 21st century, providing a new high-resolution, century-scale ensemble experiment through which to probe the processes that shape the response of climate in the United States to transient changes in global radiative forcing.

2. Models and Methods

2.1. High-Resolution Climate Modeling System

[8] In the current study, we extend the nested simulations described in the study of Diffenbaugh and Ashfaq [2010]. These simulations employ the high-resolution Abdus Salam International Centre for Theoretical Physics (ICTP) RegCM3 climate model [Pal et al., 2007]. RegCM3 is a hydrostatic, sigma coordinate, primitive equation, limited-area model. RegCM3 uses the hydrostatic dynamical core ofGrell et al. [1994], the radiation package of Kiehl et al. [1996], the land model of Dickinson et al. [1993], and the boundary layer scheme of Holtslag et al. [1990]. In our simulations, precipitation processes are parameterized using the Subgrid Explicit Moisture Scheme (SUBEX) of Pal et al. [2000] and the cumulus convection scheme of Grell [1993] (with the closure assumption of Fritsch and Chappell [1980]). Further details of RegCM3 are summarized by Pal et al. [2007]. We note that RegCM3 does not use spectral nudging. Although spectral nudging remains an active area of nested climate model development [e.g., Cha et al., 2011; Colin et al., 2010; Termonia et al., 2011], allowing the nested model to calculate its own high-resolution solution within the lateral boundaries remains a common approach (for example, two of the models in the ongoing North American Regional Climate Change Assessment Program (NARCCAP) project use spectral nudging, while four models do not (NARCCAP,http://www.narccap.ucar.edu/about/index.html, 2011)).

[9] We use the equal-area grid employed byDiffenbaugh et al. [2005, Figure S1]. The grid is centered at 39.00°N and 100.00°W, with 145 points in the latitude direction and 220 points in the longitude direction. Grid points are separated at 25 km resolution in the horizontal, with 18 levels in the vertical (up to 50 mbar).

[10] RegCM3 has been extensively applied over the continental United States including as part of the NARCCAP project [Gao et al., 2011; Gutowski et al., 2010; Mearns et al., 2009; Pryor and Barthelmie, 2011; Takle et al., 2010; Wang et al., 2009] (also NARCCAP online publication, 2011), for which data are still becoming available to the community [Gao et al., 2011; Pryor and Barthelmie, 2011; Rasmussen et al., 2011; NARCCAP, 2011]. RegCM3 captures the pattern of seasonal temperature, precipitation and atmospheric circulation observed over the continental United States, the extremes of the seasonal-scale and daily scale distribution of temperature and precipitation, and the pattern of atmospheric anomalies associated with precipitation extremes [Ashfaq et al., 2010a; Diffenbaugh et al., 2006; Diffenbaugh and Ashfaq, 2007, 2010; Gutowski, 2010; Walker and Diffenbaugh, 2009; Wang et al., 2009]. RegCM3 also captures the pattern of severe thunderstorm environments observed over the central United States [Trapp et al., 2007] and of wind power resource across the continental United States [Pryor and Barthelmie, 2011; Rasmussen et al., 2011], along with the pattern of seasonality and trend in snowmelt runoff observed over the western United States [Rauscher et al., 2008]. The most pronounced temperature and precipitation biases include (1) excessive cold season mean precipitation over the high elevations of the western United States, (2) excessive warm season mean precipitation over the southeastern United States, (3) excessive warm season mean and extreme temperatures over the central United States, and (4) excessive interannual variability of warm season mean and extreme temperatures over the central and southeastern United States [Ashfaq et al., 2010a; Diffenbaugh and Ashfaq, 2010; Walker and Diffenbaugh, 2009]. The excessive cold season precipitation over the high elevations of the western United States has been linked to excessively strong cold season surface winds, while the excessive warm season temperatures over the central United States have been linked to insufficient onshore flow over the Gulf of Mexico and insufficient warm season soil moisture and latent heat flux over the central United States [Walker and Diffenbaugh, 2009].

2.2. Climate Model Integrations

[11] We nest the high-resolution RegCM3 model within the National Center for Atmospheric Research (NCAR) Community Climate System Model (CCSM3) AOGCM [Collins et al., 2006] to create a five-member ensemble simulation of the period from 1950 to 2099 in the SRES A1B scenario [Intergovernmental Panel on Climate Change, 2000]. Although more recent versions of the NCAR global climate model are currently available [Gent et al., 2011], generation of a multimember, century-scale, high-resolution experiment requires either (1) running a new experiment at global scale (which remains prohibitive even for national modeling centers [Meehl et al., 2007a; Taylor et al., 2009]) or (2) nesting a high-resolution model within an existing global climate model experiment. Given the infeasibility of the former, we nest RegCM3 within the CCSM3 simulations described byMeehl et al. [2006], which were generated by NCAR and archived as part of the CMIP3 project [Meehl et al., 2007a]. (We note that the NARCCAP project, for which each model is simulating three decades of the mid 21st century U.S. climate at 50 km horizontal resolution, and for which 21st century simulations are still becoming available [Gao et al., 2011; Pryor and Barthelmie, 2011; Rasmussen et al., 2011; NARCCAP, 2011] also nests the respective high-resolution models within CMIP3 AOGCM simulations.)

[12] A description of the large-scale response of the CCSM3 ensemble in the A1B scenario can be found in the work ofMeehl et al. [2006]. In the CMIP3 AOGCM approach, each global ensemble member is initialized from a different point in the same multicentury, preindustrial (“unforced”) simulation. The members are then provided identical radiative forcing boundary conditions for the historical period (mid 19th century through the end of the 20th century) and the scenario period (21st century).

[13] We nest the RegCM3 high-resolution atmospheric model within the CCSM3 global atmospheric model (called the Community Atmospheric Model, or CAM3), updating large-scale atmospheric fields at the lateral boundaries of the RegCM3 grid every 6 simulated hours. We prescribe the 1 degree SST and sea ice fields from the CCSM3 simulations as the RegCM3 lower boundary condition over the ocean, updating the ocean boundary condition each simulated month. Because the three-dimensional atmospheric fields were not saved by NCAR at the subdaily time resolution that is necessary for the high-resolution climate model nesting, we rerun the atmospheric component of CCSM3 (CAM3) over the original CCSM3-generated monthly SSTs and sea ice from five of the CCSM3 simulations originally generated by NCAR (as described byDiffenbaugh et al. [2006] and Trapp et al. [2009]). We generate one global CAM3 simulation for each of the five CCSM3 simulations. As in the original CCSM3 simulations, we run CAM3 with T85 spectral truncation (approximately 1.5 degree resolution in the horizontal), with 26 levels in the vertical (up to 2.917 mbar [Collins et al., 2004]). While this “time slice” approach is necessary in order to generate the high-resolution ensemble experiment, it does not exactly replicate the original CCSM3 simulation for all locations, although statistically significant seasonal-scale differences between the CAM3 and CCSM3 simulations are primarily confined to less than ±1.5 °C for temperature and ±0.5 mm d−1 for precipitation [Ashfaq et al., 2010b]. (The importance of conformity between the global atmospheric time slice simulation and the original CCSM3 AOGCM simulation provides further motivation to use the same atmospheric model that was used in the original CCSM3 experiments (CAM3) rather than a newer version of the NCAR global atmospheric model.)

[14] We generate five high-resolution climate model simulations by nesting one RegCM3 simulation within each of the five CAM3 realizations. (We use the five CCSM3 simulations identified by NCAR as c, e, bES.01, fES.01 and gES.01.) Initial conditions for the CAM3 simulations are taken from the original CCSM3 simulations at the time of the CAM3 initiation (1 January 1948), and initial conditions for the RegCM3 simulations are taken from the CAM3 atmospheric fields and CCSM3 SST fields at the time of RegCM3 initiation (1 January 1950). The CAM3 and RegCM3 models are then run continuously to the year 2099 in the A1B scenario. The greenhouse gas concentrations applied in CAM3 and RegCM3 for the 1950–2099 period are the same as those applied in the original CCSM3 simulations, with annually varying observational concentrations applied over the 1950–1999 period, and annually varying scenario concentrations from the A1B scenario applied over the 2000–2099 period. Our analysis begins in 1970 (seesection 2.3), allowing two decades for model equilibration.

[15] Given that the physical representation and radiative forcing of the climate system are both identical between the five RegCM3 members, our nesting approach generates a “physically uniform” high-resolution ensemble. The difference between the RegCM3 members is that the SSTs and large-scale atmospheric conditions are different over the 1950–2099 period. These differences result from differences in the respective CCSM3 simulations, which themselves use identical radiative forcing and physical representation of the climate system, but are initialized from different points in the multicentury preindustrial CCSM3 simulation.

2.3. Time of Emergence Calculation

[16] We use the 30 year period from 1970 to 1999 as our baseline period. We adapt the time of emergence metric of Giorgi and Bi [2009]to quantify the permanent emergence of temperature change beyond the baseline variability. In the case of our quantification, we calculate the time at which the departure from the 1970–1999 baseline permanently exceeds two standard deviations of the 1970–1999 variability. Given the importance of different time scales of variability, we quantify the time of emergence beyond both annual-scale and decadal-scale variability.

[17] We analyze the annual time series of summer mean values to quantify the time of emergence beyond the baseline annual-scale variability. We first calculate the anomaly from the 1970–1999 mean for each year of the 21st century simulation in each realization. We then calculate the interannual standard deviation of the 1970–1999 period in each realization (after first detrending the 1970–1999 time series). We then identify the year of permanent emergence as the last year for which the seasonal anomaly is less than twice the baseline standard deviation.

[18] We also analyze the 10 year running mean of the annual time series of summer mean values to quantify the time of emergence beyond the baseline decadal-scale variability. We first calculate the 10 year running mean of the annual time series of seasonal values for both the 1970–1999 baseline period and for the 21st century period. We then calculate the anomaly from the 1970–1999 mean for each year of the smoothed 21st century simulations, along with the standard deviation of the 1970–1999 period in the smoothed baseline simulations. We then identify the decade of permanent emergence as the last decade for which the decadal anomaly is less than twice the baseline standard deviation of the smoothed time series.

[19] We calculate the time of emergence for each grid point in the high-resolution RegCM3 ensemble. To evaluate the emergence timing across the ensemble, we use the mean of the seasonal anomalies calculated for each year of the five members, along with the mean of the interannual baseline standard deviations calculated from the five members for the 1970–1999 period. We also calculate the dates of permanent emergence for selected regions in the individual realizations of both the high-resolution ensemble and the CMIP3 AOGCM ensemble [Meehl et al., 2007a]. For these regional emergence calculations, we first create an area-averaged regional temperature time series and then calculate the time of permanent emergence from that regional series.

[20] Because our high-resolution ensemble experiment does not simulate climate beyond the end of the 21st century, we followDiffenbaugh and Scherer [2011] in considering only those final emergence dates that occur at least two decades prior to the end of the simulation to be permanent, meaning that we disregard all final emergence dates that occur after the year 2080.

3. Results and Discussion

3.1. Emergence of Summer Warming

[21] Decadal-scale warming that permanently exceeds the baseline decadal-scale variability emerges prior to 2020 over most areas of the continental United States (including prior to 2010 over the western United States), and prior to 2040 over all areas of the continental United States (Figure 1). The earliest emergence of summer warming that permanently exceeds the baseline annual-scale variability occurs over the southwestern United States and northern Mexico, including widespread emergence prior to 2040 (Figure 1). Permanent annual-scale warming also emerges over parts of the Pacific Northwest prior to 2040, and over most of the western United States prior to 2050. Permanent annual-scale warming likewise emerges over much of the northeastern United States between 2030 and 2060, and over much of the north central United States between 2040 and 2080. However, summer warming that permanently exceeds the annual-scale baseline variability does not emerge prior to the end of the 21st century over much of the south central and southeastern United States, a pattern that is similar to the pattern of summer cooling and muted warming that has been observed over those regions in recent decades [Trenberth et al., 2007].

Figure 1.

Time of emergence of summer warming beyond the baseline temperature variability. The color contours show the last year in which the departure of summer temperature from the 1970–1999 baseline is less than two standard deviations of the 1970–1999 temperature variability. The departure and baseline variability are calculated as the mean of the values in the five RegCM3 members. (top) The annual departures are calculated from the 21st century annual summer temperature time series, and the baseline variability is calculated as the interannual standard deviation of the 1970–1999 annual summer temperature time series. (bottom) The annual departures are calculated from the 10 year running mean of the 21st century annual summer temperature time series, and the baseline variability is calculated as the interannual standard deviation of the 10 year running mean of the 1970–1999 annual summer temperature time series. As described in the methods, because we cannot confirm permanent emergence beyond the end of the 21st century, we disregard all final emergence dates that occur after 2080.

[22] The substantial difference in the time of emergence of annual-scale and decadal-scale summer warming can be diagnosed by comparing the time of emergence of annual-scale warming beyond the baseline decadal-scale variability and the time of emergence of decadal-scale warming beyond the baseline annual-scale variability (Figure 2). Of these two combinations, the emergence of annual-scale warming beyond the baseline decadal-scale variability occurs earlier over all areas of the continental United States (Figure 2), suggesting that the differences between the time of emergence of annual-scale and decadal-scale warming (Figure 1) result primarily from the fact that the baseline decadal-scale variability is substantially smaller than the baseline annual-scale variability.

Figure 2.

As in Figure 1 but for two combinations of time scale of 21st century time series and 21st century baseline variability. (top) The annual departures are calculated from the 10 year running mean of the 21st century annual summer temperature time series, and the baseline variability is calculated as the interannual standard deviation of the 1970–1999 annual summer temperature time series. (bottom) The annual departures are calculated from the 21st century annual summer temperature time series, and the baseline variability is calculated as the interannual standard deviation of the 10 year running mean of the 1970–1999 annual summer temperature time series.

[23] Although the time of emergence of summer warming is latest over the central and southeastern United States (Figure 1), the absolute magnitude of summer warming is also largest over those regions, and smallest over the western and northeastern United States (Figure 3). This pattern of summer warming magnitude develops in the mid 21st century, with summer warming in the 2040–2059 period reaching up to 4.5°C over parts of the central United States, but being restricted to less than 3.0°C over most of the western United States. The pattern of summer warming magnitude intensifies as greenhouse forcing increases through the end of the 21st century, with summer warming exceeding 5.5°C and 6.0°C over parts of the central United States in the 2060–2079 and 2080–2098 periods, respectively, but failing to exceed 4.5°C over most of the western United States.

Figure 3.

Twenty-first century anomalies in summer surface air temperature and precipitation. Ensemble anomalies from the 1970–1999 baseline are calculated as the mean of the anomalies in the five RegCM3 members for the 2020–2039, 2040–2059, 2060–2079, and 2080–2098 periods. The absolute magnitudes of the ensemble anomalies are shown, along with the magnitude of the ensemble anomalies as a fraction of the interannual standard deviation over the 1970–1999 baseline period and the magnitude of the ensemble anomalies as a percent of the mean over the 1970–1999 baseline period.

[24] The differences in the patterns of time of emergence of warming and absolute magnitude of warming result in part from the spatial variations in the baseline annual-scale variability, with the warming that occurs over the central and southeastern United States representing a much smaller fraction of the baseline variability than the warming that occurs over the western and northeastern United States (Figure 3). For instance, over much of the western United States and parts of the northeastern United States, the temperature anomaly in the 2020–2039 period is more than 2 standard deviations of the baseline annual-scale variability (Figure 3). In contrast, over much of the central and southeastern United States, the temperature anomaly in the 2020–2039 period is less than 2 standard deviations of the baseline annual-scale variability (Figure 3). The temperature anomalies over the central and southeastern United States remain smaller as a fraction of the baseline variability than those over the western and northeastern United States throughout the 21st century, substantially delaying the emergence of robust warming (Figure 1) in spite of the fact that those areas exhibit the largest absolute temperature anomalies during the 21st century (Figure 3).

3.2. Changes in Surface Moisture and Energy Balance

[25] The increases in summer temperature over the central United States are associated with decreases in summer precipitation, including decreases of up to 0.75 mm d−1 in the 2020–2039 period (Figure 3). The area of maximum precipitation decrease expands and intensifies as greenhouse forcing increases in the 21st century, including decreases of at least 0.5 mm d−1 over much of the central United States in the 2040–2049 period, and decreases of at least 1.0 mm d−1over much of the central and eastern United States in the 2080–2098 period. These decreases in precipitation over the central United States represent between 20% and 60% of the baseline mean precipitation, and 1.5 standard deviations of the annual-scale baseline precipitation variability. In contrast, summer precipitation is greater over much of the eastern seaboard of the United States during the 21st century than during the 1970–1999 baseline period, including up to 1.5 mm d−1over areas of the northeastern United States in the 2060–2079 and 2080–2098 periods. These increases in precipitation represent up to 40% of the baseline mean precipitation, and up to 2.0 standard deviations of annual-scale baseline variability. Changes in precipitation are limited to less than ±0.5 mm d−1over most of the western United States throughout the 21st century. Although the decreases in precipitation represent greater than 30% of the baseline mean precipitation over much of the western United States (including greater than 60% over areas of the southwestern United States), they represent less than 0.5 standard deviations of the annual-scale baseline precipitation variability over most areas of the southwestern United States throughout the 21st century.

[26] In order to better understand the processes shaping the contrasting patterns of summer warming emergence and summer warming magnitude seen in our high-resolution ensemble, we compare transient 21st century changes in the energy and moisture fluxes over two subregions of the continental United States: (1) the southwestern United States (31.6°N–43.8°N, 123.6°W–105.6°W), which exhibits rapid emergence but relatively moderate mean warming and (2) the central United States (32.0°N–44.0°N, 102.4°W–82.7°W), which exhibits delayed emergence but relatively large mean warming (Figures 1 and 3).

[27] The changes in JJA downward longwave flux are similar over the southwestern and central United States, ranging between 20 W m−2 and 40 W m−2 for most years in the late 21st century (Figures 4 and 5). However, as with JJA surface air temperature and precipitation, the other surface variables show larger JJA changes over the central United States than over the southwestern United States. For example, while late 21st century anomalies in evapotranspiration reach up to −0.7 mm d−1 over the southwestern United States (Figure 4), they reach up to −2.0 mm d−1 over the central United States (Figure 5). Likewise, anomalies in upper layer and root zone soil moisture are limited to less than −2 mm and −50 mm, respectively, over the southwestern United States but exceed −4 mm and −80 mm, respectively, over the central United States. (The land model in RegCM3 includes a total soil depth of 3000 mm, with the upper soil layer encompassing the top 100 mm and the root zone soil layer encompassing the depth of the root zone, which varies by land cover type as described by Dickinson et al. [1993].) Similarly, anomalies in sensible heat flux are limited to between −10 W m−2 and 15 W m−2 over the southwestern United States, but range between −20 W m−2 and 50 W m−2 over the central United States. Further, anomalies in JJA incident shortwave flux, net shortwave flux and net longwave flux are confined to between −20 W m−2 and 12 W m−2 throughout the 21st century over the southwestern United States, but reach between 10 W m−2 and 30 W m−2 over the central United States.

Figure 4.

Twenty-first century anomalies in summer surface air temperature and surface variables over the southwestern United States region (31.6°N–43.8°N, 123.6°W–105.6°W). Ensemble anomalies from the 1970–1999 baseline are calculated for JJA in each year in the five RegCM3 members, meaning that each decade of the 21st century includes 50 simulated years. Correlation coefficients between the surface air temperature anomalies and the respective surface variable anomalies are shown for each decade in the colored columns adjacent to each scatterplot.

Figure 5.

As in Figure 4 but for the central United States region (32.0°N–44.0°N, 102.4°W–82.7°W).

[28] The surface moisture and energy fluxes also show a much stronger correlation with surface air temperature over the central United States than over the southwestern United States (Figures 4 and 5). For example, in comparing the individual simulated years within each simulated decade over the central United States, we find that as temperature increases over the 21st century, JJA precipitation, evapotranspiration, relative humidity, upper layer soil moisture and root zone soil moisture exhibit clear increases in the occurrence of negative anomalies and strong negative correlations with JJA temperature anomalies, with the years in which larger warm anomalies occur also exhibiting larger dry anomalies (Figure 5). Likewise, we find that JJA downward longwave flux, net longwave flux, incident shortwave flux, net shortwave flux, and sensible heat flux exhibit clear increases in the occurrence of positive anomalies as the 21st century progresses, along with strong positive correlations with JJA temperature anomalies (with the years in which larger warm anomalies occur also exhibiting larger surface radiation anomalies). This coupling between anomalies in surface temperature and surface moisture and energy fluxes over the central United States contrasts with what is seen over the southwestern United States, where JJA incident shortwave flux, net shortwave flux and net longwave flux do not show a clear change in the occurrence of positive or negative anomalies over the course of the 21st century (Figure 4), and where anomalies in the surface fluxes are accompanied by much weaker correlations with JJA temperature anomalies (Figure 4).

[29] We note that, over the central United States, the correlations with JJA surface air temperature anomalies are stronger for JJA precipitation and JJA surface soil moisture than for JJA root zone soil moisture in all decades of the 21st century (Figure 5). In contrast, over the southwestern United States, the correlations with JJA surface air temperature anomalies are stronger for JJA root zone soil moisture than for JJA precipitation in the last five decades of the 21st century, and for JJA surface soil moisture in four of the last five decades of the 21st century (Figure 4). These correlations further suggest that the surface warming is more closely linked to the surface drying over the central United States than over the southwestern United States.

[30] In order to better understand the role of the hydrologic cycle in shaping the surface-atmosphere coupling seen during the summer season over the central United States, we compare surface air temperature anomalies that occur during the summer with the temperature, precipitation and soil moisture anomalies that occur during the preceding spring. These comparisons suggest that warm summer temperature anomalies are generally associated with warm spring temperature anomalies, dry spring precipitation anomalies, and dry spring soil moisture anomalies (Figure 6). The JJA temperature anomalies show stronger correlations with MAM precipitation anomalies than with MAM temperature anomalies for eight of the ten decades of the 21st century (Figure 6). In addition, the correlations between MAM precipitation and JJA temperature increase in strength over the 21st century, with each of the first four decades of the 21st century exhibiting correlations weaker than −0.5, and five of the six final decades of the 21st century exhibiting correlations stronger than −0.5. Further, the JJA temperature anomalies are more strongly correlated with MAM precipitation anomalies than with MAM soil moisture anomalies over the last four decades of the 21st century, and partial correlation analysis reveals that correlations between JJA temperature and MAM precipitation are considerably stronger when controlling for MAM root zone soil moisture (−0.54, −0.52, −0.15 and −0.54 for the 2060s, 2070s, 2080s and 2090s, respectively (not shown)) than are correlations between JJA temperature and MAM root zone soil moisture when controlling for MAM precipitation (0.10, −0.03, −0.12 and −0.14 (not shown)). Further, as with the JJA precipitation and soil moisture anomalies, the surface soil moisture correlations are closer to the precipitation correlations than to the root zone soil moisture correlations (Figure 6).

Figure 6.

As in Figure 5 but for respective correlations between summer (JJA) temperature anomalies and spring (MAM) temperature, precipitation, root zone soil moisture, and upper layer soil moisture anomalies.

[31] Similar coupling of warm season surface temperature and moisture has been extensively analyzed in previous work [e.g., Ashfaq et al., 2010a; Chang and Wallace, 1987; Diffenbaugh et al., 2007; Diffenbaugh and Ashfaq, 2010; Eltahir and Pal, 1996; Fischer et al., 2007a; Fischer and Schär, 2009; Hirschi et al., 2011; Hong and Kalnay, 2000; Koster et al., 2004; Lorenz et al., 2010; Manabe et al., 1981; Pal and Eltahir, 2001; Schubert et al., 2004b; Seneviratne et al., 2006, 2010; Wetherald and Manabe, 1995, 1999]. In particular, the central United States has been identified as a region of strong land-atmosphere coupling [Hong and Kalnay, 2000; Koster et al., 2004; Schubert et al., 2004b] with potential vulnerability to drying in response to elevated greenhouse forcing [e.g., Christensen et al., 2007; Dai, 2010; Diffenbaugh and Ashfaq, 2010; Meehl et al., 2007b; Wetherald and Manabe, 1999]. Further, as in our high-resolution ensemble experiment, the CMIP3 AOGCM ensemble exhibits increasing drying over the central United States as greenhouse forcing increases over the course of the 21st century in the A1B scenario, causing protracted “severe drought” conditions that result from changes in surface temperature, moisture and radiation fluxes [Dai, 2010]. The results of the CMIP3 multimodel ensemble therefore support the progressive shift toward warmer, drier conditions with increased net radiation to the surface that is seen over the region of peak warming in our high-resolution ensemble.

3.3. Changes in Atmospheric Circulation

[32] The changes in surface moisture and energy fluxes that occur over the central United States are associated with changes in the atmospheric circulation. As anticipated from the thermal wind balance, the midlatitude jet (zonal winds at 200 mbar) decreases in strength over the central United States over the course of the 21st century (Figure 7). Such a decrease would be expected to lead to a decrease in the occurrence of warm season storms over the central United States [e.g., Hu and Feng, 2010; Trapp et al., 2009]. The decrease in the strength of the midlatitude zonal winds over the central United States is consistent with the poleward shift in midlatitude stormtracks identified by Yin [2005] in the CMIP3 AOGCM ensemble, and with the decrease in cyclone frequency over the United States and northern Great Plains identified by Trapp et al. [2009]in the CAM3 simulations that are used in our high-resolution nested simulations.

Figure 7.

Twenty-first century anomalies in summer atmospheric circulation. Ensemble anomalies from the 1970–1999 baseline are calculated as the mean of the anomalies in the five RegCM3 members for the 2020–2039, 2040–2059, 2060–2079, and 2080–2098 periods. (top) The anomalies in 500 mbar wind vectors and (bottom) the U component of the wind over the central United States (averaged between 102.4°W and 82.7°W).

[33] The 500 mbar winds also suggest a shift toward increased anticyclonic circulation anomalies in the midtroposphere over much of the United States during the 21st century (Figure 7). Anticyclonic anomalies in the midtroposphere have been linked to 20th century hot-and-dry events over the central United States [Chang and Wallace, 1987; Chen and Newman, 1998; Choi and Meentemeyer, 2002; Hong and Kalnay, 2002]. The 500 mbar anticyclonic circulation anomalies that occur over the central United States in our 21st century ensemble experiment are associated with decreases in precipitation, soil moisture, evapotranspiration and relative humidity, along with increases in incident and net surface shortwave flux and surface sensible heat flux (Figure 5 and 7). Further, the correlations with summer temperature anomalies (Figure 5) show that the warmest summer conditions co-occur not only with the driest surface conditions but also with the lowest evapotranspiration, highest shortwave flux to the surface, and highest surface sensible heat flux. The simulated anticyclonic circulation anomalies thus suggest a shift toward more stable atmospheric conditions that increase net surface radiation both by decreasing cloud cover (thereby increasing shortwave input to the surface) and by decreasing precipitation (thereby decreasing evapotranspiration and surface latent heat flux and increasing surface sensible heat flux).

[34] In addition to the changes in circulation in the midtroposphere, the magnitude of JJA planetary boundary layer (PBL) height anomalies increase over the central United States over the course of the 21st century (including anomalies of greater than 300 m in the late 21st century), with larger PBL height anomalies associated with larger surface air temperature anomalies (Figure 5). The strong positive correlations with surface temperature for PBL height and the strong negative correlations with surface temperature for precipitation and soil moisture (Figure 5) are consistent with observational results linking dry soil conditions with a deeper boundary layer and lower total precipitation (relative to wet soil conditions) [Juang et al., 2007].

[35] Although the changes in atmospheric circulation are consistent with warmer, drier conditions at the surface over the central United States, it is not possible to definitively distinguish in our coupled atmosphere-land simulations whether the changes in the atmospheric circulation are the cause of changes in the land surface fluxes, or vice versa. The poleward shift in the midlatitude jet (Figure 7) is expected from a global warming–induced decrease in the pole-to-equator temperature gradient [e.g.,Yin, 2005], suggesting that changes in the large-scale circulation are likely to occur in response to global warming independent of changes in the regional surface energy and moisture fluxes. Alternatively, warming of the surface could cause a shift toward enhanced anticyclonic circulation aloft by warming the atmospheric column below the midtroposphere, thereby increasing the geopotential height in the midtroposphere and creating anomalously anticyclonic circulation. Indeed, dry surface conditions have been shown to enhance this midtropospheric response by enhancing surface warming through changes in the partitioning of surface latent and sensible heating [Fischer et al., 2007b]. Soil moisture anomalies can thus cause changes in the large-scale atmospheric circulation by altering the surface moisture and energy fluxes, with the potential for nonlocal atmospheric changes [Dominguez et al., 2006, 2009; Fischer et al., 2007b; Pal and Eltahir, 2002]. (In the case of our model experiments, because the input from the global climate model occurs at the lateral boundaries of the nested domain, changes in the surface moisture and energy fluxes can influence the large-scale circulation in the interior of our continental-scale domain.) Additional experiments in which the land surface model is uncoupled from the atmospheric model (as in, for example, the work ofSeneviratne et al. [2006]) will help to distinguish the relative roles of the land and atmosphere in regulating the coupled response to elevated greenhouse forcing.

3.4. Climate Model Intercomparison

[36] The time of permanent emergence of warming beyond the baseline variability is an important metric of both the statistical robustness of projected warming and the rapidity with which different areas are projected to move into a new climate space that falls outside of the background variability to which natural and human systems are currently adapted. Because the time of emergence is influenced by both the magnitude of baseline variability and the magnitude of the 21st century warming, the pattern of emergence may contrast with the pattern of warming magnitude. In comparing the southwestern and central U.S. regions, we find that the decadal-scale warming emerges beyond the decadal-scale baseline variability prior to 2020 over most areas of both regions (Figure 1). We also find that although the central United States exhibits a stronger warming trend than the southwestern United States, the transient warming over the central United States is smaller as a fraction of the annual-scale baseline variability (Figure 3), delaying emergence of warming beyond the annual-scale baseline variability (Figure 1).

[37] In order to test the robustness of the pattern of summer warming emergence to variations in model formulation, we compare the baseline variability, 21st century warming, and time of emergence of warming seen in our high-resolution ensemble with those seen the CMIP3 multi AOGCM ensemble (Figure 8). Over the southwestern United States, all of the RegCM3 and CAM3 realizations show permanent emergence of warming beyond the decadal-scale baseline variability during the first decade of the 21st century. Thirty-seven of the 52 CMIP3 realizations likewise show emergence prior to 2010, while the remaining 15 CMIP3 realizations all show emergence prior to 2030. Over the central United States, all of the RegCM3 realizations exhibit permanent emergence beyond the decadal-scale baseline variability prior to 2010, while the CAM3 realizations exhibit emergence ranging from 2010 to 2070. Thirty-two of the 52 CMIP3 realizations show permanent emergence beyond the decadal-scale baseline variability prior to 2010, and 48 show permanent emergence prior to 2030.

Figure 8.

Time of emergence of summer warming beyond the baseline variability for individual realizations of the high-resolution ensemble and the CMIP3 AOGCM ensemble. The annual and decadal timings are calculated as inFigure 1. (left) The emergence values for the southwestern United States (31.6°N–43.8°N, 123.6°W–105.6°W). (right) The emergence values for the central United States (32.0°N–44.0°N, 102.4°W–82.7°W). For these regional emergence calculations, we first create a regional temperature time series and then calculate the time of permanent emergence from that regional series. The horizontal axis is the standard deviation over the 1970–1999 period, with the vertical lines showing the observed value over the region. The vertical axis is the regional temperature trend over the 21st century.

[38] We find substantially greater spread in the time of emergence of warming beyond the baseline annual-scale variability than beyond the baseline decadal-scale variability (Figure 8). The 21st century temperature trend over the southwestern United States is similar in the RegCM3 and CAM3 ensembles. However, the annual-scale baseline variability is smaller, and closer to the observed value, in the RegCM3 ensemble, and the median permanent emergence is earlier in the RegCM3 ensemble (2040s) than in the CAM3 ensemble (2070s). The RegCM3 emergence falls near the middle of the CMIP3 ensemble over the southwestern United States, with 17 of the 52 CMIP3 realizations showing earlier emergence than the earliest RegCM3 realization, and 19 CMIP3 realizations showing later emergence than the latest RegCM3 realization. The RegCM3 21st century warming also falls near the middle of the CMIP3 ensemble, with 19 of the 52 CMIP3 realizations showing greater warming than the RegCM3 realizations, and 14 showing less warming. The CMIP3 realizations that show earlier emergence than the RegCM3 ensemble include all five of the CMIP3 realizations that exhibit lower annual-scale baseline variability and greater 21st century warming than all of RegCM3 realizations, as well as six CMIP3 realizations that exhibit baseline variability and 21st century warming that are similar to those of the RegCM3 ensemble. Likewise, five of the six CMIP3 realizations that show no permanent emergence of summer warming over the southwestern United States exhibit 21st century warming that is weaker than all of the RegCM3 realizations.

[39] RegCM3 and CAM3 both exhibit annual-scale baseline variability that is greater than the observed value over the central United States, and permanent warming does not emerge beyond two standard deviations of the baseline annual-scale variability during the 21st century in any of the RegCM3 or CAM3 realizations (Figure 8). Conversely, 34 of the 52 CMIP3 realizations do show permanent emergence beyond the baseline variability, beginning as early as the third decade of the 21st century. These include all 27 of the realizations with baseline interannual standard deviation of less than 0.9. In addition, nine of the ten CMIP3 realizations that show emergence prior to 2040 also exhibit annual-scale baseline variability that is less than or equal to the observed value. Further, all but 2 of the 21 CMIP3 realizations that exhibit 21st century warming of greater than 4.0 °C/century exhibit emergence prior to the end of the 21st century. However, of the 14 CMIP3 realizations that exhibit baseline variability that is greater than the minimum RegCM3 variability over the central United States, only four show permanent warming emergence during the 21st century.

[40] These comparisons illustrate that there can be substantial spread in the subregional-scale temperature variability and warming trend, and that model variations can influence the projected timing of warming emergence. The CAM3 and RegCM3 biases in baseline summer temperature variability clearly influence the time of emergence of robust warming over the central United States (Figure 8). Given the importance of large-scale ocean-atmosphere interactions for summer climate over the continental United States [e.g.,Barlow et al., 2001; Gershunov and Cayan, 2003; Higgins and Shi, 2000; Hong and Kalnay, 2000; McCabe et al., 2004; Schubert et al., 2004a, 2004b], it is likely that errors in the GCM representation of large-scale atmosphere-ocean variability contribute to the biases in annual-scale baseline temperature variability. CCSM3 is able to capture the pattern and magnitude of tropical Pacific and tropical Atlantic SSTs [e.g.,Ashfaq et al., 2010b; Deser et al., 2006], and the summer storm tracks simulated by CCSM3 over the North Pacific, the North Atlantic and North America are similar to those seen in reanalysis data [Alexander, 2006]. However, biases in the CCSM3-simulated tropical SSTs do cause errors in the warm season precipitation over the central United States [Ashfaq et al., 2010b]. Further, although CCSM3 exhibits ENSO amplitude and Pacific decadal-scale variability (and associated atmospheric responses) that are similar to those seen in observations [Alexander, 2006; Deser et al., 2006], the frequency of ENSO is too strong relative to observations, while tropical Atlantic variability is too weak [Deser et al., 2006].

4. Summary and Conclusions

[41] We report results from the first transient, high-resolution, century-scale, multimember ensemble climate model experiment for the full continental United States. Given the intense decision maker demand for predictions of climate change on decadal time scales and subregional spatial scales, the spread in future transient climate evolution seen within this high-resolution ensemble has important implications for climate-informed decision making. Although most areas of the continental United States exhibit emergence of warming beyond the decadal-scale baseline variability prior to 2020, emergence of warming beyond the annual-scale baseline variability exhibits much greater spatial heterogeneity. In particular, over areas that exhibit relatively little annual-scale variability in the baseline climate, such as the southwestern United States, our results suggest that near-term global warming is likely to cause robust subregional-scale warming that exceeds the baseline variability as early as within the next two decades. In contrast, where there is greater annual-scale variability in the baseline climate dynamics, the transient warming signal is less robust, including over the central United States, which exhibits the largest absolute magnitude of 21st century summer warming. Comparisons with observations and the CMIP3 ensemble of global climate model experiments suggest that this reduced subregional robustness is influenced both by the baseline variability of the real climate system, and by errors in the climate models that create excessive annual-scale variability of warm season temperature. Increased availability of high-resolution, century-scale, ensemble climate model experiments will help to elucidate the interactions between temperature variability and warming magnitude that are likely to shape the emergence of subregional-scale warming in response to transient increases in global radiative forcing.

[42] In addition, it is important to consider that other climate phenomena beyond those considered here can also be important for natural and human systems, and that those phenomena could exhibit varying emergence beyond the baseline variability. For example, analysis of the CMIP3 ensemble suggests that the fact that the baseline temperature variability is greater in winter than in summer is likely to delay 21st century emergence of a completely novel winter temperature regime over the midlatitudes, including over the continental United States [Diffenbaugh and Scherer, 2011]. Likewise, the signal-to-noise ratio of projected 21st century change in the CMIP3 ensemble is much smaller for regional precipitation change than for regional temperature change [Hawkins and Sutton, 2011], suggesting, as does our Figure 3, that the time of emergence beyond the baseline variability is likely to be later for precipitation than for temperature.

[43] However, some climatic changes could have important impacts even if they do not permanently exceed the baseline variability to which systems are currently accustomed. For instance, for systems in which relatively subtle variations in phenology and temperature tolerance can have important impacts, global warming projected over the near-term decades could be sufficient to require adaptation to temperature changes that occur both during the summer and during other parts of the growing season and/or during the cold season [Diffenbaugh et al., 2011]. Likewise, cold season warming in the western United States could cause substantial decreases in snow-to-precipitation ratio and acceleration of declines in spring snowpack prior to the mid 21st century, even if the cold season temperature change does not exceed the baseline annual-scale variability (M. Ashfaq et al., Near-term acceleration of hydroclimatic change in the western U.S., submitted toJournal of Geophysical Research, 2011). These examples provide motivation for extending our time of emergence analysis not only to a broader suite of high-resolution climate model experiments, but also to a broader suite of climate phenomena.

Acknowledgments

[44] We thank the three anonymous reviewers for their insightful and constructive comments. The CAM3 and RegCM3 simulations were generated, stored, and analyzed using computing resources provided by the Rosen Center for Advanced Computing (RCAC) at Purdue University and analyzed using computing resources provided by the Center for Computational Earth and Environmental Science (CEES) at Stanford University. We thank the NCAR CCSM3 Climate Change Working Group for access to the CCSM3 simulations at NCAR. We thank the modeling groups, WCRP's WGCM, PCMDI, and the U.S. DOE for enabling the CMIP3 archive. This work was supported by NSF awards 0541491, 0756624, and 0955283; NIH award S0183091; and DOE awards DE-FG02-08ER64649 and DE-SC0001483, along with DOE award ERKP777.