1. Introduction
Annual drying, increased aridity, and decreased streamflow have been projected for the southwestern United States (SWUS) and northwestern Mexico (MX) because of increased greenhouse gas forcing in analyses focusing on phases 3 and 5 of the the Coupled Model Intercomparison Project (CMIP3 and CMIP5, respectively) global climate model (GCM) simulations (e.g., Milly et al. 2005; Christensen et al. 2007; Hoerling and Eischeid 2007; Seager et al. 2007; Seth et al. 2013; Cook and Seager 2013). Statements about precipitation associated specifically with the North American monsoon (NAM) season for these regions are more uncertain, however. In the CMIP3 suite of GCMs, a decrease in summertime-mean precipitation was projected for the SWUS and northwestern MX, but model agreement on that projection was weak (Christensen et al. 2007, their Fig. 11.12). In the CMIP5 ensemble, decreases in monsoon season rainfall are small and insignificant overall, given a shift in the season to less early season rainfall (June–July) and more late season rainfall (September–October) due to local and remote processes creating a more unfavorable early season convective environment (Seth et al. 2011; Cook and Seager 2013; Seth et al. 2013; Torres-Alavez et al. 2014). The CMIP3 and CMIP5 models, however, have problems simulating precipitation in this region. During monsoon season, performance is mixed, with reasonable precipitation in some models but a complete lack of a monsoon in others. Late monsoon season termination is a widespread and common problem also, and the annual cycle is usually too wet, particularly in winter (Lin et al. 2008; Dominguez et al. 2010; Cook and Seager 2013; Geil et al. 2013; Torres-Alavez et al. 2014).
Uncertainty in precipitation projections for the NAM is high partly because of the dependence of the system on dynamics and finescale orography that are not well resolved by many models, particularly at typical global model scales. For example, the representation of mountains and their effects on moisture convergence has been shown to be important in projections of precipitation in this region (Gao et al. 2012). Near-surface flow and sea surface temperatures (SSTs) over the Gulf of California (GoC) are also important and not resolved at coarse resolutions (Mitchell et al. 2002; Collier and Zhang 2007; Lee et al. 2007). Note, however, that higher-resolution models that can resolve features like the GoC also do not always produce a proficient simulation of these features (e.g., Gutzler et al. 2009; Bukovsky et al. 2013). Geil et al. (2013) found no major differences in model performance between higher- and lower-resolution members in the CMIP5 ensemble for this region, where the models ranged from about 0.57° to 3.76° in latitude–longitude resolution. In that case, even the highest-resolution model was determined to be too coarse to capture smaller-scale orographically driven processes. At a resolution near that of the highest-resolution models in CMIP5 (50 km), however, Bukovsky et al. (2013) showed that some regional models can produce some of the terrain forcing and mesoscale features important to a good representation of the NAM. Perhaps this difference is due to the use of parameterizations that are adjusted for region-specific use and not generalized for global use. Castro et al. (2007a,b, 2012) have also demonstrated the potential of regional models to improve forecasts of the NAM system. Therefore, in an attempt to overcome some of the uncertainty resulting from resolution, in this study we will present precipitation projections from the set of 50-km-resolution dynamically downscaled simulations produced as a part of the North American Regional Climate Change Assessment Program (NARCCAP; Mearns et al. 2012).
This study builds off of Bukovsky et al. (2013, hereinafter BUK13), where it was shown that many of the NARCCAP regional climate models (RCMs) do reasonably simulate the NAM system and its topographically influenced mesoscale features when forced with a reanalysis product, within the limits of their given resolution. However, most of the RCMs undergo a major reduction of skill when forced by GCMs in the baseline climate scenario because of the biases they inherit from the GCMs. In BUK13, some of the identified inherited biases include atmospheric moisture content, which led to huge dry biases and no monsoon precipitation signal in some of the RCMs; warm SST biases in the GCMs, which led to more realistic (warmer) GoC SSTs in the RCMs when interpolated into GoC grid boxes in the RCMs; and large-scale circulation errors that, at a minimum, caused problems in the timing and magnitude of the monsoon. In the RCMs, biases related to the still too coarse resolution for many of the NAM system features were also identified and were shown to be RCM specific and not dependent on the driver. For example, while the RCMs provided a good terrain-driven spatial pattern of precipitation in the region for their resolution, not all of them were able to simulate a reasonable northward GoC low-level jet (LLJ), which likely contributed to those models’ low precipitation biases in Arizona (AZ).
In this study, we identify further biases in the baseline climate and discuss how these biases and those presented in BUK13 may affect the projections of NAM precipitation for the future. We also identify processes responsible for the changes in precipitation projected by the RCMs. It is this deeper analysis of the simulations that then allows us to assess the differential credibility of the RCMs.
2. Models, methods, and datasets
a. Models
Six RCMs were used to downscale four GCMs to 50 km as a part of NARCCAP. Results from all 12 of the planned combinations are included in this study. The 12 combinations were chosen using a fractional factorial design to sample the matrix of possibilities in a statistically meaningful way [see Mearns et al. (2012) for further information]. Table 1 provides an overview of the RCMs and GCMs; Table 2 presents the RCM and GCM simulation combinations. When referring to an RCM and its parent GCM, we list the forcing simulation in lowercase (e.g., WRFG–ccsm); otherwise, all acronyms are in uppercase. The ECP2-hadcm simulations were not available for use in BUK13; therefore, some precipitation verification information complementary to that in BUK13 has been included in Figs. S1 and S2 of the supplementary material.
RCMs and GCMs used in NARCCAP, their identifying acronyms (RCM acronyms are as used in the NARCCAP model archive), and relevant references. For the GCMs, horizontal resolution and CMIP3 archive ensemble member number are also listed.
NARCCAP RCM and GCM simulations. All planned combinations (all of which are used here) are marked with an X.
All future simulations utilize the Special Report on Emissions Scenarios (SRES; Nakićenović et al. 2000) A2 emissions scenario; the twentieth-century (20c3m) emission representation is used for the baseline period. All baseline simulations span 1971–99, whereas the future simulations span 2041–69. All averages herein are performed over these specified years.
The NAM region our analysis focuses on is defined in Fig. 1. This also includes two specific subregions over AZ and northwestern MX. Note that there is some variation in the size and placement of these regions in each model owing to differences in their map projections and the southward extent of their domains. Most of the RCM domains do not extend very far south of the Baja California peninsula; thus, for some consistency, but to include as much of the domain as possible, the southern edge of the analysis region is defined to be as close to 20°N as possible. In ensemble-mean plots, however, the largest common domain is used instead.
Surface elevation (m) over land from the HRM3. Ocean points are filled in blue. Names and location indicators for important topographic features indicated with white text and lines. Outlines for analysis subregions in AZ and northwestern MX are outlined in magenta. Large NAM “core” region covers the full area shown. Locations for vertical cross sections (shown later in the analysis) across the GoC and through AZ are indicated by thick black lines. Note that the southern extent of each RCM varies, and this impacts the size of the NAM core analysis region. Most NARCCAP RCM domains end around the southern tip of the Baja California peninsula.
Citation: Journal of Climate 28, 17; 10.1175/JCLI-D-14-00695.1
The core of the monsoon season is the target period of our analysis. We use a July–August (JA) average instead of the traditional June–August (JJA) or June–September (JJAS) average because of the challenge CMIP3 GCMs have in simulating monsoon onset and retreat (e.g., Geil et al. 2013). These GCM characteristics are transferred to some of the NARCCAP simulations, as documented in BUK13.
b. Verification datasets
Two reanalysis datasets are used briefly in model verification. They are the National Centers for Environmental Prediction (NCEP)–U.S. Department of Energy (DOE) AMIP-II reanalysis (hereinafter simply NCEP; Kanamitsu et al. 2002) and the North American Regional Reanalysis (NARR; Mesinger et al. 2006). NCEP was also used to force the NARCCAP RCMs and is used here only in an examination of 500-hPa winds and geopotential heights. NARR was compared to several other observationally based datasets and the NARCCAP NCEP- and GCM-forced simulations in BUK13 during the NAM season and over the same regions used herein. Its precipitation is used again in this complementary study for consistency and because it was found that the spread in the models is considerably larger than the spread in the observationally based datasets.
c. Statistical methods
1) Significance testing
Unless otherwise noted, statistical significance of the changes between the baseline and future climate is tested at the 0.1 level using bootstrapping with bias correction and acceleration following von Storch and Zwiers (1999) and Efron and Tibshirani (1993), as described in Bukovsky and Karoly (2011). For the changes in wind, the change in the magnitude of the north–south component of the wind and the change in the east–west component of the wind are tested separately for significance. If either is significant, the change in the vector wind at that point is considered significant.
2) Analysis of variance calculations
The statistical analysis that will be presented in Table 3 is done in three steps. First, we tested the hypothesis that the average rainfall, rainfall intensity, and fraction of dry days in the NARCCAP model runs have mean values equal to the corresponding NARR values using two-sided, one-sample Student’s t tests. Second, we assessed whether the differences between future and baseline average rainfall, rainfall intensity, and fraction of dry days in the NARCCAP model runs were zero. We used two-sided pairwise Student’s t tests, where a pair consists of the future and past values from the same model. Third, we tested the hypothesis that the means of the differences between baseline and future for average rainfall, rainfall intensity, and fraction of dry days differ as a function of the driving GCMs. In the case of a significant difference, we conducted a multiple-comparison procedure to identify which pairs are different. A multiple-comparison procedure adjusts for the fact that the chance of incorrectly finding a significant difference increases with the number of comparisons when comparing individual pairs and instead provides an upper bound on the probability that any comparison will be incorrectly found significant. We conducted all analyses separately for the NAM region and the two subregions, AZ and MX.
The JA future-minus-current difference in average precipitation (avg; percent and mm day−1), precipitation intensity (int; %), and the number of dry days (DD; %) for the entire analysis region over land only and the AZ and MX subregions (as shown in Fig. 1). The ensemble averages for each statistic are given in the last five rows for the full ensemble (average), and then for subensembles of models grouped by forcing GCM. In the average row only, boldface values indicate a statistically significant change from the baseline to the future, and italicized values indicate a strong bias in the baseline value [see section 2c(2) for details]. The values followed by an exclamation point indicate where mean differences between the baseline and future climate differ significantly as a function of the driving GCMs, and these are explained further in the text.
3) Analysis of wet and dry years
Wet and dry years are defined using the JA season mean for each year averaged across all land-based grid points in the full NAM region. Wet years are defined as years that exceed a standardized precipitation index (SPI; McKee et al. 1993) value of 1.0 and dry years as those that fall below −1.0 SPI. This is calculated separately for the baseline (1971–99) and future (2041–69) periods. Most simulations have at least five wet and five dry years using this definition. Quantiles for wet and dry years are then calculated from JA daily total precipitation values that have been averaged across the NAM region (for land points only). The quantiles are calculated for separate cumulative distribution functions from the baseline and future periods. As such, we can present the change in the magnitude for daily precipitation at the 0.90 quantile, for example, during wet or dry years (as will be done in Tables 4 and 5). For the change in frequency, in the 0.25 quantile, the frequency is measured as the number of days with precipitation less than or equal to the 0.25 quantile threshold. For other quantiles, the frequency is the number of days greater than or equal to the given threshold.
The JA percent change in the magnitude and frequency of daily precipitation at the given quantiles for wet years only between the baseline and the future. The upper table shows the change in magnitude, where each column represents a specific quantile threshold. The lower table shows the frequency change in daily precipitation defined for the quantiles in the upper table. The ensemble average for each quantile is given in the last five rows for the full 12-simulation ensemble (average) and then for subensembles grouped by forcing GCM.
3. Precipitation projections
As illustrated in Fig. 2, the 12-RCM mean projects a decrease in JA average precipitation across the region. While most of the changes in precipitation are within the bounds of natural variability in the majority of the models, there is strong agreement on the sign of the change in much of the region, particularly in southwestern AZ and northwestern MX. The majority of the models agree that changes are significant and decreasing, as indicated by the hatching in Fig. 2, in the central plains and at a few locations along the west coast of Mexico and the Baja California peninsula.
Average JA precipitation change (%) from the baseline period for the 12-model ensemble mean. Precipitation is presented following methodology proposed by Tebaldi et al. (2011), with slight modification: hatching indicates where more than 50% of the models show change that is significant at the 0.10 level (as determined by a Student’s t test) and where more than 75% of the models agree on the sign of change (thus, where the majority of the models agree on significance and sign). White grid cells indicate where more than 50% of the models show change that is significant but also where 75% of the models or less agree on the sign of the change (thus indicating true disagreement and little information). Additionally, the number of models that agree on the sign of the change is indicated by the color saturation and value (the vertical axis on the color bar). To facilitate creating this ensemble average, all models were regridded to a common 0.5° × 0.5° latitude–longitude grid.
Citation: Journal of Climate 28, 17; 10.1175/JCLI-D-14-00695.1
The ensemble mean, however, does not capture the large variability in magnitude and spatial distribution of the precipitation projections across the 12 simulations (Fig. 3). There are some broad similarities across RCMs that have the same parent GCM, but among those, one still finds substantial variation across RCMs when the details are examined. As the spatial distribution of convection in this region is governed largely by local orography, one might expect the pattern of change to be influenced by the orography as well, but that is not clearly the case. For example, the Sierra Madre Occidental (SMO) and Mogollon Rim appear to influence the pattern of change in some of the RCMs (more or less precipitation on one side or the other), but the change is not consistent across the models.
The JA average precipitation change (%) from the baseline period. Hatching indicates where the change is statistically significant at the 0.1 level. Each row is organized by parent GCM, with the change from the GCM in the left panels.
Citation: Journal of Climate 28, 17; 10.1175/JCLI-D-14-00695.1
Generally, the CCSM- and CGCM-forced simulations project less future precipitation for most of the region, whereas the signal for change is more mixed in the GFDL- and HADCM-driven simulations. In the CCSM-forced simulations, this is opposite to what the CCSM GCM projects. This is not the only region where RCMs forced with this CCSM simulation produce precipitation projections that are contrary to what the CCSM has, nor are these the only RCM simulations that do this (Bukovsky and Karoly 2011; Mearns et al. 2013).
To aid in the interpretation of these precipitation projections, we present projection summaries for average precipitation, precipitation intensity, and the number of dry days (DD) in Table 3. The values are averaged over the full NAM region, and the values for the AZ and MX subregions in Fig. 1 are averaged over land only. The upper part of Table 3 shows the individual model values, and the lower part shows the means for both the full model ensemble and the GCM-driven subensembles. Overall, most models indicate a small increase in the number of dry days over the full region and over AZ, with a larger increase over MX. The number of dry-day projections is more consistent in magnitude and sign across the full ensemble and the subensembles than for other precipitation metrics, and in the full-ensemble mean, this change is significantly different from zero. However, the number of dry days is also significantly biased relative to NARR. For mean precipitation change, it is clear here, as in Fig. 3, that the CCSM- and CGCM-forced simulations produce the greatest decreases. The same is true for projected decreases in intensity. Regarding intensity, the projections from the CCSM- and CGCM-driven ensembles are significantly different from the HADCM-driven ensemble in the NAM region (as indicated by the values followed by an exclamation point in Table 3). Similarly, in MX, both the HADCM- and GFDL-driven intensity projections are significantly different than the CCSM- and CGCM-driven ensembles. There are no significant differences in projections between the GCM-driven subensembles for any other variable or region, however.
Within the GFDL-driven group in Table 3, projections are less in agreement over all regions, particularly with regard to the HRM3–gfdl and the ECP2–gfdl. The same is true, to some extent, in the two HADCM-forced simulations. The HRM3–hadcm simulates a strong percent decrease in precipitation average and intensity over AZ and an increase in the number of dry days, with a slight increase in average and intensity over MX. This is in disagreement with the MM5I–hadcm, which projects little change in AZ and a stronger decrease in average and intensity in MX.
The change in the frequency of 3-hourly precipitation rates or events during JA is illustrated in Fig. 4 for the MX and AZ subregions. All but one of the RCMs simulate decreases in the frequency of events of nearly every magnitude; the ECP2 simulations are outliers. The decrease in frequency is strongest in the CCSM- and CGCM-driven simulations. The decrease is smaller, though not always insignificant, in the other simulations. Half the simulations project an increase in the frequency of events that are classified at or above the 99.9th percentile in the baseline period, including the three GFDL-driven simulations, the CRCM–ccsm, and two others depending on the region. These increases range from around 7% to 212%.
Percent change from the baseline period to the future in the frequency of 3-hourly precipitation rates in JA for (a) AZ and (b) MX subregions. Rates are binned according to their percentiles in the baseline climate. The given number associated with a bin is the starting point for values within that bin; for example, the blue 90th-percentile bin examines the change in the frequency of events with a magnitude greater than or equal to the 90th-percentile magnitude and less than the 95th-percentile magnitude from the current climate period. A dark block under a given bin at the bottom of each panel indicates that the change in that bin is statistically significant at the 0.1 level. Note that the 99.9th-percentile bars are cut off in the ECP2 simulations to make the smaller changes more visible. The changes are 212% (GFDL driven) and 155% (HADCM driven).
Citation: Journal of Climate 28, 17; 10.1175/JCLI-D-14-00695.1
Precipitation magnitude and frequency projections for wet and dry years only are shown in Tables 4 and 5, respectively. In the full-ensemble mean, in Tables 4 and 5, there is a decrease in the magnitude of total daily precipitation in all quantiles except those at the 99th percentile and above in both wet and dry years. The HADCM-driven ensemble is a relative outlier, with increases in the magnitude of all quantiles in both wet and dry years. However, the more interesting results occur in the 0.99 quantile for wet years and the 0.25 quantile for dry years. In wet years, an increase in magnitude of about 5% and an almost 90% increase in the frequency of days with precipitation totals at or above the 99th percentile is seen in the full-ensemble mean. Likewise, in three of the four subensembles these heavy precipitation days also become more frequent and more intense during wet years (the CGCM-driven ensemble is the outlier), though there are differences among individual simulations. In dry years (Table 5), days with low precipitation totals (at the 0.25 quantile or less) become more frequent and even less intense in all but one RCM (both ECP2 simulations). In the ensemble mean, there is around 26% less precipitation on these days and they occur about 34% more often. These results simultaneously suggest that extreme wet years get wetter and extreme dry years get drier. Results here are also consistent with those above in that in the CCSM-driven and CGCM-driven simulations, the extreme wet years projection is suppressed compared to the other subensembles, and the extreme dry years projection is enhanced; that is, the models that have the driest bias have drier projections.
Overall, combining these precipitation projections with the analysis of BUK13, we find that the RCM simulations that have the greatest biases in precipitation during the monsoon season also have some of the greatest decreases in future precipitation total, intensity, and frequency. This is emphasized in Fig. 5 for average JA precipitation. The simulations that are most biased in Fig. 5 are not consistently biased throughout the year, as is shown in BUK13 (their Figs. 11–13). In fact, the CCSM-driven simulations, which have no signal for the monsoon in their annual precipitation climatology, are strongly wet-biased during the cool season. While the cool season wet bias is inherited from the CCSM GCM, the lack of monsoon season precipitation is not directly inherited (the CCSM does have monsoon-related precipitation); this is due to a combination of inherited errors that inhibit precipitation in the RCMs only. The CGCM-driven simulations perform well during the rest of the year, as does the CGCM GCM, but the RCMs are too dry during monsoon season (with the exception of the CRCM–cgcm, which is not as heavily biased in JA). This, again, is due to a combination of inherited errors that prohibit RCM precipitation only; the CGCM does not contain this bias.
The JA average precipitation change (%) from the baseline to the future period vs the precipitation bias (%). Bias is defined as the models’ baseline period average (1971–99) simulation minus NARR (1980–2003). Values are the average of land points only over the NAM core region. The linear fit applied to the points does not include the driving GCM results (open black symbols).
Citation: Journal of Climate 28, 17; 10.1175/JCLI-D-14-00695.1
4. Understanding the precipitation projections
In this section, we examine the processes driving the precipitation projections. The aim is to determine if the projected precipitation change is reasonable and credible, despite the known biases in the baseline simulations.
a. The CCSM-driven simulations
All CCSM-driven simulations project an increase in low-level specific humidity over the GoC and in AZ (shown for AZ only in Figs. 6c–e; color fill indicates the change in specific humidity). This is likely due in part to increasing temperature (not shown), but the low-level increase in AZ may also be as a result of changes in the GoC LLJ. The GoC LLJ is an important NAM feature as it transports tropical Pacific moisture into northwestern Mexico and the SWUS. It runs northward along the GoC, parallel to its axis. It is confined to the boundary layer, below about 850 hPa, with a core around 250–400 m above the surface (Douglas 1995; Mo et al. 2005). The CRCM and the WRFG do generate a GoC LLJ in the NCEP-driven and baseline GCM-driven runs, although it is weak in the WRFG GCM-driven runs (Figs. 7c,e; baseline mean in color fill, northward is blue; BUK13, their Figs. 8 and 17). Both of these also increase the strength of their northward flow in the future (Figs. 7c,e; solid magenta lines indicate an increase in northward flow). The MM5I does not have a mean GoC LLJ (Fig. 7d), and flow becomes slightly more southward in the future (Fig. 7, dashed magenta lines indicate an increase in southward flow). This difference explains why the increase in specific humidity in the MM5I is not as strong in AZ as in the CRCM and WRFG (Figs. 6c–e, color fill). It also supports the more uniformly negative precipitation change across AZ in the MM5I compared to the CRCM and WRFG and the small, insignificant increase on the windward side of the Mogollon Rim in the latter two. In the CRCM–ccsm and WRFG–ccsm, the changes in moisture and local flow alone imply potential for an increase in NAM system precipitation in the future in AZ and MX. However, all of these simulations start out with a strong low bias in specific humidity (BUK13), inherited from the CCSM, and the projected increase in humidity is not enough to compensate for the starting bias, particularly in the MM5I and WRFG; that is, relative to historical period observations, the future simulation would still be biased dry despite the increase in humidity.
Vertical cross sections through AZ (location shown in the bottom-right panel with orientation) of JA average change from the baseline to the future climate in specific humidity (g kg−1; color fill) and flow parallel to the cross section (thick black vectors; reference vector in bottom right of left panels). JA baseline-mean specific humidity (magenta contours, interval of 1 g kg−1) and flow parallel to the cross section (light gray vectors) are shown for reference. Panels are organized to maximize the number of related simulations that are adjacent. Rows are organized by GCM driver, and RCMs are positioned as indicated in the key in the top-right corner. Note that vertical motion has been multiplied by a factor of 1000 for visibility and that the vertical wind component is not available and not shown for the HRM3–hadcm or both RCM3 simulations.
Citation: Journal of Climate 28, 17; 10.1175/JCLI-D-14-00695.1
Vertical cross sections through the GoC (location shown in the bottom-right panel with orientation) of JA average change from the baseline to the future climate in flow parallel to (thick black vectors; reference vector in bottom right of left panels) and orthogonal to (magenta contours, interval of 0.25 m s−1; negative contours are dashed and zero contour is thickened) the cross sections. JA baseline-mean flow parallel to the cross section (light gray vectors) and baseline-mean flow orthogonal to the cross section (color fill) are shown for reference. Positive orthogonal values are approximately northward and into the page, as indicated in the bottom-right panel of the figure. Panels are organized to maximize the number of related simulations that are adjacent to one another. Rows are organized by GCM driver, and RCMs are positioned as indicated in the key in the top-right corner. Gray areas indicates topography. Note that vertical motion has been multiplied by a factor of 1000 for visibility and that the vertical wind component is not available and not shown for the HRM3–hadcm or both RCM3 simulations.
Citation: Journal of Climate 28, 17; 10.1175/JCLI-D-14-00695.1
Compounding the humidity bias, and at least partly explaining the strong decrease in precipitation projected for the future regionally, are the upper-level monsoon anticyclone and changes in larger-scale flow. The monsoon anticyclone, or “high,” exists in the middle-to-upper troposphere and is caused by enhanced heating of the atmosphere over land, particularly the elevated terrain in MX and the western United States (e.g., Higgins et al. 1997, 1999). During June, the monsoon high migrates north along the west coast of MX, coincident with the northward migration of precipitation along the SMO to the SWUS in the beginning of July. This migration of the monsoon high leads to a wind shift that favors landward moisture transport. The mean location and strength of the monsoon anticyclone and its corresponding wind field in July and August at 500 hPa in NCEP is given in Fig. 8 by the solid gray circle and gray vectors. The convection associated with the monsoon also modulates the strength of the high through latent heating, with increased precipitation leading to a stronger and broader anticyclone (e.g., Higgins et al. 1998; Stensrud 2013).
The JA average location and strength of the 500-hPa monsoon anticyclone center in the baseline (filled circle) and future (open circle) for GCM and GCM-driven simulations for (a) CCSM, (b) CGCM, (c) GFDL, and (d) HADCM. The size of the filled and open circles represents the magnitude of the maximum geopotential height in the anticyclone center, following the key on the right. This includes NCEP, the gray-filled circle in all panels, at 5931 m, the central circle size. Vectors indicate the JA average baseline flow at 500 hPa from all simulations and NCEP. Note that geopotential height is not available from the RCM3; therefore, the geopotential height of the anticyclone center in this figure for RCM3 only is set to that of NCEP for the current and future, and the location of the anticyclone center is taken as the center of the circulation in the 500-hPa wind field instead of as the maximum in the 500-hPa geopotential height field. Also, except for the 500-hPa geopotential height field, no other upper-level information is available from the ECP2 at the time of writing; therefore, no wind vectors are plotted for these simulations.
Citation: Journal of Climate 28, 17; 10.1175/JCLI-D-14-00695.1
In the RCMs, the strength and location of the monsoon anticyclone and the corresponding flow are generally inherited from the GCMs. In the CCSM-driven simulations, this results in the high being too far south (over northwestern MX instead of northeastern New Mexico) and too strong (Fig. 8a). There is also an additional, separate cutoff anticyclonic circulation branch in the south-central United States in a few of the simulations (which do show up in Fig. 8a, as they are distinct, closed-circulation height maximums). The CCSM, the MM5I–ccsm, and the WRFG–ccsm, specifically, contain the greatest location errors. In reality, a southward displacement like that seen in these simulations is usually associated with dry monsoon years in the SWUS, as this position is not favorable for moisture flux convergence in the SWUS (e.g., Higgins et al. 1998; Higgins and Shi 2000; Johnson et al. 2007). This location bias is clearly inherited by the MM5I–ccsm and the WRFG–ccsm, leading to flow into the SWUS that is not as southerly, is less tropical in origin, and has a stronger fetch from the Pacific Ocean. This likely explains the stronger low-level dry bias in AZ compared to the CRCM–ccsm (Figs. 6c–e). The magnitude bias in the monsoon high is also clearly inherited (the initial cause of which remains undiagnosed), as well as the high bias in the westerlies (because of a north–south pressure/height gradient that is too strong; not shown). It is interesting that the CRCM–ccsm is the least incorrect here, with the center of the circulation as the center of one very elongated anticyclone that stretches from central AZ into Arkansas, as this is the only one of these three RCMs to include nudging (weakly, at 500 hPa and above). This would generally make it more likely to match the large-scale pattern from the CCSM.
In the future, while the monsoon anticyclone does not substantially change its mean position (Fig. 8; cf. open and filled circles), the continental-scale geopotential height pattern changes such that there is an increase in flow that is continental in origin above 900 hPa over the SWUS and northwestern MX. This is illustrated at 700 hPa in Fig. 9 (and in Figs. 6c–e over AZ). In the CRCM and WRFG, this is associated with a well-defined future-minus-current anticyclonic flow anomaly at 700 hPa centered near northern AZ or over Nevada (depending on the RCM; Fig. 9) or at 500 hPa centered near or just northeast of the Great Salt Lake (not shown). There is also a stronger inverted trough at 700 hPa east of the SMO (Fig. 9).
The (a) CCSM and (b)–(d) CCSM-driven RCMs for JA from 1971–99 to 2041–69 average change in 700-hPa wind speed (m s−1) and direction [1 m s−1 reference vector inset at bottom right in (d)]. Light gray shading indicates that the change is significant at the 0.1 level.
Citation: Journal of Climate 28, 17; 10.1175/JCLI-D-14-00695.1
The future flow anomaly resembling a stronger inverted trough is similar to an anomaly that would precede inverted trough/tropical easterly wave (TEW) passage and often-associated gulf surges1 (Schiffer and Nesbitt 2012). However, given that there is little-to-no TEW activity in this version of the CCSM (McCrary et al. 2014), at least as far as TEWs originating over Africa are concerned, this is unlikely to be associated with a future change in TEW activity. The increased strength of the inverted trough to the east of the SMO in the RCMs is likely forced by an incidental change in the CCSM and is only seen in the RCMs because it is inherited. It is not a propagating feature. In the future, in the CCSM, the southward flow on the eastern side of the monsoon anticyclone increases and the northerly flow on the western side of the south-central U.S. anticyclone center (which does not exist in observations) increases. This gives the false sense of a stronger inverted trough in the anomaly field between the two anticyclonic centers (Fig. 9; the cyclonic anomaly centered over far western Texas in Fig. 9a). Furthermore, in July and August, precipitation is forced, most evenings, on the east and southeast slope of the coarse-resolution terrain that represents the Rocky Mountains and the eastern slope of the SMO related to the Gulf of Mexico (GoM) LLJ in the CCSM (not shown). This precipitation significantly increases in the future (Fig. 3a), unlike the already strong monsoon-related precipitation on the western slope of the SMO in MX, and may be increasing the strength of or causing the anomalous cyclonic circulation.
The anomalous future-minus-current anticyclonic flow in the Rocky Mountains, which forces an increase in continental flow in the SWUS in the future and the decrease in precipitation, is likely tied to El Niño in this case. Castro et al. (2001) showed that anomalous ridging over the Rocky Mountains in late June and early July, associated with the positive phase of the Pacific transition (PT) pattern, is significantly and strongly correlated with a negative/cool SST anomaly over the Niño-3 region. According to Meehl and Arblaster (2002), June–September mean negative SST anomalies in the central-to-eastern equatorial Pacific follow December–February mean positive/warm SST anomalies that are usually associated with El Niño. It follows then that a positive winter SST anomaly (El Niño) would be associated with the PT height anomaly pattern at the beginning of the NAM season seen in Castro et al. (2001). As the CCSM does project a shift to more El Niño–like conditions in the future (van Oldenborgh et al. 2005), it is possible that the PT-like flow anomaly we observe is forced by this shift. However, the CCSM also has a poor representation of El Niño–Southern Oscillation (ENSO) variability to start (too frequent and too weak; van Oldenborgh et al. 2005), and the RCMs forced by it do not well simulate ENSO-related variability of monsoon precipitation as a result (Carrillo et al. 2015, manuscript submitted to Int. J. Climatol.), leaving confidence in these projections even lower. Note that while this particular change appears to be forced by an increasingly frequent El Niño, and we consider ENSO an important factor to consider in the simulations of the recent past, the ENSO teleconnection with NAM precipitation in general is somewhat uncertain as it is not stable over long periods of time (Griffin et al. 2013).
In the MM5I–ccsm, the pattern of midlevel change is different (Fig. 9c). It produces anomalous cyclonic flow in its 700-hPa projected difference, centered over east Texas. The reason for the divergence of this simulation from its driver, to a much greater extent than in the WRFG and CRCM projections, may be related to the “drift” that occurs in all MM5I simulations (and only the MM5I simulations); that is, the MM5I has a warming bias that causes it to slowly depart from its driver, any driver, rather linearly over the course of a run at all levels. In the MM5I–ccsm simulations, this leads to about a 1.14 m yr−1 increase in 500-hPa geopotential heights over the CCSM during the baseline simulation and a 0.86 m yr−1 increase over the CCSM in the future, averaged over the full model domain. For additional discussion of this bias, see Bukovsky (2012).
While changes in the ingredients necessary for convection in the CCSM-driven simulations are somewhat mixed, the starting, inherited biases in these simulations that lead to little precipitation in the current period, particularly the biases in the monsoon anticyclone and specific humidity, likely lead to unrealistic and unreliable decreases in precipitation amount, frequency, and intensity over the region. Confidence in the changes in future upper-level, larger-scale flow that would at least support a decrease in precipitation of an unknown magnitude over the SWUS is also low, as the CCSM is one of many CMIP3 GCMs with a poor simulation of ENSO variability (Collins et al. 2005), and the NAM system precipitation change in the CCSM-driven simulations appears to be related to a questionable increase in El Niño frequency. Overall, all of these biased starting conditions act to strongly inhibit convection in the current period, to the extent that there is no monsoon precipitation signal in the annual cycle of precipitation in these models, and these biased conditions become stronger in the future, leading to an unreliably large decrease in midcentury precipitation.
b. The CGCM-driven simulations
Current-to-future differences in wind show a strong anticyclonic circulation anomaly off of the west coast of MX (Fig. 10). This could suggest a decrease in the strength or frequency of the inverted trough off the west coast of MX, supporting the decreases in specific humidity over the GoC (not shown) and precipitation (Figs. 3e–h). The trough is associated with tropical easterly waves or, on occasion, tropical cyclones. It acts to transport moisture into the NAM region. Unlike the CCSM, this version of the CGCM does simulate easterly wave activity that is similar to that seen in reanalysis over Africa, at least, so TEW forcing may be included here (Skinner and Diffenbaugh 2013). This circulation change anomaly is approximately collocated with a reduced/cooler SST warming signal that occurs just west of the tip of the Baja California peninsula in these simulations (not shown). This future change in flow leads to a lesser increase in specific humidity in the region during the season compared to other simulations (e.g., Fig. 6 for AZ), as flow is generally less southerly. Low-level flow in the GoC also becomes less southerly in the RCM3 and WRFG; however, the flow associated with the GoC LLJ in the CRCM (which has a good representation of the LLJ to start) becomes more northerly (Figs. 7f–h). Combined, these changes drive decreases in precipitation in the CRCM and WRFG simulations in the SWUS and MX (Figs. 3f,g). The RCM3–cgcm does not have as widespread a strong precipitation decrease as the other CGCM-driven simulations (Fig. 3h).
As in Fig. 9, but for the CGCM and CGCM-driven RCMs.
Citation: Journal of Climate 28, 17; 10.1175/JCLI-D-14-00695.1
The RCM3, however, is one of two RCMs that typically have a large bias in precipitation intensity (the other being ECP2), as illustrated in Fig. 11. To better explain this intensity bias and its potential effect on the precipitation projections, particularly for wet and dry monsoon years, we present a Hovmöller diagram in Fig. 12 for one dry and one wet year (as defined in section 3) in the baseline and future simulations of the RCM3–cgcm and the CRCM–cgcm. In 1988, the wet RCM3 produces intense precipitation, but events have a short duration (Fig. 12); however, the dry CRCM produces less precipitation, but convection persists longer as it propagates westward over time, which is consistent with the observed propagation of precipitation in this region (e.g., Gochis et al. 2007; Lang et al. 2007; Nesbitt et al. 2008). In wet and dry years, the timing and frequency of events is similar between RCMs because they have the same parent GCM; however, in Fig. 12, whether it is a dry or wet monsoon year, current or future, the RCM3 precipitates more heavily during any individual event than the CRCM. Therefore, the less widespread and smaller decreases in mean precipitation in the RCM3 versus the CRCM (or WRFG) are likely the result of these differing changes in the intensity of individual events along with the changes in intensity and frequency seen in section 3.
The JA average change in precipitation intensity from the baseline to the future period vs the precipitation intensity bias (mm day−1). Bias is defined as a model’s baseline period average (1971–99) minus NARR (1980–2003). Values are the average over (a) the NAM core region land points only and (b) AZ and (c) MX subregions as defined in Fig. 1.
Citation: Journal of Climate 28, 17; 10.1175/JCLI-D-14-00695.1
Hovmöller diagrams of daily precipitation (mm day−1) from the RCM3–cgcm (higher precipitation intensity RCM) and CRCM–cgcm (lower precipitation intensity RCM) for an extreme dry year (1981, baseline; 2053, future) and wet year (1988, baseline; 2060, future) in the baseline and future simulations during June–August. Precipitation is averaged over 30°–37.5°N.
Citation: Journal of Climate 28, 17; 10.1175/JCLI-D-14-00695.1
The larger-scale changes in the CGCM-forced simulations do imply less precipitation, but the magnitude of the precipitation projections is still questionable, as these simulations have some of the same basic problems as the CCSM-forced simulations and may also lead to a deceptively large decrease in precipitation. The CGCM-driven simulations also place the mean location of the anticyclone too far south, and its position does not change much in the future. This location is not ideal for good moisture transport in the SWUS, as in section 4a, and it is made even less ideal by a plausibly realistic change in flow in the future (the anticyclonic flow anomaly described above). The CGCM-driven simulations, like the CCSM-driven simulations, also start with a large dry bias in low-level specific humidity, which also causes a large low precipitation bias (BUK13). Warmer and slightly drier conditions, plus the other changes discussed above, contribute overall to an environment that is even less favorable for convection, as seen in the decrease in the frequency of precipitation of nearly all magnitudes (Fig. 4), and particularly for convective initiation. For example, convective inhibition (CIN) in JA monthly mean profiles for 1981–99 and 2051–69,2 near Los Mochis, MX, in the CRCM–cgcm projection increases from 483 to 578 J kg−1, a 19% increase. In the CRCM-ncep simulation, however, mean CIN is 241 J kg−1 at this location, and if we simply apply the mean temperature and specific humidity changes from the CGCM-driven simulation to the CRCM-ncep profile (as in the “delta” method, but not with the climate change applied to observations), this increases to 358 J kg−1. The latter is a larger increase in CIN, from current to future, but the delta-method-like future value is still lower than the value from the CRCM–cgcm in the baseline period by 125 J kg−1 and would be easier to overcome with any given level of forcing.
Additionally, the pattern of more anticyclonic flow west of the Baja California peninsula seen in the future is correlated in observations with El Niño years and a positive North Pacific Oscillation, as shown in Castro et al. (2001). However, while the CGCM is similar to the CCSM in that it has an ENSO cycle that is too frequent and has too weak an amplitude, it instead produces a more La Niña–like state in the future, like other low-resolution GCMs (van Oldenborgh et al. 2005). Little confidence is assigned to this future projection of more La Niña–like conditions because of the poor ENSO simulations in the GCMs that project it (van Oldenborgh et al. 2005), and here it is contrary to this pattern of change, implying that there is another cause for this flow anomaly. Last, an increase in La Niña–like years would imply more favorable conditions (e.g., enhanced GoC LLJ) and more precipitation (Castro et al. 2001), not seen here.
While it is impossible to say what the magnitude of the precipitation decrease would be if the possibly plausible larger-scale changes in flow from the CGCM-driven simulations were applied to more realistic starting conditions, it is likely that the decreases projected in these simulations are not representative of those values. Furthermore, while a decrease in precipitation would likely occur given the change in flow, the biases existing in these simulations may be leading to a greater decrease in precipitation than if the larger-scale changes were applied to nonbiased starting conditions, as the convective environment is bad to start and only becomes worse in the future.
c. The GFDL-driven simulations
The GFDL-driven simulations’ main problem, caused by the GFDL GCM, which forces an incredibly excessive amount of precipitation in the NAM region from September through December, only starts to appear as a problem in August (BUK13). It is unclear what effect this bias has on projections for the core of the monsoon season. However, Carrillo et al. (2015, manuscript submitted to Int. J. Climatol.) found that this misrepresentation of the NAM-region annual cycle may cause a poor representation of the spatial variability of JA precipitation at a continental scale associated with ENSO and PDO. Relative to the CCSM- and CGCM-driven simulations, most of the GFDL-driven simulations do not have a large bias in the magnitude or location of the monsoon high (Fig. 8c), except the ECP2–gfdl, in which the anticyclone is too weak.3 The GFDL-driven simulations also do not have other, precipitation-exterminating biases in their driving fields that they inherit during JA. However, the parent GCM is known to have very weak TEW activity (Skinner and Diffenbaugh 2013), which likely contributes at least to the dry AZ precipitation biases seen in the RCMs. The other known problems in JA are largely tied to the RCMs. HRM3, for example, is the only simulation of the three that reasonably reproduces the GoC LLJ, and it projects a decrease in northward flow (Fig. 7i). The RCM3 does not produce a LLJ in this simulation (Fig. 7j) and does not produce a good signal for the monsoon in AZ precipitation as a result (Fig. 12 in BUK13). There is little-to-no upper-level information from the ECP2–gfdl simulation available; however, it might be assumed that it also does not have the GoC LLJ, since the ECP2 does not produce one when driven with NCEP, and this feature remains fairly consistent in quality in the other RCMs when driven with various GCMs. The HRM3 is also the only simulation of the three that does not have a high bias in the intensity of the precipitation it produces. The RCM3 and, especially, the ECP2 do (Fig. 11). Moreover, it is possible that this intensity bias is contributing to the increase in precipitation seen in the ECP2–gfdl, particularly in the SWUS, where its intensity is most biased to start and it projects the greatest increase in the future (Fig. 11). Unfortunately, it is not possible to further examine the ECP2–gfdl to see what is driving the relatively large precipitation increases owing to the unavailability of many of its output fields. The intensity bias might also be contributing to the precipitation projections from the RCM3, as when forced with the CGCM. Without this intensity bias, it is possible that the areas where less precipitation is projected would be drier and that the increases would be weaker, given the same changes in frequency.
The decreases in future precipitation seen in the HRM3, however, are warranted, given the changes in circulation and its lack of a large precipitation intensity bias. The small decreases in southerly flow (see Figs. 7i and 6i), particularly over the GoC, help explain the small but significant decreases in precipitation in this simulation. The GFDL as well as the HADCM discussed in the next section do not have significant future changes in ENSO, or significant problems in simulating it, as in the CGCM and CCSM (van Oldenborgh et al. 2005); therefore, the RCMs they force do not lose credibility from this point of view.
d. The HADCM-driven simulations
The HADCM-driven simulations inherit fewer biases from their parent global model than the rest (BUK13). They contain realistic NAM system precipitation during the NAM season, and although the RCMs inherit an early onset problem from the HADCM, this bias is much less fatal to the precipitation simulations than what is seen in the other GCM-driven simulations (BUK13). The HADCM also contains reasonable African TEW activity (not shown). This version of the HADCM is also one of two models in the CMIP3 suite that was found to most realistically represent ENSO variability [see Dominguez et al. (2010) and van Oldenborgh et al. (2005), although these analyses did not focus on the realization used for the NARCCAP simulations]. However, despite having the same parent GCM and fewer initial biases, there are noticeable differences in the precipitation projections between the HADCM-forced simulations (Figs. 3n–p), particularly in AZ and the Four Corners region as well as near the west coast of MX and the GoC. This is partly due to differences in how mid- to upper-level flow evolves in the future (e.g., Fig. 13). The mean position of the monsoon anticyclone in the MM5I–hadcm is good, relative to many of the other simulations, and it does strengthen and shift slightly northeast in the future, closer to the correct position in the baseline climate, as illustrated in Fig. 8d. The overall change in mid- to upper-level flow in the MM5I–hadcm in the future is that of an anomalous cyclone centered over west-central Texas (Fig. 13c). This causes a switch to predominantly northeast flow near the west coast of MX in the future in the MM5I, explaining the decreased precipitation there. However, the reason for this peculiar larger-scale change, which is quite different from what the parent GCM does, may be related to the unrealistic drift in the MM5I, as discussed in section 4a. Also, the MM5I–hadcm has no GoC LLJ, and it projects an increase in northerly flow over the southern half of the GoC (Fig. 7b) and a slight increase in onshore flow into AZ (Fig. 6b). The latter may be due to an increase in the strength of the daily sea breeze resulting from an increase in the land–sea temperature contrast.
As in Fig. 9, but for the HADCM and HADCM-driven RCMs.
Citation: Journal of Climate 28, 17; 10.1175/JCLI-D-14-00695.1
In the HRM3–hadcm simulation, the precipitation decrease centered on the Four Corners region is likely due to the increase in mid- to upper-level northerly flow (Fig. 13b). The change in the winds resembles an anomalous inverted ridge that covers the western half of the United States, with a ridge axis running along the U.S. West Coast and thus leading to enhanced northerly flow over the western half of the United States. Increased moisture and an increase in low-level southerly flow over the northern half of the GoC are not enough to counter the increased, unfavorable flow aloft, leading to the decrease in precipitation in the Four Corners region. The change in flow from the ECP2–hadcm simulation is assumed to be similar to that in the HRM3–hadcm simulation, as changes in the 500-hPa height field are similar in pattern (not shown; also, other upper-level fields from the ECP2 are not available). Despite this similarity, precipitation increases are much more widespread in the ECP2 than in the HRM3, and decreases are less widespread in the Four Corners region (Figs. 3n,o). This may be due in part to the precipitation intensity bias seen in the ECP2 (Fig. 11), as discussed with the ECP2–gfdl simulation. With the ECP2 GFDL- and HADCM-driven simulations, it is interesting that the precipitation change fields more closely resemble each other than the other RCMs with common GCM drivers; that is, the changes in Figs. 3j and 3n more closely resemble each other than those in Figs. 3k and 3l or Figs. 3o and 3p. This suggests that this RCM may have more control over the changes in precipitation than the other RCMs (even if just through an intensity bias).
5. Discussion and conclusions
While model agreement sometimes leads to increased confidence, it can also be irrelevant and potentially misleading, as our examination demonstrates. We have shown that the NARCCAP ensemble projects decreased mean precipitation and less frequent precipitation during the NAM season in the SWUS and northwestern MX with good agreement. However, after an in-depth analysis of the NAM system in the 12 NARCCAP RCMs, we find that the ensemble-mean precipitation projection lacks credibility. Some of the more important features analyzed, and their contribution to our conclusion on credibility, are summarized in Table 6. Combining this study with results from BUK13, we find that some of the most credible simulations, regarding their baseline performance and their projections, are the HADCM-driven simulations and the HRM3 simulations (including the implied overlap). These four simulations also obtain the highest numbers of positive scores in Table 6. However, the HRM3–gfdl contains the undiagnosed effect of the GFDL “extended” monsoon season, the MM5I–hadcm contains the similarly undiagnosed effect of the MM5I “drift,” and the ECP2–hadcm has a strong precipitation intensity bias, leaving the HRM3–hadcm as the most credible simulation in the set. This one simulation projects small but significant decreases in mean precipitation during the core of the NAM season across the SWUS, small increases in the number of dry days regionally, and an increase in the frequency of the heaviest precipitation events with a decrease in the frequency of precipitation of lesser intensities (Figs. 3 and 4; Table 3).
Question: Is the specific feature well enough represented such that it contributes to the credibility of the final precipitation projection? The more “yes” answers, the more credible the simulation.
The WRFG–cgcm and CRCM–cgcm simulations could be considered “runners up” behind the previously described simulations, but they have biases inherited from the CGCM that cause their projections to be much more questionable. Given that the WRFG and CRCM perform well when forced with NCEP for the NAM system, it would be ideal to complete simulations where they are forced with a less biased set of GCMs (e.g., HADCM), but this is outside the scope of this study and the planned set of NARCCAP simulations. Here, the value added by the WRFG and CRCM to their coarse-resolution drivers is the addition of finer-scale forcing and appropriate mesoscale features (e.g., local orography like the Mogollon Rim and GoC and RCM-developed circulations like the GoC LLJ), but this is eclipsed by the problems caused by the biased boundary conditions from the CGCM (and the CCSM).
The poorest simulation is the MM5I–ccsm (Table 6). Note that this simulation includes the large-scale disadvantages of the CCSM (which leads to the lowest average positive responses in all of the RCMs it forces in Table 6) along with the relatively poor performance of the MM5I regarding subregional-scale phenomena (e.g., the GoC LLJ). Certainly we discourage the use of the MM5I–ccsm results in this region for, say, an impacts analysis.
It is important to note that while our more credible simulations generally produced a smaller signal for a decrease in mean NAM precipitation amount by midcentury, this would not necessarily preclude drying in the region, as temperatures are also projected to rise, and soil moisture evaporation would increase. Exploring this effect is beyond the scope of this work, however.
The effect of the GCM bias on our RCM simulations encapsulates the well-known “garbage in, garbage out” effect (e.g., Rummukainen 2010), and it governs four of the six specifically named features in Table 6. This can be used to argue that a GCM cannot be too skillful for further downscaling [contrary to a statement in Shindell et al. (2014) that GCMs “should not be too skillful… or there will be little opportunity for added value”] and that the careful selection of GCMs for downscaling is warranted. However, picking a “good” GCM for downscaling is clearly not a straightforward task, particularly for large, diverse regions, like the NARCCAP domain.
It has been noted in numerous publications that it is difficult to evaluate GCM and RCM simulations in order to either eliminate ensemble members (of too poor quality) or differentially weight them for the sake of coming up with more robust estimates of future climate on regional scales (e.g., Gleckler et al. 2008; Knutti et al. 2010; Flato et al. 2013). This problem persists, and we would likely have a difficult time determining if some of these NARCCAP simulations should be used for any purpose over this region aside from general research on model results. Yet we do believe we have made headway in applying regional, process-based methods to evaluate the quality of future projections (Barsugli et al. 2013). We note that the research described here, albeit very scientifically rewarding, has also been very time consuming. However, we believe that such intensive investigations are required to make definitive progress in determining the credibility of climate simulations.
We have determined both the best and the worst simulations, but it is still difficult to make specific recommendations for the use of a single best model; what could be done with just one credible simulation remains problematic. While the NARCCAP matrix was based on a specific statistical design, still different combinations could have been used that would have been consistent with that design (e.g. WRFG-hadcm). This might have resulted in a greater number of higher-quality simulations. An additional approach may be to undertake an expert elicitation exercise to assemble a wider range of expert opinions on the quality of the simulations (e.g., Mearns et al. 2015, manuscript submitted to Bull. Amer. Meteor. Soc.).
Finally, we hope to take what we have learned in this work with NARCCAP and some of the CMIP3 GCMs and expand on it in the near future with the CMIP5 and Coordinated Regional Climate Downscaling Experiment (CORDEX) ensembles.
Acknowledgments
We wish to thank NARCCAP for providing the data used in this paper. NARCCAP is funded by the National Science Foundation, the U.S. Department of Energy, the National Oceanic and Atmospheric Administration (NOAA), and the U.S. Environmental Protection Agency Office of Research and Development. We would also like to thank the entire NARCCAP modeling team for useful discussions regarding this work, as well as the two reviewers who helped improve this manuscript. The authors also acknowledge the support of the NOAA Climate Program Office Modeling, Analysis, Predictions, and Projections (MAPP) Program. Work was supported under Grant NA11AOR4310111.
REFERENCES
Anderson, J. L., and Coauthors, 2004: The new GFDL global atmosphere and land model AM2–LM2: Evaluation with prescribed SST simulations. J. Climate, 17, 4641–4673, doi:10.1175/JCLI-3223.1.
Barsugli, J. J., and Coauthors, 2013: The practitioner’s dilemma: How to assess the credibility of downscaled climate projections. Eos, Trans. Amer. Geophys. Union, 94, 424–425, doi:10.1002/2013EO460005.
Bukovsky, M. S., 2012: Temperature trends in the NARCCAP regional climate models. J. Climate, 25, 3985–3991, doi:10.1175/JCLI-D-11-00588.1.
Bukovsky, M. S., and D. J. Karoly, 2011: A regional modeling study of climate change impacts on warm-season precipitation in the central United States. J. Climate, 24, 1985–2002, doi:10.1175/2010JCLI3447.1.
Bukovsky, M. S., D. J. Gochis, and L. O. Mearns, 2013: Towards assessing NARCCAP regional climate model credibility for the North American monsoon: Current climate simulations. J. Climate, 26, 8802–8826, doi:10.1175/JCLI-D-12-00538.1.
Castro, C. L., T. B. McKee, and R. A. Pielke, 2001: The relationship of the North American monsoon to tropical and North Pacific sea surface temperatures as revealed by observational analyses. J. Climate, 14, 4449–4473, doi:10.1175/1520-0442(2001)014<4449:TROTNA>2.0.CO;2.
Castro, C. L., R. A. Pielke, and J. O. Adegoke, 2007a: Investigation of the summer climate of the contiguous United States and Mexico using the regional atmospheric modeling system (RAMS). Part I: Model climatology (1950–2002). J. Climate, 20, 3844–3865, doi:10.1175/JCLI4211.1.
Castro, C. L., R. A. Pielke, J. O. Adegoke, S. D. Schubert, and P. J. Pegion, 2007b: Investigation of the summer climate of the contiguous United States and Mexico using the regional atmospheric modeling system (RAMS). Part II: Model climate variability. J. Climate, 20, 3866–3887, doi:10.1175/JCLI4212.1.
Castro, C. L., H. Chang, F. Dominguez, C. Carrillo, J. K. Schemm, and H. M. H. Juang, 2012: Can a regional climate model improve the ability to forecast the North American monsoon? J. Climate, 25, 8212–8237, doi:10.1175/JCLI-D-11-00441.1.
Caya, D., and R. Laprise, 1999: A semi-implicit semi-Lagrangian regional climate model: The Canadian RCM. Mon. Wea. Rev., 127, 341–362, doi:10.1175/1520-0493(1999)127<0341:ASISLR>2.0.CO;2.
Christensen, J. H., and Coauthors, 2007: Regional climate projections. Climate Change 2007: The Physical Science Basis, S. Solomon et al., Eds., Cambridge University Press, 847–940. [Available online at http://www.ipcc.ch/pdf/assessment-report/ar4/wg1/ar4-wg1-chapter11.pdf.]
Collier, J. C., and G. J. Zhang, 2007: Effects of increased horizontal resolution on simulation of the North American monsoon in the NCAR CAM3: An evaluation based on surface, satellite, and reanalysis data. J. Climate, 20, 1843–1861, doi:10.1175/JCLI4099.1.
Collins, M., and Coauthors, 2005: El Niño- or La Niña-like climate change? Climate Dyn., 24, 89–104, doi:10.1007/s00382-004-0478-x.
Collins, W. D., and Coauthors, 2006: The Community Climate System Model version 3 (CCSM3). J. Climate, 19, 2122–2143, doi:10.1175/JCLI3761.1.
Cook, B. I., and R. Seager, 2013: The response of the North American monsoon to increased greenhouse gas forcing. J. Geophys. Res. Atmos., 118, 1690–1699, doi:10.1002/jgrd.50111.
Dominguez, F., J. Cañon, and J. Valdes, 2010: IPCC-AR4 climate simulations for the southwestern US: The importance of future ENSO projections. Climatic Change, 99, 499–514, doi:10.1007/s10584-009-9672-5.
Douglas, M., 1995: The summertime low-level jet over the Gulf of California. Mon. Wea. Rev., 123, 2334–2347, doi:10.1175/1520-0493(1995)123<2334:TSLLJO>2.0.CO;2.
Efron, B., and R. Tibshirani, 1993: An Introduction to the Bootstrap. Chapman and Hall/CRC, 450 pp.
Flato, G. M., G. J. Boer, W. G. Lee, N. A. McFarlane, D. Ramsden, M. C. Reader, and A. J. Weaver, 2000: The Canadian Centre for Climate Modelling and Analysis global coupled model and its climate. Climate Dyn., 16, 451–467, doi:10.1007/s003820050339.
Flato, G., and Coauthors, 2013: Evaluation of climate models. Climate Change 2013: The Physical Science Basis, T. F. Stocker, et al., Eds., Cambridge University Press, 741–866. [Available online at http://www.ipcc.ch/pdf/assessment-report/ar5/wg1/WG1AR5_Chapter09_FINAL.pdf.]
Gao, Y., L. R. Leung, E. P. Salathé, F. Dominguez, B. Nijssen, and D. P. Lettenmaier, 2012: Moisture flux convergence in regional and global climate models: Implications for droughts in the southwestern United States under climate change. Geophys. Res. Lett., 39, L09711, doi:10.1029/2012GL051560.
Geil, K. L., Y. L. Serra, and X. Zeng, 2013: Assessment of CMIP5 model simulations of the North American monsoon system. J. Climate, 26, 8787–8801, doi:10.1175/JCLI-D-13-00044.1.
Giorgi, F., M. R. Marinucci, and G. T. Bates, 1993a: Development of a second-generation Regional Climate Model (RegCM2). Part I: Boundary-layer and radiative transfer processes. Mon. Wea. Rev., 121, 2794–2813, doi:10.1175/1520-0493(1993)121<2794:DOASGR>2.0.CO;2.
Giorgi, F., M. R. Marinucci, G. T. Bates, and G. de Canio, 1993b: Development of a second-generation Regional Climate Model (RegCM2). Part II: Convective processes and assimilation of lateral boundary conditions. Mon. Wea. Rev., 121, 2814–2832, doi:10.1175/1520-0493(1993)121<2814:DOASGR>2.0.CO;2.
Gleckler, P. J., K. E. Taylor, and C. Doutriaux, 2008: Performance metrics for climate models. J. Geophys. Res., 113, D06104, doi:10.1029/2007JD008972.
Gochis, D. J., C. J. Watts, J. Garatuza-Payan, and J. C. Rodriguez, 2007: Spatial and temporal patterns of precipitation intensity as observed by the NAME event rain gauge network from 2002 to 2004. J. Climate, 20, 1734–1750, doi:10.1175/JCLI4092.1.
Gordon, C., C. Cooper, C. A. Senior, H. Banks, J. M. Gregory, T. C. Johns, J. F. B. Mitchell, and R. A. Wood, 2000: The simulation of SST, sea ice extents and ocean heat transports in a version of the Hadley Centre coupled model without flux adjustments. Climate Dyn., 16, 147–168, doi:10.1007/s003820050010.
Grell, G. A., J. Dudhia, and D. R. Stauffer, 1993: A description of the fifth-generation Penn State/NCAR Mesoscale Model (MM5). NCAR Tech. Note NCAR/TN-398+STR, 121 pp. [Available online at http://nldr.library.ucar.edu/repository/assets/technotes/TECH-NOTE-000-000-000-214.pdf.]
Griffin, D., and Coauthors, 2013: North American monsoon precipitation reconstructed from tree-ring latewood. Geophys. Res. Lett., 40, 954–958, doi:10.1002/grl.50184.
Gutzler, D. S., and Coauthors, 2009: Simulations of the 2004 North American monsoon: NAMAP2. J. Climate, 22, 6716–6740, doi:10.1175/2009JCLI3138.1.
Hales, J. E., Jr., 1972: Surges of maritime tropical air northward over the Gulf of California. Mon. Wea. Rev., 100, 298–306, doi:10.1175/1520-0493(1972)100<0298:SOMTAN>2.3.CO;2.
Higgins, R. W., and W. Shi, 2000: Dominant factors responsible for interannual variability of the summer monsoon in the southwestern United States. J. Climate, 13, 759–776, doi:10.1175/1520-0442(2000)013<0759:DFRFIV>2.0.CO;2.
Higgins, R. W., Y. Yao, and X. L. Wang, 1997: Influence of the North American monsoon system on the U.S. summer precipitation regime. J. Climate, 10, 2600–2622, doi:10.1175/1520-0442(1997)010<2600:IOTNAM>2.0.CO;2.
Higgins, R. W., K. C. Mo, and Y. Yao, 1998: Interannual variability of the U.S. summer precipitation regime with emphasis on the southwestern monsoon. J. Climate, 11, 2582–2606, doi:10.1175/1520-0442(1998)011<2582:IVOTUS>2.0.CO;2.
Higgins, R. W., Y. Chen, and A. V. Douglas, 1999: Interannual variability of the North American warm season precipitation regime. J. Climate, 12, 653–680, doi:10.1175/1520-0442(1999)012<0653:IVOTNA>2.0.CO;2.
Hoerling, M., and J. Eischeid, 2007: Past peak water in the Southwest. Southwest Hydrol., 18, 18–19.
Johnson, R. H., P. E. Ciesielski, B. D. McNoldy, P. J. Rogers, and R. K. Taft, 2007: Multiscale variability of the flow during the North American monsoon experiment. J. Climate, 20, 1628–1648, doi:10.1175/JCLI4087.1.
Jones, R. G., D. C. Hassell, D. Hudson, S. S. Wilson, G. J. Jenkins, and J. F. B. Mitchell, 2003: Workbook on generating high resolution climate change scenarios using PRECIS. UNDP/Met Office Hadley Centre Rep., 32 pp. [Available online at http://www.uncclearn.org/sites/www.uncclearn.org/files/inventory/UNDP17.pdf.]
Juang, H. M., S. Y. Hong, and M. Kanamitsu, 1997: The NCEP regional spectral model: An update. Bull. Amer. Meteor. Soc., 78, 2125–2143, doi:10.1175/1520-0477(1997)078<2125:TNRSMA>2.0.CO;2.
Kanamitsu, M., W. Ebisuzaki, J. Woollen, S. K. Yang, J. J. Hnilo, M. Fiorino, and G. L. Potter, 2002: NCEP–DOE AMIP-II Reanalysis (R-2). Bull. Amer. Meteor. Soc., 83, 1631–1643, doi:10.1175/BAMS-83-11-1631.
Knutti, R., G. Abramowitz, M. Collins, V. Eyring, P. J. Gleckler, B. Hewitson, and L. Mearns, 2010: Good practice guidance paper on assessing and combining multi model climate projections. IPCC Rep., 13 pp. [Available online at https://www.ipcc-wg1.unibe.ch/guidancepaper/IPCC_EM_MME_GoodPracticeGuidancePaper.pdf.]
Lang, T. J., D. A. Ahijevych, S. W. Nesbitt, R. E. Carbone, S. A. Rutledge, and R. Cifelli, 2007: Radar-observed characteristics of precipitating systems during NAME 2004. J. Climate, 20, 1713–1733, doi:10.1175/JCLI4082.1.
Lee, M. I., and Coauthors, 2007: Sensitivity to horizontal resolution in the AGCM simulations of warm season diurnal cycle of precipitation over the United States and northern Mexico. J. Climate, 20, 1862–1881, doi:10.1175/JCLI4090.1.
Lin, J. L., B. E. Mapes, K. M. Weickmann, G. N. Kiladis, S. D. Schubert, M. J. Suarez, J. T. Bacmeister, and M. I. Lee, 2008: North American monsoon and convectively coupled equatorial waves simulated by IPCC AR4 coupled GCMs. J. Climate, 21, 2919–2937, doi:10.1175/2007JCLI1815.1.
McCrary, R. R., D. A. Randall, and C. Stan, 2014: Simulations of the West African monsoon with a superparameterized climate model. Part 2: African easterly waves. J. Climate, 27, 8323–8341, doi:10.1175/JCLI-D-13-00677.1.
McKee, T. B., N. J. Doeskin, and J. Kleist, 1993: The relationship of drought frequency and duration to time scales. Proc. Eighth Conf. on Applied Climatology, Boston, MA, Amer. Meteor. Soc., 179–184.
Mearns, L. O., W. J. Gutowski, R. Jones, R. Leung, S. McGinnis, A. M. B. Nunes, and Y. Qian, 2009: A regional climate change assessment program for North America. Eos, Trans. Amer. Geophys. Union, 90, 311, doi:10.1029/2009EO360002.
Mearns, L. O., and Coauthors, 2012: The North American Regional Climate Change Assessment Program: Overview of phase I results. Bull. Amer. Meteor. Soc., 93, 1337–1362, doi:10.1175/BAMS-D-11-00223.1.
Mearns, L. O., and Coauthors, 2013: Climate change projections of the North American Regional Climate Change Assessment Program (NARCCAP). Climatic Change, 120, 965–975, doi:10.1007/s10584-013-0831-3.
Meehl, G. A., and J. M. Arblaster, 2002: The tropospheric biennial oscillation and Asian–Australian monsoon rainfall. J. Climate, 15, 722–744, doi:10.1175/1520-0442(2002)015<0722:TTBOAA>2.0.CO;2.
Mesinger, F., and Coauthors, 2006: North American Regional Reanalysis. Bull. Amer. Meteor. Soc., 87, 343–360, doi:10.1175/BAMS-87-3-343.
Milly, P. C. D., K. A. Dunne, and A. V. Vecchia, 2005: Global pattern of trends in streamflow and water availability in a changing climate. Nature, 438, 347–350, doi:10.1038/nature04312.
Mitchell, D. L., D. Ivanova, R. Rabin, T. J. Brown, and K. Redmond, 2002: Gulf of California sea surface temperatures and the North American monsoon: Mechanistic implications from observations. J. Climate, 15, 2261–2281, doi:10.1175/1520-0442(2002)015<2261:GOCSST>2.0.CO;2.
Mo, K. C., M. Chelliah, M. L. Carrera, R. W. Higgins, and W. Ebisuzaki, 2005: Atmospheric moisture transport over the United States and Mexico as evaluated in the NCEP Regional Reanalysis. J. Hydrometor., 6, 710–728, doi:10.1175/JHM452.1.
Nakićenović, N., and Coauthors, 2000: Special Report on Emissions Scenarios. Cambridge University Press, 599 pp. [Available online at https://www.ipcc.ch/pdf/special-reports/emissions_scenarios.pdf.]
Nesbitt, S. W., D. J. Gochis, and T. J. Lang, 2008: The diurnal cycle of clouds and precipitation along the Sierra Madre Occidental observed during NAME-2004: Implications for warm season precipitation estimation in complex terrain. J. Hydrometeor., 9, 728–743, doi:10.1175/2008JHM939.1.
Pal, J. S., and Coauthors, 2007: Regional climate modeling for the developing world: The ICTP RegCM3 and RegCNET. Bull. Amer. Meteor. Soc., 88, 1395–1409, doi:10.1175/BAMS-88-9-1395.
Pope, V. D., M. L. Gallani, P. R. Rowntree, and R. A. Stratton, 2000: The impact of new physical parametrizations in the Hadley Centre climate model: HadAM3. Climate Dyn., 16, 123–146, doi:10.1007/s003820050009.
Rummukainen, M., 2010: State-of-the-art with regional climate models. Wiley Interdiscip. Rev.: Climate Change, 1, 82–96, doi:10.1002/wcc.8.
Schiffer, N. J., and S. W. Nesbitt, 2012: Flow, moisture, and thermodynamic variability associated with Gulf of California surges within the North American monsoon. J. Climate, 25, 4220–4241, doi:10.1175/JCLI-D-11-00266.1.
Seager, R., and Coauthors, 2007: Model projections of an imminent transition to a more arid climate in southwestern North America. Science, 316, 1181–1184, doi:10.1126/science.1139601.
Seth, A., S. A. Rauscher, M. Rojas, A. Giannini, and S. J. Camargo, 2011: Enhanced spring convective barrier for monsoons in a warmer world? Climatic Change, 104, 403–414, doi:10.1007/s10584-010-9973-8.
Seth, A., S. A. Rauscher, M. Biasutti, A. Giannini, S. J. Camargo, and M. Rojas, 2013: CMIP5 projected changes in the annual cycle of precipitation in monsoon regions. J. Climate, 26, 7328–7351, doi:10.1175/JCLI-D-12-00726.1.
Shindell, D., P. Racherla, and G. Milly, 2014: Reply to comment by Laprise on “The added value to global model projections of climate change by dynamical downscaling: A case study over the continental U.S. using the GISS-ModelE2 and WRF models.” J. Geophys. Res., 119, 3882–3885, doi:10.1002/2013JD020732.
Skamarock, W. C., J. B. Klemp, J. Dudhia, D. O. Gill, D. M. Barker, W. Wang, and J. G. Powers, 2005: A description of the Advanced Research WRF version 2. NCAR Tech. Note NCAR/TN-468+STR, 88 pp. [Available online at http://www2.mmm.ucar.edu/wrf/users/docs/arw_v2_070111.pdf.]
Skinner, C. B., and N. S. Diffenbaugh, 2013: The contribution of African easterly waves to monsoon precipitation in the CMIP3 ensemble. J. Geophys. Res. Atmos., 118, 3590–3609, doi:10.1002/jgrd.50363.
Stensrud, D. J., 2013: Upscale effects of deep convection during the North American monsoon. J. Atmos. Sci., 70, 2681–2695, doi:10.1175/JAS-D-13-063.1.
Stensrud, D. J., R. L. Gall, and M. K. Nordquist, 1997: Surges over the Gulf of California during the Mexican monsoon. Mon. Wea. Rev., 125, 417–437, doi:10.1175/1520-0493(1997)125<0417:SOTGOC>2.0.CO;2.
Tebaldi, C., J. M. Arblaster, and R. Knutti, 2011: Mapping model agreement on future climate projections. Geophys. Res. Lett., 38, L23701, doi:10.1029/2011GL049863.
Torres-Alavez, A., T. Cavazos, and C. Turrent, 2014: Land–sea thermal contrast and intensity of the North American monsoon under climate change conditions. J. Climate, 27, 4566–4580, doi:10.1175/JCLI-D-13-00557.1.
van Oldenborgh, G. J., S. Y. Philip, and M. Collins, 2005: El Niño in a changing climate: A multi-model study. Ocean Sci., 1, 81–95, doi:10.5194/os-1-81-2005.
von Storch, H., and F. W. Zwiers, 1999: Statistical Analysis in Climate Research. Cambridge University Press, 484 pp.
Coastally trapped wave that propagates up the GoC forced by convection associated with inverted trough/tropical easterly wave passage near the south end of the GoC (e.g., Hales 1972; Stensrud et al. 1997).
An overlapping period with the NCEP-driven simulation in the baseline period and an equivalent number of years in the future.
Given that the bias is fairly consistent across all ECP2 simulations throughout the NARCCAP domain (not shown), this bias may be a result of the method in which the 500-hPa geopotential height was postcalculated from the model’s standard output.
