Contribution of land-atmosphere coupling to summer climate variability over the contiguous United States

Authors


Abstract

[1] The Weather Research and Forecasting (WRF) model has been used to study the role of land-atmosphere coupling in influencing interannual summer climate variability over the contiguous United States. Two long-term climate simulations are performed: a control experiment (CTL) allows soil moisture to interact freely with the atmosphere, and an additional experiment uncouples the land surface from the atmosphere by replacing summer soil moisture at each time step with the climatology of CTL. The CTL simulation reproduces well the observed summer temperature and precipitation variability, despite some discrepancies in daily mean and maximum temperature variability in the midwest/Ohio Valley region and the adjacent areas, and precipitation variability in the Great Plains and some other areas. Strong coupling of soil moisture with daily mean temperature appears mainly over the zone from the southwest to the northern Great Plains to the southeast, contributing up to about 30–60% of the total interannual variance of temperature. There is a significantly different influence on daily maximum and minimum temperatures. Soil moisture plays a leading role in explaining the variability of maximum temperature over this zone whereas minimum temperature variability is highly constrained by external factors including atmospheric circulation and sea surface temperature almost everywhere over land. Soil moisture, mainly through its effects on convection, makes a dominant contribution to precipitation variability over about half of the northern United States. This result does not support the Global Land-Atmosphere Coupling Experiment (GLACE) hot spot hypothesis over the central United States, at least on the interannual timescale. The model's behavior agrees to a large extent with land-atmosphere relationships diagnosed using the observations.

1. Introduction

[2] Increased temperature and precipitation variability can potentially cause more climate extremes such as more frequent and intense heat waves, an increased chance of drought, and increased intensity of precipitation over the midlatitude continents in summer [Easterling et al., 2000; Räisänen, 2002; Meehl and Tebaldi, 2004]. Land surface is thought to be an important slowly varying component of the Earth system that affects climate variation and variability, especially involving variable of soil wetness [e.g., Dirmeyer, 1995].

[3] The soil can “remember” a dry or wet anomaly and further maintain low or high evaporation and transpiration anomalies for several months, which may in turn play an important role in the atmosphere evolution [e.g., Shukla and Mintz, 1982; Delworth and Manabe, 1988]. The coupling between soil moisture conditions and the atmosphere is found to be sensitive to regional climate regimes [e.g., Koster et al., 2000]. The Global Land-Atmosphere Coupling Experiment (GLACE) [Koster et al., 2004, 2006a; Guo et al., 2006] study demonstrated that strong land–atmosphere coupling mainly appears in the transitional zones between dry and wet climates where evaporation is suitably high but still sensitive to soil moisture in the boreal summer season. Zhang et al. [2008] recently showed evidences for soil moisture feedback on precipitation in the climatic/ecological transitional zones by statistical analyses to observed precipitation and soil moisture data from the Global Land Data Assimilation System (GLDAS) [Rodell et al., 2004], ERA-40 Reanalysis [Uppala et al., 2005], and limited-area observations. The importance of soil moisture to climate variability also varies with season. While soil moisture could have a weak impact on wintertime climate variability which is largely influenced by the large-scale circulation associated with sea surface temperature (SST) anomalies, it may play a leading role in summertime variability over the midlatitude continents [Koster et al., 2000; Douville and Chauvin, 2000; Kushnir et al., 2002; Conil et al., 2007].

[4] Surface air temperature is determined by surface energy balance involving radiation, latent and sensible heat fluxes and heat storage, and is thus expected to be highly sensitive to localized interactions between land surface and the overlying atmosphere. Timbal et al. [2002] demonstrated the crucial role of soil moisture in temperature variability and predictability over Australia. Koster et al. [2006b] suggested that land moisture variables have a first-order impact on temporal variability of temperature. Seneviratne et al. [2006] found that the land-atmosphere coupling could contribute up to about two thirds of total summer temperature variance over the transitional zone in both recent and future European climates. While most previous studies focused on soil moisture feedback on average temperature, a few addressed soil moisture influence on daily maximum and minimum temperatures [Dai et al., 1999; Durre et al., 2000; Alfaro et al., 2006].

[5] Earlier research in soil moisture feedback on climate variation and variability (including many mentioned above) was largely based on simulations with atmospheric general circulation models (AGCMs), which may only provide limited information on land-atmosphere interactions and climate variability at a regional scale. On the basis of a comparison of observational and simulated relationships between land surface and atmospheric state variables at a few locations, Dirmeyer et al. [2006] suggested that most of the 12 participating AGCMs in the GLACE study do not represent the land-atmosphere coupling correctly. Nevertheless, available data are not sufficient to establish observational land-atmosphere relationships at regional and global scales, and the statistical approach applied to limited-area observations only can offer limited insight into physical mechanisms. Therefore, if enough care is taken in analysis of models' behavior, models still are useful for understanding the land-atmosphere relationship and provide capacity for the causality analysis. With high spatial resolution and realistic representation of key physical processes, regional climate models (RCMs) are more skillful at resolving land surface heterogeneity and other physical processes, and may thus better represent land-atmosphere interactions and the associated regional climate characteristics as compared to AGCMs [e.g., Dickinson et al., 1989; Giorgi, 1990; Leung and Ghan, 1998; Schär et al., 1999; Wang et al., 2000, 2003]. For example, Castro et al. [2007a, 2007b] found that the Regional Atmospheric Modeling System (RAMS) can add value to the representation of summer climate and long–term climate variability beyond the driving global reanalysis. By analyses of observations and a 20-year regional climate simulation, Qian and Leung [2007] demonstrated that their RCM can capture surface hydrological cycle and precipitation characteristics over East Asia.

[6] RCM simulations have previously been used to investigate the role of soil moisture conditions in North American summer climate [e.g., Giorgi et al., 1996; Bosilovich and Sun, 1999; Hong and Pan, 2000; Kanamitsu and Mo, 2003]. These studies, on the whole, exhibited a positive soil moisture feedback on precipitation over North America, complying with the GLACE result. Regarding soil moisture feedback on temperature, Diffenbaugh et al. [2005] recently demonstrated that soil moisture−temperature interactions enhance the occurrence of extreme high temperature events over the United States. The enhancement was also identified over the Europe [Diffenbaugh et al., 2007; Fischer et al., 2007].

[7] The purpose of the present study is to investigate the role of the land-atmosphere coupling in interannual summer climate variability over the contiguous United States using two long-term climate simulations produced by the Weather Research and Forecasting (WRF) model. Besides mean temperature (Tmean) variability, we are also motivated to examine if the role of the land-atmosphere coupling differs for maximum temperature (Tmax) versus minimum temperature (Tmin) because surface fluxes associated with large variation of solar radiation are quite different between day and night.

[8] The paper is organized as follows. Section 2 describes model and experiments, as well as methods to assess the land-atmosphere coupling. Section 3 evaluates the performance of the WRF model in simulating interannual climate variability, and identifies model biases. In section 4, climate variability is split into respective contributions induced by the land-atmosphere coupling and external factors (atmospheric circulation, SST). Section 5 examines if the results are model-dependent by comparing model's behavior with observational land-atmosphere relationships. Finally, conclusion and discussion are given in section 6.

2. Approach

[9] On the basis of previous experimental designs to uncouple land surface from the atmosphere [Koster et al., 2000; Seneviratne et al., 2006], two experiments are performed in this study: a control integration (CTL), which covers the period of March 1981 to August 1996, allows soil moisture vary freely according to the land surface scheme; an additional simulation (SoilM), in which the soil moisture evolution at each time step is replaced with the climatology of CTL, consist of 15-summer (1982–1996) integration which restarts on 1 June of each year and integrates for 3 months to 31 August. This effectively separates the contribution of the land-atmosphere coupling to interannual summer climate variability from the external forcings. The SoilM experiment shares the same model configuration as the CTL simulation.

[10] The RCM used in the study is the Weather Research and Forecasting (WRF) model version 2 [Skamarock et al., 2005] that has been adapted by Leung et al. [2006] for climate simulations. The physical parameterizations used include the WSM-5 class microphysics [Hong et al., 1998], the new Kain-Fritsch convective parameterization [Kain, 2004], the Community Atmospheric Model (CAM3) radiation package [Collins et al., 2006], the Yonsei University planetary boundary layer scheme [Noh et al., 2003], and the Noah land surface scheme [Chen and Dudhia, 2001].

[11] The Noah land surface scheme calculates soil processes at 4 soil layers with 10, 30, 60, and 100 cm thickness, and includes one canopy layer. It simulates soil moisture (both liquid and frozen), soil temperature, skin temperature, snowpack depth, snowpack water equivalent, canopy water content, and energy flux and water flux terms of surface energy balance and surface water balance.

[12] Model domain and topography utilized in the experiments, as well as the analysis domain are shown in Figure 1. The model domain covers the whole contiguous United States and the surrounding ocean areas, and extends far enough south to entirely include the Gulf of Mexico. The simulations use a horizontal grid spacing of 60 km and 31 vertical sigma levels. Initial condition and lateral and lower boundary conditions (SST) are derived from the NCEP-NCAR global reanalyses, and updated every 6 h. As described by Leung et al. [2006], our simulations include seasonally varying vegetation fraction and surface albedo. The buffer zone in the lateral boundaries consists of 10 grid points; a relaxation scheme is used to blend the boundary conditions with the model solution with a linear–exponential functional form for the relaxation coefficients. We use the first 15 months (from March 1981 to May 1982) as model spin-up period to minimize the initialization effects of soil moisture and soil temperature. The 15 summers that followed the spin-up period are the periods analyzed in our study.

Figure 1.

WRF model domain and topography (in meters), and analysis domain (marked by the rectangle) defined in this study.

[13] We use two methods, variance analysis and the GLACE-type coupling strength parameter, to objectively quantify the land-atmosphere coupling and its contribution to climate variability.

[14] In variance analysis, percentage of interannual variance of summer (June–August) mean climate for a variable V owing to the land-atmosphere coupling is estimated as follows:

equation image

where σV2(CTL) and σV2(SoilM) are interannual variances of summer mean V in the CTL and SoilM simulations, respectively. The coupling strength parameter method was initially used by Koster et al. [2000] and recently followed in the GLACE study and other studies. For each specific climate variable, we ignore the first 10 days and group the following 80 days into pentads (5-day periods). This is similar to the GLACE study and gives 16 pentads per summer. However, the following data arrangement is different from the GLACE study because the focal timescales are different in the two studies. We arrange each pentad for 15 years into one group (15 values in total), representing interannual variation of the climate variable at each pentad, and thus get 16-group data. The coupling strength parameter for the climate variable is defined as ΔΩVV (CTL) − ΩV (SoilM)), and ΩV value is calculated as follows:

equation image

where σV2 is the variance of pentad mean climate variable computed from all values available in 16-group data (across 16 group by 15 pentads or 240 values in total) and σV2 is the variance of pentad mean climate variable from the 16-group average time series (across 15 values in total). In our study, ΔΩV measures the degree to which the land-atmosphere coupling induces similarity of interannual changes for one specific climate variable between pentads. Consider, for example, a wet anomaly induced by a precipitation event at one pentad may in turn induce additional precipitation at subsequent pentads, and thus increase the precipitation similarity across pentads. In the SoilM simulation, such similarity should be absent as climatological land surface conditions are prescribed and the land surface is not allowed to respond to the precipitation event.

[15] Because the real world cannot present us the states of uncoupling of land and atmosphere system, the results obtained with the variance analysis and coupling strength analysis using the model outputs cannot be verified directly on the basis of observations. To check if simulated land-atmosphere relationships are dependent on the model, we further investigate the correlations between antecedent precipitation and temperature in simulations and observations.

3. Model Evaluation

[16] The observed daily Tmean, Tmax, and Tmin data are from the U.S. Historical Climatology Network (USHCN) [Karl et al., 1990]. For observed precipitation, we take the Climate Prediction Center (CPC) Merged Analysis of Precipitation (CMAP) data, which are derived from rain gauge observations and satellite estimates over the land [Xie and Arkin, 1997]. The precipitation data set exhibits similar interannual variability pattern and range over the United States to surface-based records [Ruiz-Barradas and Nigam, 2005]. Therefore we only present CMAP interannual summer precipitation variability in this study.

[17] We first look at the interannual variability of summer Tmean, Tmax and Tmin, expressed in terms of standard deviation with respect to the period of 1982–1996 (Figure 2). The Tmean, Tmax and Tmin variability in the observations has a similar spatial distribution with high values over the areas from the northern Rockies to the midwest/Ohio Valley region, which is well simulated by the WRF model. For magnitude, although the model produces larger Tmean and Tmax variability over the midwest/Ohio Valley region and adjacent areas, it simulates reasonably well Tmean and Tmax variability over other regions and Tmin variability over most areas of the United States as compared to observations.

Figure 2.

Standard deviation of temperature (in °C) during summer (June–August) based on 1982–1996 data from (left) U.S. Historical Climatology Network (USHCN) and (right) CTL: (a, d) Tmean, (b, e) Tmax, and (c, f) Tmin.

[18] Figure 3 shows 15-year summer mean precipitation, standard deviation of summer precipitation, and the coefficient of variation (standard deviation divided by the mean) in the observations from the CMAP and model simulations. The WRF model results are in good agreement with the observed mean precipitation over western United States, much of eastern United States, and northern Mexico, but show some differences in central United States and some eastern states. The WRF model reproduces poorly the observational local maximum over the midwest, and also underestimates precipitation over other areas in the Great Plains and over the eastern United States. The standard deviation of summer precipitation is greatly affected by the precipitation mean. Overall, it exhibits a similar pattern to the summer mean precipitation in both WRF simulation and observations. The WRF model simulates well a general increase in precipitation standard deviation from west to east, but fails to catch the local maximum over the midwest. Because the coefficient of variation can remove the dependency of the standard deviation on the mean precipitation, it is a more independent measure of interannual variability [Giorgi et al., 2004]. If we adopt this index, the WRF model can simulate well the spatial distribution of precipitation variability though some discrepancies exist over some areas. In particular, it reproduces the local maximum of precipitation variability over the midwest though with a too large magnitude. As is well known, it is a common difficulty for climate models to reproduce realistically Great Plains precipitation variations [e.g., Anderson et al., 2003], mainly attributed to complexity in diurnal rainfall associated with mesoscale convective complexes propagating from the Rocky Mountain [e.g., Carbone et al., 2002]. The WRF model exhibits poor skills in simulating Great Plains precipitation climatology but not the interannual variability (as measured by the coefficient of variation).

Figure 3.

The 1982–1996 summer mean precipitation, standard deviation of summer precipitation, and the coefficient of variation (standard deviation divided by the mean) in CMAP and CTL: (a, b) mean, (c, d) standard deviation, and (e, f) coefficient of variation.

[19] Precipitation is the dominant driver of surface water balance. We further examine the climatology of the model soil moisture and the interannual anomalies from that climatology. In a given land model, the range of absolute value of soil moisture depends on the maximum holding capacity, the field capacity, the soil moisture threshold below which transpiration starts to become soil moisture limited, and the soil wilting point. These values are closely associated with soil and other land surface properties. Therefore, spatial pattern of soil moisture climatology is not necessarily consistent with that of precipitation though, at a given grid cell, soil moisture is greatly determined by precipitation. Figure 4 shows 15-year summer mean and standard deviation fields of soil moisture in CTL. In general terms, mean total soil moisture tends to be high in forest-cover areas over the northwest and eastern United States. Note that the precipitation deficits over the Great Plains and the Gulf of Mexico coastal regions may cause the dry model biases in soil moisture. However, the biases could not be objectively judged owing to the lack of observations. The largest variability of soil moisture, as represented by its standard deviation, is found over eastern United States, the portion of the southwest, and parts of the northwest and northern Great Plains. Compared to mean precipitation impact on soil moisture climatology, the pattern of soil moisture variability is affected by that of precipitation to a larger extent. The ability of soil moisture to affect temperature and precipitation variability definitely depends on the value of soil moisture itself and its variability. With this in wind, we subsequently look at the role of the land-atmosphere coupling.

Figure 4.

The 1982–1996 summer mean total soil moisture and standard deviation of summer soil moisture in CTL: (a) the 1982–1996 summer mean total soil moisture (in centimeters) and (b) standard deviation of summer soil moisture (in centimeters).

4. Land-Atmosphere Coupling and Climate Variability

4.1 Coupling of Soil Moisture With Temperature

[20] Figure 5 shows differences between CTL and SoilM in standard deviations of summer Tmean, Tmax and Tmin. Because the land-atmosphere system is uncoupled and interannual variation of soil moisture is removed in SoilM, the fields reflect changes in summer temperature variability owing to the land-atmosphere coupling.

Figure 5.

Difference in standard deviation of summer temperature (in °C) between CTL and SoilM: (a) Tmean, (b) Tmax, and (c) Tmin. Grid cells with values significant at the 90% level by F-test are marked by the solid circles.

[21] Overall, the land-atmosphere coupling results in an increase in the standard deviation of summer Tmean, especially over a region extending from the southwest to the northern Great Plains to the southeast. There is a sharp contrast between Tmin and Tmax with respect to soil moisture influence. Changes in Tmax variability have a similar spatial distribution to that of Tmean, but with larger amplitude, whereas Tmin variability changes are generally small and insignificant. This is not unexpected since sensible and latent heat changes induced by soil moisture feedback are far larger during day than during night. The sharp day-night contrast dictates that Tmax and Tmin behave differently.

[22] We further quantify the role of the land-atmosphere coupling in interannual summer temperature variability using the variance method and the GLACE-type coupling strength parameter described in section 2. Figure 6 shows that the two methods present similar features.

Figure 6.

(left) Percentage of interannual summer temperature variance due to land-atmosphere coupling and (right) the GLACE-type coupling strength parameter: (a, d) Tmean, (b, e) Tmax, and (c, f) Tmin.

[23] 1. The land-atmosphere coupling makes a large contribution to summer Tmean variability in a zone extending from the southwest to the northern Great Plains to the southeast, explaining about 30–60% of the total interannual variance. In particular, the Tmean variability is greatly affected by soil moisture feedback over Arkansas, Tennessee, and portions of Mississippi, Alabama, Missouri, Illinois, Kentucky, and New Mexico.

[24] 2. The land-atmosphere coupling plays a leading role in summer Tmax variability almost over all regions where Tmean is highly influenced, generally accounting for more than 50% of the total variance.

[25] 3. Summer Tmin variability is highly constrained by external factors over most areas of the United States; soil moisture exerts some impacts mainly in southern California, Arizona and New Mexico that are influenced by the North American summer monsoon, and some isolated small areas.

[26] Compared with Figure 4b, the zone in which strong coupling of soil moisture with Tmean and Tmax exists also has high soil moisture variability in CTL. The existence of high soil moisture variability is a necessary, but not sufficient, condition for strong land-atmosphere feedback on temperature to exist, as a region with small variability would not be expected to have strong feedback. For example, some areas over the northwest that exhibit high soil moisture variability do not reveal strong soil moisture−temperature coupling. Further analyses find that the zone with strong coupling generally corresponds to the transitional zone between cold and warm climates (Figure 7b). Most of grid cells with ΔΩTmean > 0.08 and/or more than 40% of total variance owing to the land-atmosphere coupling fall within the temperature range of 23°–29°C. The sensitivity of soil moisture impacts to temperature regime (in addition to soil moisture regime) suggests that if a RCM or AGCM behaves unrealistically in simulating Tmean, it may be not able to represent the interactions between soil moisture and temperature correctly. Comparison with observations (Figure 7a) shows that the WRF model reproduces Tmean features properly despite a local warm bias over central United States and some other differences in magnitudes. With the warm bias, the overlapping region with the model transitional zone in the observations has colder temperature bounds of 1°–2°C compared to the CTL simulations over the central United States.

Figure 7.

The 1982–1996 summer mean Tmean (in °C) in (a) USHCN and (b) CTL. In Figure 7b, grid cells in Figure 6a with values larger than 40% and in Figure 6d with ΔΩTmean > 0.08 are marked with circles and crosses, respectively. Note that strong coupling mainly appears the transition zone between cold and warm climate (23–29°C in model and 22–27°C in observations).

[28] This leads us to ask why strong soil moisture−temperature coupling preferably appears in the transitional temperature zone with high soil moisture variability. A series of previous studies have demonstrated that the temperature variability is closely associated with evapotranspiration (ET) anomalies [e.g., Diffenbaugh et al., 2005; Koster et al., 2006a; Seneviratne et al., 2006]. The GLACE study used the product of the coupling strength parameter for ET and the standard deviation of ET to characterize the ability of a local ET signal to support land-atmosphere feedback [Guo et al., 2006]. It was found that an evaporation rate that varies strongly and consistently with soil moisture tends to lead to a higher coupling strength. Here we follow the same procedure to examine the coupling of soil moisture with ET and sensible heat (Figure 8). Betts [2004] found that the relationship between sensible heat and soil wetness is usually stronger than that for latent heat and soil wetness across several domains from the deep tropics to boreal forests. This characteristic is borne out in our study as geographical pattern of the diagnostic product for sensible heat agree to a larger degree with pattern of soil moisture anomalies (Figure 4b) than that of latent heat. Clearly, the product for sensible heat appears to explain well the geographical variation in the coupling of soil moisture with Tmean and Tmax. In addition, over the zone with strong soil moisture−temperature coupling the ET signal also tends to vary strongly and consistently with soil moisture. From the perspective of surface energy budget, soil moisture, along with other land surface conditions, determines the partitioning of available energy into latent heat and sensible heat. The sensible heat directly warms near-surface air while the latent heat is released at higher level to fuel precipitation. Therefore, it is not surprising that soil moisture anomalies affect temperature variability more via sensible heat than latent heat. The consistency between the diagnostic product for sensible heat and the soil moisture−temperature coupling also implies that soil moisture impacts on temperature are largely local.

Figure 8.

Distribution of the product of the GLACE-type coupling strength parameter and standard deviation: (a) latent heat and (b) sensible heat.

4.2. Coupling of Soil Moisture With Precipitation

[29] Figure 9 presents differences between CTL and SoilM in standard deviations of total precipitation, convective precipitation, and large-scale precipitation. Overall, the land-atmosphere coupling causes an increase in interannual summer variability of total precipitation and convective precipitation over northern United States (north of 36°N), eastern sea board, and the southwest. In contrast, neither positive nor negative sign dominates in central United States and other regions. Changes in large-scale precipitation variability, though significant at the 90% level by F-test over many areas, are generally small as compared to convective precipitation variability changes.

Figure 9.

Differences in standard deviation of summer precipitation (in mm/d) between CTL and SoilM: (a) total precipitation, (b) convective precipitation, and (c) large-scale precipitation. Grid cells with values significant at the 90% level by F-test are marked by solid circles.

[30] There are generally more spatial heterogeneity and uncertainty with respect to changes in precipitation variability than changes in temperature variability. For example, differences in standard deviations of precipitation between CTL and SoilM often jump from positive to negative values across very small areas over southern Great Plains and southeastern states. To avoid the effects of precipitation heterogeneity, model data are smoothed in each direction with a nine-point filter prior to quantifying the coupling of soil moisture with precipitation.

[31] Figure 10 examines the percentage of total interannual summer variance induced by the land-atmosphere coupling and the GLACE-type coupling strength parameter for total precipitation, convective precipitation, and large-scale precipitation using the spatially smoothed data. Both measures agree that the land-atmosphere coupling significantly contributes to total precipitation variability in a swath covering about 50% of northern United States, accounting for about half of the interannual variance. Further analysis of respective contributions of convective precipitation and large-scale precipitation shows that the effects of convection dominate the coupling with total precipitation over most areas of the swath, while large-scale precipitation is only important in north central United States and portion of the Great Lakes region. Similarly, the GLACE study also found that convective precipitation bears most of the signal of the soil moisture's impact on precipitation on the global scale, due in large part to the dominance of convective precipitation during boreal summer.

Figure 10.

(left) Percentage of interannual summer precipitation variance due to land-atmosphere coupling and (right) the GLACE-type coupling strength parameter: (a, d) total precipitation, (b, e) convective precipitation, and (c, f) large-scale precipitation.

[32] Although positive soil moisture feedback is dominant over most areas, negative contribution of soil moisture to precipitation variability is also seen, but limited to small areas. Possible reasons for negative changes may include large-scale effects, negative soil moisture feedback, and statistical sampling.

[33] From Figure 10, the locations of strong soil moisture−precipitation coupling are generally identical to those diagnosed from the GLDAS soil moisture analysis of Zhang et al. [2008]. However, there are some disagreements with the GLACE study that estimated a hot spot of the land-atmosphere coupling in the Great Plains over North America, although the northern Great Plains are identified as a common key region in the two studies. Dirmeyer et al. [2006] demonstrated that, compared to the observed relationships between surface and atmospheric state variables in few locations, most of the AGCMs in the GLACE study cannot validate well, suggesting that these models do not represent the land-atmosphere coupling correctly. Over the Great Plains, it is evident that the AGCMs often produce poor simulations of climate variations [e.g., Fennessy and Xue, 1997]. This implies that the estimated hot spot in the Great Plains in the GLACE study could be more uncertain. Indeed, only half of the participating models in the GLACE study showed strong soil moisture impacts over the Great Plains [Koster et al., 2004]. Ruiz-Barradas and Nigam [2005, 2006] found the dominance of large-scale moisture flux in accounting for Great Plains precipitation variations both from the observations and the North American Regional Reanalysis (NARR) [Mesinger et al., 2006], and suggested that the GLACE hot spot in the Great Plains could be attributed to the undue influence of a minority of models (3 out of 12).

[34] Meanwhile, the results of this study agree qualitatively with those of the GLACE study at two aspects. The two studies consistently agree that, as a whole, there exists a positive soil moisture−precipitation feedback over the contiguous United States. In addition, the GLACE found that strong land-atmosphere coupling appears over the transitional zones between dry and wet climates, emphasizing the role of soil moisture regime [e.g., Koster et al., 2006a]. Similarly, our results also show that strong coupling with temperature and precipitation are not visible over many wet areas over the northwest and the northeast and dry southern Great Plains in model.

[35] Strong soil moisture−precipitation coupling would not appear over the areas where highest product of the coupling strength parameter for ET and the standard deviation of ET occurs (Figure 8a). Rather, over regions of strong coupling the soil moisture only exhibits moderate ability to affect the ET signal. This suggests that the ET-precipitation link that involving the interactions among surface processes, the atmospheric boundary layer, and clouds also play an important role in the geographic variations of soil moisture–precipitation coupling strength.

5. Comparison With Observational Land-Atmosphere Relationships

[36] Validation of the simulated coupling is limited by a lack of observational soil moisture, ET, and other surface fluxes data at regional and global scales. Zhang et al. [2008] recently assessed Northern Hemisphere summer land surface and precipitation coupling using available observations and soil moisture data from the GLDAS. They found that strong coupling exists in the northern part of the United States. As mentioned in section 4.2, their diagnostic conclusion and the quantified coupling in our simulations verify well with each other, although the former focused on local subsurface soil moisture impacts, and estimated a smaller percentage of total variance owing to soil moisture feedback.

[37] To test if simulated relationships between soil moisture and temperature are model-dependent, we follows the same strategy used in previous studies [e.g., Huang and Van den Dool, 1993; Koster et al., 2003]. First, we examine correlations between antecedent precipitation and temperature in the observations, and a significant correlation is hypothesized to point to soil moisture impacts. We then determine if the significant correlation agrees with that calculated from the CTL simulation. Finally, the same calculation is performed using data from SoilM to determine whether or not the consistent correlation sign, if any, found in the observations and CTL, is attributed to soil moisture feedback.

[38] Here we use CMAP precipitation and Willmott and Matsuura [1995] temperature data. Prior to correlation calculations, observed temperature data at a resolution of 0.5° × 0.5° and model data are first aggregated onto the same 2.5° × 2.5° spatial resolution as CMAP precipitation. A three-point filter is then applied in both meridional and zonal directions at each grid cell with mean annual cycles and interannual trends of each month removed. Finally, correlations are calculated using June–July and July–August data (That is, 2 × 15 years = 30 samples in total). Note that the smoothness adopted allows us to obtain better significance because the correlations depend partly on spatial scales.

[39] Figure 11 presents the correlations between antecedent precipitation and temperature in the observations and CTL and SoilM simulations. Observational correlations are significant at 95% confidence level over northern and eastern United States and portion of the southwest. The overall pattern in CTL agrees with that in the observations to a large extent. At the same time, there exist some differences in the northern Rockies and midwest/Ohio Valley region, where CTL has stronger correlations. Figure 11c shows that the atmospheric circulation and SST can explain part of the correlations in the two regions. If their effects are removed from the CTL simulation, the pattern in Figure 11b tends to agree more with Figure 11a. On one hand, the better consistency indicates that simulated land-atmosphere relationships are not model specific. On the other hand, it suggests that most of the observed correlations are induced solely by soil moisture feedback.

Figure 11.

Correlations between antecedent precipitation (1-month lead) and temperature in (a) observations, (b) CTL, and (c) SoilM. Correlations of ±0.36, ±0.46, ±0.5, and ±0.57 are significant at the 95%, 99%, 99.5%, and 99.9% levels, respectively. The correlation is calculated using June–July precipitation and July–August temperature over the period 1982–1996 (That is, 2 × 15 years = 30 samples in total). Grid cells with correlation in SoilM significant at the 95% level are marked by solid circles in Figure 11b.

6. Conclusion and Discussion

[40] This study isolates the role of the land-atmosphere coupling in climate variability from external factors including atmospheric circulation and SST over the contiguous United States with two RCM simulations. The land-atmosphere coupling makes a significant (dominant) contribution to interannual summer Tmean (Tmax) variability over the zone extending from the southwest to the northern Great Plains to the southeast, but plays a very limited role in Tmin variability. Summer precipitation variability is dominated by the land-atmosphere coupling in a swath over northern United States. Convective precipitation, as a key component of the pathway linking soil moisture variations and precipitation, is more sensitive to land surface moisture variations than large-scale precipitation. Comparison with diagnosed land-atmosphere relationships from Zhang et al. [2008] and observational correlations between antecedent precipitation and temperature shows that the WRF model realistically represents the land-atmosphere coupling over the contiguous United States to a large extent.

[41] It is noteworthy that the existence of high soil moisture variability, though is a necessary, does not by itself guarantee a strong coupling of soil moisture with Tmean and Tmax. The coupling of soil moisture with Tmean and Tmax may also be sensitive to temperature regime (in addition to soil moisture regime), which was not identified in earlier AGCM studies. While sensible heat tends to be limited at low temperature, high temperature may make the soil dry out very quickly, and thus damp soil moisture anomalies. Therefore, in both cold and warm climates, either small soil moisture changes or small sensible heat changes limit the temperature response to soil moisture anomalies. This is traced to the coupling between soil moisture and sensible heat, which is larger in the temperature range of 23°–29°C between the cold and warm regimes. At a higher spatial resolution that resolves regional-scale forcings, the WRF model can represent more detailed climate features and more realistic land-atmosphere interactions, and may thus allow the role of temperature regime to stand out. Conversely, because the resolution of AGCMs is too coarse to capture some important regional climate details, they may lack the ability to identify the role of temperature regime in soil moisture−temperature coupling correctly.

[42] Strong coupling between soil moisture and precipitation appear over northern United States in which the ET signals are in general moderately, but not most highly, sensitive to soil moisture anomalies. This suggests that, in addition to the ability of soil moisture to affect ET, the ET-precipitation connection also plays an important role in soil moisture−precipitation feedbacks over the United States. Further analysis of variations of the atmospheric boundary layer and clouds will offer insight into this connection.

[43] Meanwhile, there are several relevant issues that warrant discussion. Although the WRF model generally reproduces the major characteristics of temperature and precipitation variability, it contains biases in simulating Tmean and Tmax variability in the midwest/Ohio Valley region and adjacent areas, and precipitation variability in the Great Plains and some other areas. In the midwest/Ohio Valley region which has large biases in temperature variability, soil moisture also has high variability and exhibits strong feedbacks. If too high soil moisture variability is simulated over this region, the model errors would be amplified by surface moisture effects. Further study is needed to improve our understanding of the complex interplay between the land-atmosphere coupling and the model errors. Previous studies demonstrated that RCM simulations of precipitation are highly sensitive to the choice of cumulus parameterization [e.g., Giorgi and Shields, 1999; Leung et al., 2003]. Liang et al. [2004] found that the Kain-Fritsch cumulus scheme yields a large precipitation deficit over the midwest when compared to the Grell scheme. Experiments with different physical representations especially cumulus options should be pursued in the future to clarify model uncertainties over areas with large precipitation biases. Giorgi and Bi [2000] found that the difference in summer mean rainfall, or internal model variability, due to random perturbations of initial atmospheric conditions can reach 5–15% of the average rainfall at the subregional level. They also compared the results of the perturbation experiments with corresponding simulations in which the Leaf Area Index (LAI) throughout the model domain was divided by a factor of 3. Precipitation changes induced by the decreasing LAI were essentially not distinguishable from those of the perturbation experiments. Qian and Leung [2007] suggested that the internal model variability may contribute to low RCM skill in simulating interannual summer precipitation anomalies. More research is clearly needed to identify that, to what extent, the internal model variability biases the contribution of land-atmosphere coupling to summer precipitation variability over the United States.

[44] Nevertheless, this study represents an early attempt to use long-term RCM simulations to assess the role of the land-atmosphere coupling in climate variability over the United States. The main conclusions are encouraging because they are not model specific, at least qualitatively, when compared to the observational land-atmosphere relationships. Given the importance of the land-atmosphere coupling to interannual summer climate variability, and hence initial land surface states to climate prediction at seasonal to interannual timescales, improved monitoring of soil moisture and land surface fluxes especially over key regions with strong land surface feedback is highly desirable.

Acknowledgments

[45] We thank two anonymous reviewers for helpful comments and suggestions. This work is supported by grants (to SUNY Albany) from the U.S. Department of Energy's Office of Science Biological and Environmental Research and the Climate Dynamics Division, National Science Foundation. Pacific Northwest National Laboratory is operated for the U.S. DOE by Battelle Memorial Institute under contract DE-AC06-76RLO1830.

Advertisement