Abstract

The fifth-generation Pennsylvania State University–NCAR Mesoscale Model (MM5)-based regional climate model (CMM5) capability in simulating the interannual variations of U.S. precipitation and surface air temperature during 1982–2002 is evaluated with a continuous baseline integration driven by the NCEP–Department of Energy (DOE) Second Atmospheric Model Intercomparison Project Reanalysis (R-2). It is demonstrated that the CMM5 has a pronounced downscaling skill for precipitation and temperature interannual variations. The EOF and correlation analyses illustrate that, for both quantities, the CMM5 captures the spatial pattern, temporal evolution, and circulation teleconnections much better than the R-2. In particular, the CMM5 more realistically simulates the precipitation pattern centered in the Northwest, where the representation of the orographic enhancement by the forced uplifting during winter (rainy season) is greatly improved over the R-2.

The downscaling skill, however, is sensitive to the cumulus parameterization. This sensitivity is studied by comparing the baseline with a branch summer integration replacing the Grell with the Kain–Fritsch cumulus scheme in the CMM5. The dominant EOF mode of the U.S. summer precipitation interannual variation, identified with the out-of-phase relationship between the Midwest and Southeast in observations, is reproduced more accurately by the Grell than the Kain–Fritsch scheme, which largely underestimates the variation in the Midwest. This pattern is associated with east–west movement of the Great Plains low-level jet (LLJ): a more western position corresponds to a stronger southerly flow bringing more moisture and heavier rainfall in the Midwest and less in the Southeast. The second EOF pattern, which describes the consistent variation over the southern part of the Midwest and the South in observations, is captured better by the Kain–Fritsch scheme than the Grell, whose pattern systematically shifts southward.

1. Introduction

Interannual variations of precipitation and surface air temperature are integral parts of the climate system, where remote controls by planetary circulation and global surface anomalies act together with local influences by regional mesoscale surface characteristics. These two broad factors (remote versus local) can be distinguished by using a regional climate model (RCM), in which planetary signals are integrated through lateral boundary conditions while mesoscale impacts are internally resolved (e.g., Giorgi et al. 1993; Liang et al. 2001). Given the inadequacies of general circulation models (GCMs) to simulate regional climate variability, the RCM downscaling has become a powerful alterative and is widely applied in studies of seasonal–interannual climate prediction and future climate change projection.

It is imperative that any RCM must be rigorously validated in reproducing historical observations, including both mean climate and temporal variability, before credible application for climate change projection. In the published literature, numerous studies have demonstrated the RCM skill enhancement for downscaling regional characteristics, especially of precipitation and surface air temperature, focusing on mean climate (long-term averaged) biases such as in the annual and diurnal cycles (Pan et al. 2001; Roads et al. 2003; Leung et al. 2003a; Liang et al. 2004a, b; and references cited therein). There have been, however, a very limited number of studies validating the RCM capability in reproducing observed interannual variations.

Over the United States, a few studies compared RCM result differences between particular years, mainly the record 1988 drought and 1993 flood cases (Giorgi et al. 1996; Hong and Leetmaa 1999; Fennessy and Shukla 2000; Xue et al. 2001; Roads et al. 2003; Liang et al. 2004b). Dutton and Barron (2000) compared interannual variations and pattern correlations between an RCM and its driving GCM during a 10-yr period; since the forcing conditions were from a model simulation without a real-time correspondence, the comparison was not a direct validation of the RCM downscaling against observations. Gutowski et al. (2004) briefly discussed the 1979–88 interannual evolution of seasonal precipitation biases averaged in the Mississippi Delta, although the RCM integration was conducted over the whole United States. Zhu and Liang (2005) evaluated the RCM skill in downscaling interannual variations of soil moisture and soil temperature, limited to regional averages over the areas with sufficient observational data. Leung et al. (2003b), the most comprehensive study of the kind thus far, documented the RCM skill in reproducing the precipitation and temperature interannual variability over the western United States and their association with the El Niño–Southern Oscillation observed during 1981–2000.

The present study, a continuation of previous work by Liang et al. (2004a, b), systematically documents the fifth-generation Pennsylvania State University–National Center for Atmospheric Research (PSU–NCAR) Mesoscale Model (MM5)-based regional climate model (CMM5) downscaling skill, as compared with the driving reanalysis, against observations of interannual variations of precipitation and surface air temperature during 1982–2002 over the whole United States and their underlying physical mechanisms. This is facilitated by using empirical orthogonal function (EOF) and correlation analyses. Section 2 describes the model simulated and observational data used in the comparison. Section 3 focuses on the mean statistics, including interannual deviations explained by the dominant EOFs and gross spatial and temporal correlations between the simulated and observed patterns. Section 4 compares the spatial structure and temporal evolution of the dominant EOF patterns. Section 5 discusses the physical mechanisms associated with the dominant patterns of interannual variations.

2. Model simulations and observations

Liang et al. (2004b) described the CMM5 model formulation and precipitation annual cycle performance using a continuous baseline integration during 1982–2002. They demonstrated that the CMM5, with a horizontal resolution of 30 km, has considerable precipitation downscaling skill, producing more realistic regional details and overall smaller biases than the driving global reanalysis, the National Centers for Environmental Prediction–Department of Energy (NCEP–DOE) Second Atmospheric Model Intercomparison Project Reanalysis (R-2; Kanamitsu et al. 2002). The precipitation simulation is most skillful in the Northwest, where orographic forcing dominates throughout the year; in the Midwest, where mesoscale convective complexes prevail in summer; and in the central Great Plains, where nocturnal low-level jet (LLJ) and rainfall peaks occur in summer.

Liang et al. (2004a) later documented the CMM5 skill in downscaling the precipitation diurnal cycle and its sensitivity to the choice of the cumulus parameterization. A branch CMM5 simulation was conducted in parallel to the baseline integration, where the Grell (1993) cumulus scheme was replaced by the Kain and Fritsch (1993) scheme with everything else being identical. The branch simulation included each summer during 1982–2002 started from the baseline condition on 1 April. The results demonstrated the importance of cumulus schemes and provided a realistic simulation of the central U.S. nocturnal precipitation maxima, which remains a challenging problem in the global and regional climate modeling.

The above two studies have shown that the CMM5 performance is sensitive to the choice of cumulus schemes, whose skills are highly climate regime selective. The Grell scheme realistically simulates the nocturnal precipitation maxima over the central United States and the associated eastward propagation of convective systems from the Rockies to the Great Plains, where the diurnal timing of convection is controlled by large-scale tropospheric forcing, whereas the Kain–Fritsch scheme is more accurate for the late afternoon peaks in the Southeast, where moist convection is governed by near-surface forcing (Liang et al. 2004a). Summer rainfall amounts in the North American monsoon region are simulated very poorly by the Grell scheme but are well reproduced by the Kain–Fritsch scheme, whereas rainfall amounts from moist convection in the Southeast are underestimated by the former and overestimated by the latter (Liang et al. 2004b). This study compares the baseline integration and branch simulation to further demonstrate the impact of the cumulus schemes on the CMM5 capability in downscaling interannual variations of precipitation and surface air temperature. The following analysis refers to the R-2-driven CMM5 simulations using the Grell and Kain–Fritsch schemes denoted as RGR and RKF, respectively.

Several daily observational datasets are utilized for the validation. For the circulation fields (850-hPa wind and 500-hPa height), the R-2 data are taken as the best proxy of observations because they assimilated all available measurements. They are available on a global 2.5° latitude × 2.5° longitude grid mesh. Daily sea surface temperature (SST) data are based on conservative spline fit from the weekly optimum interpolation SST (OISST) analysis (Reynolds et al. 2002). The OISST is available over the global oceans on a 1° latitude × 1° longitude grid mesh and used consistently in both the R-2 reanalysis and the CMM5 downscaling (see Liang et al. 2004b). Given their relatively coarse resolutions, for a quantitative comparison, all these fields were mapped onto the CMM5 grids using bilinear spatial interpolation, and no area weighting is applied in subsequent EOF and correlation analyses.

The precipitation observational data are derived from an the objective analysis of daily measurements from the National Weather Service cooperative observer stations over the contiguous United States mapped onto the CMM5 30-km grid mesh [see Liang et al. (2004b) for details of the data source and analysis procedure]. There are 7235 cooperative stations, where the density is generally high and compatible with the CMM5 30-km grid except for mountainous regions in the Rockies. For surface air temperature, daily mean values were constructed from the average of daily maximum and minimum temperature measurements at the same cooperative stations. A constant lapse rate factor based on the individual station altitude first was subtracted from the station observation, and the resulting values were then mapped onto the CMM5 grid via objective analysis in a manner similar to precipitation. Finally, the lapse rate factor based on the local CMM5 terrain height was added back to the objectively analyzed values. For a quantitative comparison, the R-2-generated precipitation and surface air temperature are also mapped onto the CMM5 grids using bilinear spatial interpolation.

3. Mean statistics

Figure 1 illustrates the frequency distributions of mean biases and root-mean-square (rms) errors of daily surface air temperature and precipitation over the whole contiguous United States as simulated by the R-2, RGR, and RKF. Both the mean biases and rms errors are first calculated at each grid, and the final frequency distributions are obtained from the resulting values at all land grids over the United States during 1982–2002. In winter, the R-2 has more grids with greater temperature biases than the RGR, which results mainly from the systematically colder R-2 values, especially over the northern states and the Rockies. In summer, the R-2 temperature biases are the smallest, whereas the RGR has more grids of slightly warmer temperature mainly over the Rockies, and the RKF overestimation is much more extensive and substantial over the Rockies and most of the central-eastern United States. For winter precipitation, the R-2 has more grids with dry biases over the Southeast, while the RGR generates more grids of light precipitation (<1.0 mm day−1) than observations over the central and northern United States. When these grids are not counted, the RGR is actually more realistic than the R-2. In summer, the frequency distribution shifts toward wetter mean biases from the RGR to RKF to R-2, which are mainly contributed in the RGR by large underestimations in the southern and eastern states, in the RKF by medium underestimations (overestimations) in the central-southern Great Plains and southern Rockies (the Southeast), and in the R-2 by substantial overestimation over the Southeast.

Fig. 1.

Frequency distributions of mean biases and rms errors of winter (DJF) and summer (JJA) (left) surface air temperature (°C) and (right) precipitation (mm day−1) over the whole contiguous United States as simulated by the driving R-2 (thin solid) and downscaled by the CMM5 using the Grell (RGR, thick solid) and Kain–Fritsch (RKF, thick dashed) cumulus schemes. Shown also are the DJF frequency distributions of rms errors by the R-2 (thin dotted) and RGR (thin dashed) where the grids with light precipitation (<1.0 mm day−1) are not counted. All frequencies are defined in bins of a half-unit interval.

Fig. 1.

Frequency distributions of mean biases and rms errors of winter (DJF) and summer (JJA) (left) surface air temperature (°C) and (right) precipitation (mm day−1) over the whole contiguous United States as simulated by the driving R-2 (thin solid) and downscaled by the CMM5 using the Grell (RGR, thick solid) and Kain–Fritsch (RKF, thick dashed) cumulus schemes. Shown also are the DJF frequency distributions of rms errors by the R-2 (thin dotted) and RGR (thin dashed) where the grids with light precipitation (<1.0 mm day−1) are not counted. All frequencies are defined in bins of a half-unit interval.

More important is the skill of the models in simulating the observed interannual variation patterns and the associated physical mechanisms. This is facilitated here by using EOF and correlation analyses. Our intention is not to document all EOF patterns but to focus on those dominant ones that are representative of model performance. The threshold where the dominant EOF modes are cut off from the indistinguishable ones can be determined as the point where the percentage of the explained deviation starts to flatten. For surface air temperature, this occurs at EOF4 in both summer and winter (Table 1), and thus only the first three EOFs are considered as the dominant modes. For precipitation, this appears at EOF3 in summer and EOF4 in winter (Table 2), and hence the first two and three EOFs are identified as the dominant modes, respectively. The subsequent analyses elaborate more on these dominant modes.

Table 1.

The variance (%) explained by the first four EOFs of surface air temperature in summer and winter as observed, downscaled by the CMM5 using the Grell (RGR) and Kain–Fritsch (RKF) cumulus scheme, and simulated by the driving reanalysis (R-2), as well as the corresponding temporal and spatial correlation coefficients of the RGR, RKF, and R-2 with observations.

The variance (%) explained by the first four EOFs of surface air temperature in summer and winter as observed, downscaled by the CMM5 using the Grell (RGR) and Kain–Fritsch (RKF) cumulus scheme, and simulated by the driving reanalysis (R-2), as well as the corresponding temporal and spatial correlation coefficients of the RGR, RKF, and R-2 with observations.
The variance (%) explained by the first four EOFs of surface air temperature in summer and winter as observed, downscaled by the CMM5 using the Grell (RGR) and Kain–Fritsch (RKF) cumulus scheme, and simulated by the driving reanalysis (R-2), as well as the corresponding temporal and spatial correlation coefficients of the RGR, RKF, and R-2 with observations.
Table 2.

Same as in Table 1, except for precipitation. The original EOF order numbers in terms of their explained variances are listed in parentheses.

Same as in Table 1, except for precipitation. The original EOF order numbers in terms of their explained variances are listed in parentheses.
Same as in Table 1, except for precipitation. The original EOF order numbers in terms of their explained variances are listed in parentheses.

For both temperature (Table 1) and precipitation (Table 2), the winter patterns generally are easier to reproduce than summer patterns, where temporal correlations with observations are higher and differences in temporal and spatial structure between the CMM5 downscaling and the driving R-2 are small. As such the subsequent comparison focuses more on the summer patterns. In addition, the interannual variation patterns are more complicated for precipitation than temperature, where the variance explained by the observed winter (summer) EOF1 is only 36.1% (20.3%) in the former but almost doubled as 50.9% (40.6%) in the latter. It is thus important to better understand the model skill in simulating the precipitation patterns by comparing the responsible physical mechanisms (see section 5).

4. Dominant interannual variation patterns

Figure 2 shows the first three EOF modes of summer temperature and the corresponding principal component (PC) observed and simulated by the RGR, RKF, and R-2, where summer is defined as the average of June–August (JJA). Table 1 lists the deviation explained by these modes as well as temporal correlations between the models and observations. The observed dominant mode represents the in-phase variation over most areas of the United States centered at the northern Great Plains, the second mode depicts the out-of-phase variation between the western and eastern United States, and the third mode shows that the opposite variation between the northern and the southern United States. These first three modes explain a total of 74.5% of the interannual variation. The CMM5 captures these patterns with the total explained variance of 81.8% by the RGR and 78.5% by the RKF. The spatial correlation coefficients with the observed first three modes are 0.91, 0.83, and 0.87 by the RGR and 0.97, 0.92, and 0.89 by the RKF, respectively. The corresponding temporal correlation coefficients are 0.65, 0.79, and 0.64 by the RGR and 0.80, 0.92, and 0.65 by the RKF, respectively. Although the RKF produces high correlation coefficients of the spatial pattern and temporal variation, its credibility is lower because of substantial systematic warm biases (2.7°C), which are almost double that of the RGR (1.5°C).

Fig. 2.

(a)–(c) The PC during 1982–2002 of the first three dominant EOF modes of summer surface air temperature. The respective geographic distributions (eigenvectors) of these EOF patterns (d)–(f) observed (OBS), downscaled by the (g)–(i) RGR and (j)–(l) RKF, and (m)–(o) simulated by the R-2.

Fig. 2.

(a)–(c) The PC during 1982–2002 of the first three dominant EOF modes of summer surface air temperature. The respective geographic distributions (eigenvectors) of these EOF patterns (d)–(f) observed (OBS), downscaled by the (g)–(i) RGR and (j)–(l) RKF, and (m)–(o) simulated by the R-2.

The R-2 simulation has larger differences from observations than the RGR and RKF. Its EOF spatial pattern correlations are 0.66, 0.58, and 0.81 and corresponding temporal correlations are 0.53, 0.53, and 0.83, respectively, for the first three modes. The total variance explained by these modes is 74.3%. Note that the R-2 assimilates all observations and thus represents the best proxy of the observed structure. Given the lack of such data assimilation, the CMM5 would be expected to have relatively larger mean errors than the R-2. It is surprising that the CMM5 captures the interannual patterns and associated physical processes (see section 5) in a more realistic fashion than the R-2.

Figure 3 shows the spatial patterns of the first three EOF modes of summer precipitation and corresponding interannual variations observed and simulated by the RGR, RKF, and R-2. In observations, the dominant mode represents the out-of-phase variation between the Midwest and the Southeast; the second mode depicts the coherent variation over most of the domain centered in the central United States; the third mode is a triplet where the Northwest and the Gulf states vary in opposition to the central region extending from the northeast to southwest. The variances explained by the first three modes are 20.3%, 14.5%, and 8.0%, respectively, with a total of 42.8% that is much smaller than that for temperature (74.5%). As discussed above, the interannual variation of precipitation is more complex and also more difficult to model than temperature. In particular, the explained percentage of variance represented by each mode in the models may not be in the same order as the correlation coefficient with observations. For a meaningful comparison, the simulated modes are rearranged by the largest spatial pattern correlation coefficient with the respective observed modes, although Table 2 also lists the original order numbers in terms of their explained variances.

Fig. 3.

Same as in Fig. 2, except for summer precipitation. Outlined are the subdomains of the (d) Southeast, (e) Midwest, and (f) Northwest referred to in the text.

Fig. 3.

Same as in Fig. 2, except for summer precipitation. Outlined are the subdomains of the (d) Southeast, (e) Midwest, and (f) Northwest referred to in the text.

The first mode in the RGR explains 19.2% of the variance, very close to observation. The spatial pattern correlation coefficient with observations is 0.41 (statistically significant at the 99% level), although the RGR overestimates (underestimates) the variance in the Midwest (Southeast). The RKF pattern explains 16.3% of the variance in the data and has a spatial correlation of 0.38 with the observed mode due to a much weak center in the Midwest. The RKF model run reproduces the structure over the Southeast better than the RGR, but underestimates the observed variance in the Midwest. Both the RGR and RKF unrealistically shift the variance center from the Southeast states to the coast of Texas. The R-2 result is much less realistic than the CMM5: the pattern having the highest spatial correlation with the observed first mode explains the second largest variance (much less than the first one); the observed Midwest center is dislocated to the northwest and the variance along the Gulf coast is substantially underestimated. Clearly, the CMM5 downscaling corrects important biases in the driving R-2.

For the observed second mode, the RKF pattern explains 14.0% of the variance and has the highest spatial correlation coefficient (0.66) with observations. The RKF overestimates (underestimates) the variance in the lower Mississippi (the central United States). The RGR pattern is relatively less realistic, explaining 9.3% of the variance and having a spatial correlation with observations of 0.47; the pattern tends to focus more toward the south coast. The R-2 produces an even stronger center over the central to southern states, explaining 28.7% of precipitation variance (the dominant mode). The CMM5 overestimation toward the south coast may be a reflection of the driving R-2 error.

For the observed third mode, the RGR, RKF, and R-2 explain 8.0%, 7.0%, and 8.9% of the variance and the spatial pattern correlation are 0.36, 0.31, and 0.22, respectively. The biggest differences in this mode between the models and observations are identified over the central United States, and the R-2 simulation is the worst.

Figure 4 compares the simulated with observed interannual variations of summer mean precipitation averaged over the southeast, midwest, and northwest United States (see Fig. 3 for the boundaries). They are the representative centers of the first three dominant EOF patterns as identified by the areas of high interannual correlation coefficients (exceeding ±0.6) with the actual precipitation distribution (not shown). Listed also are the corresponding mean statistics, including climate mean, interannual variability, and correlation coefficients.

Fig. 4.

Interannual variations during 1982–2002 of summer precipitation (mm day−1) averaged over the (a) Southeast, (b) Midwest, and (c) Northwest regions outlined in Fig. 3. The numbers in the parentheses following each model name show the mean statistics in the order of long-term average, interannual standard deviations, and correlation coefficients with observations. All calculations are based on seasonal means of regional averages.

Fig. 4.

Interannual variations during 1982–2002 of summer precipitation (mm day−1) averaged over the (a) Southeast, (b) Midwest, and (c) Northwest regions outlined in Fig. 3. The numbers in the parentheses following each model name show the mean statistics in the order of long-term average, interannual standard deviations, and correlation coefficients with observations. All calculations are based on seasonal means of regional averages.

In the Southeast, the R-2 substantially overestimates the observed summer mean precipitation (>70%). In contrast, the CMM5 downscaling is more realistic, where the RKF value is 13% larger and RGR is 20% lower than observations. The CMM5 underestimates interannual deviation by 23% (38%) in the RGR (RKF), whereas the R-2 overestimates that by 73%. Although the correlation coefficient with the observed interannual variation is the highest (0.62) by the R-2, followed by the RGR (0.48) and RKF (0.35), the R-2 credibility is low because it produces much greater errors in both mean and deviation than the CMM5.

In the Midwest, the summer mean precipitation and interannual variability simulated by the RGR is the most realistic, having the highest correlation coefficient (0.75) with the observed interannual variation, while the RKF (R-2) underestimates (overestimates) the mean precipitation and has a smaller correlation with observations of 0.65 (0.55). All correlation coefficients are significant at the 99% level.

In the Northwest, the RGR and RKF underestimate the summer mean precipitation and interannual variability, while a much larger overestimation is produced by the R-2. In contrast, the correlation coefficient with the observed interannual variation is the highest by RGR (0.90), followed by RKF (0.85) and R-2 (0.81).

The above result indicates that the RGR performs better overall in the central United States and Midwest for both seasonal mean and interannual variation, while the RKF is more realistic over the Southeast. This is consistent with our previous findings (Liang et al. 2004a, b). On the other hand, the driving R-2 is poor in simulating the dominant modes of observed precipitation interannual variations. The noticeable R-2 problem is that its evolution of the first three modes becomes more unrealistic after 1996. This problem also has been noticed in the study of soil temperature and moisture (Zhu and Liang 2005).

5. Physical mechanisms causing precipitation patterns

The remaining important question is whether the models reproduce the observed physical mechanisms that govern the dominant modes of interannual variability. This section focuses on those major precipitation patterns, where the correlation analysis with key planetary circulation features is conducted for a possible physical interpretation.

The LLJ plays an extremely important role in explaining the observed first EOF mode where precipitation in the Midwest varies out of phase with the Southeast. Figure 5 shows the geographic distribution of correlation coefficients of the precipitation EOF1 with pointwise meridional wind at 850 hPa. This pattern represents mainly the effect of the LLJ east–west movement. In observations, when the LLJ is shifted to the west of the normal position, stronger southerly flows prevail along the southern Great Plains and thus transport more moisture toward the Midwest, producing heavier rainfall there and inversely less precipitation over the Southeast. The opposite occurs when the LLJ moves to the east. The RGR reproduces the main feature of positive correlations over the southern Great Plains; this center, however, is stretched toward the Northeast and distorting the precipitation pattern, while the negative correlation center over the Southeast is absent causing the lack of precipitation response there. In contrast, the RKF improves the negative correlation somewhat by producing a more realistic (still weaker) precipitation response in the Southeast; the positive correlation center, however, becomes weaker and less extensive, causing a weaker precipitation response in the Midwest. The R-2 reproduces the negative center along the east coast, but simulates only a very weak positive center over the southern Great Plains. The existence of important biases in this teleconnection across all the models indicates the challenge confronting predictability of the LLJ and the associated precipitation pattern over the eastern United States.

Fig. 5.

Geographic distributions of the temporal correlation coefficients (× 10) between summer precipitation EOF1 and pointwise 850-hPa meridional wind during 1982–2002: (a) OBS, (b) RGR, (c) RKF, and (d) R-2. Contour intervals are 2, where negative correlations are dashed. The shaded areas are where correlations are statistically significant at the 95% confidence level.

Fig. 5.

Geographic distributions of the temporal correlation coefficients (× 10) between summer precipitation EOF1 and pointwise 850-hPa meridional wind during 1982–2002: (a) OBS, (b) RGR, (c) RKF, and (d) R-2. Contour intervals are 2, where negative correlations are dashed. The shaded areas are where correlations are statistically significant at the 95% confidence level.

SSTs are one of the most important factors identified with the interannual variation of summer precipitation in the United States. (Ropelewski and Halpert 1986; Ting and Wang 1997; Barlow et al. 2001). Figure 6 shows the geographic distribution of interannual correlation coefficients of the summer precipitation EOF2 with pointwise SSTs over global oceans as observed, simulated by the R-2 and downscaled by the RGR and RKF. The observed precipitation in the central United States is closely related to the SST anomalies with a north–south triplet pattern distributed over the northeastern Pacific, where the negative correlation center along 30°N extends westward to the mid-Pacific. The RGR realistically reproduces this teleconnection pattern except that the positive correlation center in the central Pacific 10°S–10°N is too strong and widespread. The RKF feature is also realistic except that the negative correlation center in the mid-Pacific is too strong. The R-2 pattern is much less realistic with a broader but weaker negative center in the northern mid-Pacific and an excessively strong positive center in the tropical Pacific.

Fig. 6.

Same as in Fig. 5, except for correlations between summer precipitation EOF2 and pointwise sea surface temperature.

Fig. 6.

Same as in Fig. 5, except for correlations between summer precipitation EOF2 and pointwise sea surface temperature.

This SST teleconnection is identified with a distinct planetary circulation pattern. Figure 7 depicts the geographic distribution of correlation coefficients of the summer precipitation EOF2 with pointwise geopotential height at 500 hPa. In observations, the low (high) anomaly geopotential height over the north-central United States (Southeast) is identified with the heavier precipitation over the central United States. The RGR faithfully reproduces this relationship except with a slightly broader/stronger center over the Gulf Coast. The RKF simulation is also realistic except that the Gulf center is slightly shifted toward the northeast. The R-2 simulation is much poorer where the two centers have weaker correlations and the axis across them is turned anticlockwise.

Fig. 7.

Same as in Fig. 5, except for correlations between summer precipitation EOF2 and pointwise 500-hPa geopotential height.

Fig. 7.

Same as in Fig. 5, except for correlations between summer precipitation EOF2 and pointwise 500-hPa geopotential height.

The observed patterns of the precipitation EOF2 correlations with SST (Fig. 6a) and 500-hPa height (Fig. 7a) closely resemble those of Ting and Wang (1997), corresponding to their Fig. 6 and Fig. 10b, respectively. Although the differences exist in data periods (1982–2002 versus 1950–90) and the analysis domains, the close similarity indicates that these patterns are physically the same. This pattern seems to contain signals associated with both tropical and extratropical SST anomalies as well as a wave train of circulation anomalies over the Pacific–North American region, suggesting the role of the Rossby wave propagation from the tropical central Pacific. As explained by Ting and Wang (1997), when a deeper trough located over the north-central United States (Fig. 7a), stronger northerly flows from interior Canada carry cold air toward the United States, and combined with the southward shift of the storm track, this brings the summer atmospheric circulation over the central United States into spring-like conditions: relatively cold and wet. This is consistently reflected in our pattern correlation with surface air temperature (not shown).

The rainy season for the Northwest is winter [December–February (DJF)]. Figure 8 compares the dominant EOF mode of winter precipitation interannual variation between the observed, simulated by the R-2 and downscaled by the RGR; also shown are the PC of the EOF1 eigenvalue and mean precipitation averaged over the representative central Cascades (box in Fig. 8c). The observed, RGR, and R-2 patterns explain 36.1%, 37.4%, and 35.3% of the variances, respectively (Table 2). The spatial correlation coefficients with observations are 0.88 by the RGR and 0.73 by the R-2. The corresponding principal component of this mode simulated by the RGR matches observations almost perfectly with a correlation coefficient of 0.98, whereas systematic biases are produced by the R-2 with a smaller correlation of 0.87. The regional mean (interannual deviation) averaged over the Cascades is 5.71 (1.73) mm day−1 by the RGR, close to the observed 5.80 (1.90) mm day−1. On the other hand, the R-2 significantly underestimates both the mean (4.02 mm day−1) and variability (1.31 mm day−1). Although the driving R-2 poorly simulates the precipitation structure over the Northwest, the RGR downscaling corrects most of the errors and produces realistically the detailed distribution characterized by distinct orographic features. In addition, the R-2 substantially overestimates the positive precipitation response over the Southeast. This deficiency is also largely corrected by the RGR. The result is consistent with Liang et al. (2004b) for the precipitation annual cycle.

Fig. 8.

(a) The PC during 1982–2002 of the first dominant EOF mode of winter precipitation and (b) the regional mean precipitation rate (mm day−1) over the Cascades. The numbers in the parentheses in (b) following each model name show the mean statistics (see Fig. 4). (c)–(e) The respective geographic distributions of this EOF mode and (f)–(h) its temporal correlation coefficients (×10) with pointwise 500-hPa geopotential height are shown for (c), (f) OBS, (d), (g) RGR, and (e), (h) R-2. Contour intervals are 2. The shaded areas are where correlations are statistically significant at the 95% confidence level. In (c) the subdomain of the Cascades is outlined.

Fig. 8.

(a) The PC during 1982–2002 of the first dominant EOF mode of winter precipitation and (b) the regional mean precipitation rate (mm day−1) over the Cascades. The numbers in the parentheses in (b) following each model name show the mean statistics (see Fig. 4). (c)–(e) The respective geographic distributions of this EOF mode and (f)–(h) its temporal correlation coefficients (×10) with pointwise 500-hPa geopotential height are shown for (c), (f) OBS, (d), (g) RGR, and (e), (h) R-2. Contour intervals are 2. The shaded areas are where correlations are statistically significant at the 95% confidence level. In (c) the subdomain of the Cascades is outlined.

This winter precipitation pattern is caused mainly by the orographic uplifting effect on the westerly flow over the western mountains. Figure 8 also shows the geographic distribution of correlation coefficients of this precipitation pattern with the geopotential height at 500 hPa. The dipole structure along the western United States and adjacent Pacific Ocean indicates that stronger westerly flow brings more moisture that is forced upward by the blocking mountains to enhance condensation and subsequently heavier precipitation. The RGR realistically captures this orographic lifting effect. Although the R-2 also correctly simulates the 500-hPa height (and flow) pattern, the coarse resolution diminishes the mountain peaks, and thus the model provides insufficient precipitation details.

Interestingly, an opposite dipole structure occurs along the eastern United States and adjacent Atlantic Ocean in the 500-hPa height correlation distribution. This again indicates anomalous easterly flow bringing more rain on shore. Given the relatively low mountains, the orographic uplifting is also weak. The R-2 excessive rainfall over the Southeast results mainly from its weaker subtropical Atlantic.

Note that the two precipitation centers over the Northwest and Southeast during winter are not coupled but are an artifact of the EOF analysis. The correlation coefficient of actual precipitation interannual variations averaged over these centers is very low for both observations (0.10) and RGR (0.07). The R-2, however, incorrectly depicts this independence, with a high correlation of 0.34.

6. Conclusions

The CMM5 integration for 1982–2002 driven by the R-2 reanalysis with Grell and Kain–Fritsch schemes was diagnosed to determine the RCM capability and uncertainty in reproducing the observed interannual variability of U.S. precipitation and surface air temperature. The EOF and correlation analyses are performed to determine the dominant patterns, based on which model validations and intercomparisons are made. To better understand the RCM skill enhancement in simulating the dominant modes of interannual variability, a teleconnection analysis is then conducted to link these regional patterns with global circulation and surface climate anomalies.

For summer precipitation, the observed dominant mode depicts the out-of-phase correlation between the Midwest and the Southeast. This precipitation pattern is identified with the LLJ east–west displacement. When the LLJ is located to the west of its climatological position, stronger southerly flow prevails over the southern Great Plains, producing heavier rainfall in the Midwest and less precipitation over the Southeast. Overall the RGR better reproduces the positive correlations over the Midwest, while the RKF improves the negative correlations in the Southeast. In contrast, the R-2 captures the negative center along the East Coast but simulates a very weak positive center over the southern Great Plains. The result indicates that the predictability of the LLJ and associated precipitation pattern remains a challenging problem in global and regional climate modeling. The second mode of summer precipitation represents the coherent interannual variation over most of the domain centered in the central United States, which is closely related to the north–south triplet SST anomalies over the northeastern Pacific and the planetary circulation pattern over North America at 500-hPa height. The RGR best reproduces this teleconnection pattern, whereas the R-2 pattern is the least realistic.

In winter, the dominant mode of precipitation interannual variability describes the heavy precipitation anomalies over the western United States centered along the coastal states. It is identified with the dipole structure in 500-hPa height anomalies along the western United States and adjacent Pacific Ocean, indicating stronger westerly flow that brings more moisture to be forced upward by the blocking mountains, enhancing condensation, and subsequently producing heavier precipitation. The RGR realistically captures this orographic lifting effect with its spatial pattern matching observations almost perfectly, whereas the R-2 lacks important structural details as a result of its coarse resolution that diminishes mountain peaks.

The above results demonstrate that the CMM5 runs capture the spatial pattern and temporal variation of the dominant patterns and are more realistic than the driving R-2 for both precipitation and temperature. The CMM5 downscaling is able to correct important R-2 biases, especially in precipitation. Consistent with our previous findings (Liang et al. 2004a, b), the actual downscaling skill is sensitive to the choice of the cumulus parameterization. In particular, for summer precipitation, the RGR better reproduces interannual variations in the Midwest, while the RKF is more realistic over the Southeast. For summer temperature, although the RKF produces higher correlation coefficients of the spatial pattern and temporal variations, its credibility is lower because of substantial systematic warm biases (2.7°C) almost double that of the RGR (1.5°C). These sensitivities will have important consequences for the CMM5 applications in seasonal–interannual climate prediction, future climate change projection, and impact studies including air quality modeling.

Acknowledgments

We thank Kenneth Kunkel and Michael Palecki for constructive discussions. We thank Prof. David M. Straus and three anonymous reviewers for their valuable comments. We acknowledge FSL/NOAA and NCSA/UIUC for the supercomputing support and NCAR for access to the R-2 data. This study was supported in part by U.S. Environmental Protection Agency Award RD-83096301-0. The views expressed are those of the authors and do not necessarily reflect those of the sponsoring agencies or the Illinois State Water Survey.

REFERENCES

REFERENCES
Barlow
,
M.
,
S.
Nigam
, and
E. H.
Berbery
,
2001
:
ENSO, Pacific decadal variability, and U.S. summertime precipitation, drought, and streamflow.
J. Climate
,
14
,
2105
2128
.
Dutton
,
J. F.
, and
E. J.
Barron
,
2000
:
Intra-annual and interannual ensemble forcing of a regional climate model.
J. Geophys. Res.
,
105
,
29523
29538
.
Fennessy
,
M. J.
, and
J.
Shukla
,
2000
:
Seasonal prediction over North America with a regional model nested in a global model.
J. Climate
,
13
,
2605
2627
.
Giorgi
,
F.
,
M. R.
Marinucci
, and
G. T.
Bates
,
1993
:
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
.
Giorgi
,
F.
,
L. O.
Mearns
,
C.
Shields
, and
L.
Mayer
,
1996
:
A regional model study of the importance of local versus remote controls of the 1988 drought and the 1993 flood over the central United States.
J. Climate
,
9
,
1150
1162
.
Grell
,
G. A.
,
1993
:
Prognostic evaluation of assumptions used by cumulus parameterizations.
Mon. Wea. Rev.
,
121
,
764
787
.
Gutowski
,
W. J.
,
F. O.
Otieno
,
A. W.
Raymond
,
E. S.
Takle
, and
Z.
Pan
,
2004
:
Diagnosis and attribution of a seasonal precipitation deficit in a U.S. regional climate simulation.
J. Hydrometeor.
,
5
,
230
242
.
Hong
,
S-Y.
, and
A.
Leetmaa
,
1999
:
An evaluation of the NCEP RSM for regional climate modeling.
J. Climate
,
12
,
592
609
.
Kain
,
J. S.
, and
J. M.
Fritsch
,
1993
:
Convective parameterization in mesoscale models: The Kain-Fritsch scheme. The Representation of Cumulus Convection in Numerical Models, Meteor. Monogr., No. 46, Amer. Meteor. Soc., 165–170
.
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
.
Leung
,
K. R.
,
Y.
Qian
, and
X.
Bian
,
2003a
:
Hydroclimate of the western United States based on observations and regional climate simulation of 1981–2000. Part I: Seasonal statistics.
J. Climate
,
16
,
1892
1911
.
Leung
,
K. R.
,
Y.
Qian
,
X.
Bian
, and
A.
Hunt
,
2003b
:
Hydroclimate of the western United States based on observations and regional climate simulation of 1981–2000. Part II: Mesoscale ENSO anomalies.
J. Climate
,
16
,
1912
1928
.
Liang
,
X-Z.
,
K. E.
Kunkel
, and
A. N.
Samel
,
2001
:
Development of a regional climate model for U.S. Midwest applications. Part I: Sensitivity to buffer zone treatment.
J. Climate
,
14
,
4363
4378
.
Liang
,
X-Z.
,
L.
Li
,
A.
Dai
, and
K. E.
Kunkel
,
2004a
:
Regional climate model simulation of summer precipitation diurnal cycle over the United States.
Geophys. Res. Lett.
,
31
.
L24208, doi:10.1029/2004GL021054
.
Liang
,
X-Z.
,
L.
Li
,
K. E.
Kunkel
,
M.
Ting
, and
J. X. L.
Wang
,
2004b
:
Regional climate model simulation of U.S. precipitation during 1982–2002. Part I: Annual cycle.
J. Climate
,
17
,
3510
3528
.
Pan
,
Z.
,
J. H.
Christensen
,
R. W.
Arritt
,
W. J.
Gutowski
Jr.
,
E. S.
Takle
, and
F.
Otieno
,
2001
:
Evaluation of uncertainties in regional climate change simulations.
J. Geophys. Res.
,
106
,
17735
17751
.
Reynolds
,
R. W.
,
N. A.
Rayner
,
T. M.
Smith
,
D. C.
Stokes
, and
W.
Wang
,
2002
:
An improved in situ and satellite SST analysis for climate.
J. Climate
,
15
,
1609
1625
.
Roads
,
J.
,
S-C.
Chen
, and
M.
Kanamitsu
,
2003
:
U.S. regional climate simulations and seasonal forecasts.
J. Geophys. Res.
,
108
.
8606, doi:10.1029/2002JD002232
.
Ropelewski
,
C. F.
, and
M. S.
Halpert
,
1986
:
North American precipitation and temperature patterns associated with the El Niño/Southern Oscillation (ENSO).
Mon. Wea. Rev.
,
114
,
2352
2362
.
Ting
,
M.
, and
H.
Wang
,
1997
:
Summertime U.S. precipitation variability and its relation to Pacific sea surface temperature.
J. Climate
,
10
,
1853
1873
.
Xue
,
Y.
,
F. J.
Zeng
,
K. E.
Mitchell
,
Z.
Janjic
, and
E.
Rogers
,
2001
:
The impact of land surface processes on simulations of the U.S. hydrological cycle: A case study of the 1993 flood using the SSiB land surface model in the NCEP Eta regional model.
Mon. Wea. Rev.
,
129
,
2833
2860
.
Zhu
,
J.
, and
X-Z.
Liang
,
2005
:
Regional climate model simulation of U.S. soil temperature and moisture during 1982–2002.
J. Geophys. Res.
,
110
.
D24110, doi:10.1029/2005JD006472
.

Footnotes

Corresponding author address: Dr. Jinhong Zhu, Illinois State Water Survey, 2204 Griffith Dr., Champaign, IL 61820-7495. Email: zjh@uiuc.edu