Effects of Initial Soil Moisture on Rainfall Generation and Subsequent ...

644

JOURNAL OF HYDROMETEOROLOGY

VOLUME 10

Effects of Initial Soil Moisture on Rainfall Generation and Subsequent Hydrologic Response during the North American Monsoon ENRIQUE R. VIVONI AND KINWAI TAI Department of Earth and Environmental Science, New Mexico Institute of Mining and Technology, Socorro, New Mexico

DAVID J. GOCHIS National Center for Atmospheric Research, Boulder, Colorado (Manuscript received 25 June 2008, in final form 9 November 2008) ABSTRACT Through the use of a mesoscale meteorological model and distributed hydrologic model, the effects of initial soil moisture on rainfall generation, streamflow, and evapotranspiration during the North American monsoon are examined. A collection of atmospheric fields is simulated by varying initial soil moisture in the meteorological model. Analysis of the simulated rainfall fields shows that the total rainfall, intensity, and spatial coverage increase with higher soil moisture. Hydrologic simulations forced by the meteorological fields are performed using two scenarios: (i) fixed soil moisture initializations obtained via a drainage experiment in the hydrologic model and (ii) adjusted initializations to match conditions in the two models. The scenarios indicate that the runoff ratio increases with higher rainfall, although a change is observed from a linear (fixed initialization) to a nonlinear response (adjusted initialization). Variations in basin response are attributed to controls exerted by rainfall, soil, and vegetation properties for varying initial conditions. Antecedent wetness significantly influences the runoff response through the interplay of different runoff generation mechanisms and also controls the evapotranspiration process. The authors conclude that a regional increase in initial soil moisture promotes rainfall generation, streamflow, and evapotranspiration for this warm-season case study.

1. Introduction and background Soil moisture initialization in meteorological models can significantly affect the performance of weather forecasts (e.g., Mahfouf 1991; Beljaars et al. 1996; Scha¨r et al. 1999; Koster et al. 2004a). Since various studies have indicated a strong sensitivity of rainfall on soil moisture (e.g., Mintz 1984; Mo et al. 2006; Aligo et al. 2007), a proper initialization can potentially enhance precipitation predictability (Miyakoda et al. 1979; Beljaars et al. 1996; Scha¨r et al. 1999; Seuffert et al. 2002; Wang et al. 2007). Further, the spatial variation of soil moisture has been found to influence precipitation patterns at the regional scale across a range of different climates (e.g., Mahfouf et al. 1987; Avissar and Liu

Corresponding author address: Enrique R. Vivoni, Department of Earth and Environmental Science, New Mexico Institute of Mining and Technology, 801 Leroy Place, MSEC 244, Socorro, NM 87801. E-mail: [email protected] DOI: 10.1175/2008JHM1069.1 Ó 2009 American Meteorological Society

1996; Weaver 2004; Taylor et al. 2007; Kim and Wang 2007). Soil moisture effects on rainfall generation are especially important in transitional areas between dry and wet climates (Koster et al. 2004b). One such region is the southwest United States, where summer convective rainfall is promoted and evapotranspiration (ET) is sensitive to soil moisture during the North American monsoon (NAM; Hong and Pan 2000; Kurc and Small 2007; Vivoni et al. 2007c, 2008b). In this region, high soil moisture tends to lower surface albedo and the Bowen ratio (i.e., sensible to latent heat flux ratio), which in turn increases net radiation (Zheng and Eltahir 1998; Eltahir 1998). The increase of total available energy promotes a larger moist static energy and increased convective potential (Eltahir 1998; Pal and Eltahir 2001). It should be noted that the feedback between soil moisture and rainfall may be reversed under different conditions (Ek and Holstag 2004). In numerical simulations, soil moisture and the physical parameterizations related to convection have

JUNE 2009

VIVONI ET AL.

been shown to influence simulated rainfall (e.g., Xu and Small 2002; Gochis et al. 2002, 2003; Ratnam and Kumar 2005). Small (2001) and Xu et al. (2004) demonstrated a positive soil moisture–rainfall feedback in the NAM season (i.e., higher soil moisture leads to higher rainfall). Numerical studies in other regions (Bosilovich and Sun 1999; Hong and Pan 2000; Pal and Eltahir 2001; Oglesby et al. 2002; Schubert et al. 2004; Kim and Wang 2007) have also identified a positive feedback. The strength of this feedback is likely to depend on the location and season of interest, with more pronounced conditions during the summer (Scha¨r et al. 1999; Hong and Pan 2000), and in transition regions between dry and wet climates (Koster et al. 2004b; Wang et al. 2007). The soil moisture–rainfall feedback also has important implications on hydrologic forecasting at the watershed scale in the NAM region. It is well known that spatial soil moisture distributions in semiarid basins influence the prediction of surface fluxes (e.g., Kemp et al. 1997; Vivoni et al. 2008b) and runoff generation (e.g., Goodrich et al. 1994; Castillo et al. 2003). Less is known with respect to how soil moisture patterns promote rainfall generation that subsequently feeds back to soil water content. A study by Maxwell et al. (2007), however, showed that the soil moisture–rainfall feedback resulted in a more realistic soil moisture pattern simulated in a hydrologic model. Similarly, Seuffert et al. (2002) found that representation of detailed hydrologic processes improved simulations of precipitation and land surface fluxes, as compared to groundbased observations. Despite recent progress, the effect of soil moisture– rainfall interactions on basin-scale hydrologic predictability, in particular runoff generation, streamflow response, and evapotranspiration, has received limited attention. In fact, hydrologic simulations using meteorological model forcing have difficulty in yielding accurate streamflow predictions as a result of the stringent requirements in reproducing the space–time rainfall distribution (e.g., Chancibault et al. 2006; Hay et al. 2006). The propagation of rainfall forecast errors into the hydrologic response can be a significant challenge to overcome (e,g, Westrick and Mass 2001; Decharme and Douville 2006; Lin et al. 2006), in particular for distributed hydrologic models that are sensitive to small errors in the location or timing of forecasted precipitation (Vivoni et al. 2007a,b). Nevertheless, distributed models can be used to assess hydrologic predictability by transforming meteorological forecasts into detailed land surface states and fluxes, such as soil moisture, runoff production, and evapotranspiration. In this study, we perform a sensitivity analysis of the basin hydrologic response using a collection (or en-

645

FIG. 1. Schematic representation of the WRF and tRIBS offline one-way model coupling and initialization strategies.

semble) of meteorological model forcings generated by varying the initial soil moisture. We link the atmospheric model to a distributed hydrologic model in a one-way offline mode over a short forecast period (Fig. 1). Our main objectives are to (i) analyze the simulated rainfall fields under varying initial soil moisture in the meteorological model and (ii) evaluate the basin hydrologic response to the ensemble of meteorological forcings under a different set of initializations in the hydrologic model. In addition, we explore the differences in basin response induced by matching the initial soil moisture in the hydrologic and atmospheric models. This highresolution distributed modeling approach is an early attempt to quantify the effect of the soil moisture–rainfall feedback on runoff generation, streamflow, and evapotranspiration in the NAM region.

2. Methods a. Study area and simulation event The study site is the Upper Rıó Puerco (URP) basin in New Mexico (Fig. 2). We delineated the basin using a U.S. Geological Survey (USGS) 30-m digital elevation model (DEM) and defined a stream network with a drainage density of 0.25 km21 by comparing it with 1-m orthophotos of the region (Wyckoff 2007). The 1117km2 basin has a range of elevation from 1802 to 3222 m, with the Sierra Nacimiento along the eastern divide and decreasing elevations toward the basin outlet. The

646


VOLUME 10

FIG. 2. Study site represented at different modeling scales. (a) Nested domains for WRF model. Domain 1 (D01) has 30-km coarse grid spacing, nested domain 2 (D02) has 10-km grid spacing, and domain 3 (D03) has ;3.33-km finer grid spacing. (b) Location of Rıó Puerco basin in north-central New Mexico in D03.

URP is composed of a mixture of land use: forest (39%), shrubland (29%), and grassland (29%) (Wyckoff 2007). Soil texture is loam (43%), clay loam (33%), sandy loam (16%), unweathered bedrock (7%), and silt loam (;1%) (Wyckoff 2007). A long-term rain gauge is located in the URP at Cuba, NM. Being a semiarid environment, the URP basin receives an average annual rainfall of 323 mm yr21 in Cuba (Vivoni et al. 2006a). Streamflow at the basin outlet is measured by a USGS stream gauge (Rıó Puerco above Arroyo Chico near Guadalupe, NM). Other hydrologic observations are not available in the rural, semiarid basin. Vivoni et al. (2006a) studied a major storm event during 8–12 September 2003 in the URP based on Next Generation Weather Radar (NEXRAD), rain gauge, and stream gauge observations. The large monsoon storm developed over Arizona and moved into New Mexico, resulting in high rainfall accumulations and a major flood. Figure 3 presents the basin-averaged rainfall and flood response for this summer event. We selected this period for the numerical experiments because of the extended rainfall and significant flood response. The large storm size and duration possibly leads to higher predictability using an atmospheric model.

In addition, the storm timing in September facilitates simulations using a short forecast period, as high evapotranspiration resets soil moisture between storms. Although the study is limited to September 2003, the experiments provide insight into the soil moisture effects on rainfall and flood response in a broader NAM region.

b. Atmospheric model The atmospheric model used here is the Weather Research and Forecasting (WRF) model (Skamarock et al. 2005). For this study, the WRF model version 2.2 was set up with an outer domain and two nested grids at 30, 10, and 3.33 km, respectively (Fig. 2). Thirty vertical levels were used in the atmospheric columns. The model was driven by North American Regional Reanalysis (NARR) data at 32-km resolution (Fig. 1) at 3-h intervals for the initial and outer domain boundary conditions (Mesinger et al. 2006). The following parameterizations were used: (i) cloud microphysics from Thompson et al. (2004), (ii) longwave radiation based on the Rapid Radiative Transfer Model scheme (Mlawer et al. 1997), (iii) shortwave radiation using Dudhia (1989), (iv) the planetary boundary layer was based on Mellor– Yamada–Janjic approach (Janjic 2001), (v) the Noah

JUNE 2009

VIVONI ET AL.

647

FIG. 3. (top axis) Rainfall hyetograph and (bottom axis) streamflow hydrograph in the URP. The radar rainfall hyetograph (mm h21) represents the basin-averaged rainfall rate from 4-km NEXRAD Stage III hourly data. The discharge hydrograph (m3 s21) is shown for the USGS gauge in the Rıó Puerco above Arroyo Chico near Guadalupe. The runoff ratio computed using the flood volume (4.48 3 106 m3) and the radar rainfall volume (4.20 3 107 m3) yields a value of 0.106 (10.6%) (Vivoni et al. 2006a).

land surface model (LSM) was used (Chen and Dudhia 2001), and (vi) cumulus parameterization was based on the Kain–Fritsch scheme (Kain and Fritsch 1990). No convective parameterization was used for the finest (3.33 km) domain, leaving the model to attempt to resolve convective cloud systems at this scale. To vary the initial conditions in WRF, we multiplied the NARR volumetric soil moisture field by a factor (a) ranging from 0 to 2.25 [16 different cases, from fully dry (a 5 0) to fully wet (a 5 2.25); Fig. 1]. The soil column is represented using four layers (0–10, 10–40, 40–100, and 100–200 cm thick) and parameterized based on hydraulic properties inferred from soil texture. The a factor was applied across all soil layers in each domain and was allowed to evolve in the Noah LSM. The uniform multiplier ensures that a full range of soil moisture profile conditions was tested. Simulations with a , 1 capture conditions drier than the initial NARR values for the study period in the NAM region, while simulations with a . 1 represent wetter conditions. Figure 4 describes the statistical properties of the initial soil moisture field (m3 m23) for the top layer (0–10 cm) over the three domains. Soil moisture statistics are sensitive to the spatial distribution of soil hydraulic properties. Note that resolution changes across the domains lead to differences in soil texture that affect the statistical properties of soil moisture. The mean and maximum increase linearly with a (up to a 5 1.5) and then increase nonlinearly (up to a 5 2.25). This is a result of the initial soil moisture reaching the maximum

FIG. 4. Effect of the soil moisture multiplier (a from 0 to 2.25) on the initial soil moisture field in WRF for the top layer (0–10 cm). The statistical properties of the initial soil moisture for (a) domain 1, (b) domain 2, and (c) domain 3 include the minimum, maximum, and spatial mean volumetric soil moisture (m3 m23) with 61 standard deviation of the spatial soil moisture (m3 m23; vertical bars).

storage capacity, specified by the soil porosity, for high a. Minimum soil moisture for some domains remains constant, since there are areas with low porosity soils that are unaffected by a. The sharp decrease in minimum soil moisture (a 5 2.25 for domain 3) is due to an interaction with exposed rock. More importantly, the soil moisture multiplier allows for testing of a full range of

648


initial soil moisture that spans observed conditions in the region. Soil moisture initializations toward the drier end (a , 1) are considered more realistic with respect to field observations. For example, Vivoni et al. (2008b) found that NARR data (a 5 1) overestimates surface soil moisture with respect to observations at four different NAM ecosystems during the summer season. In this offline experiment, the initial soil moisture in WRF was not spun up to a new equilibrium by a longterm simulation or periodic forcing. As a result, the uniform soil moisture profile may not accurately depict the effect of soil wetting or drying. This simple strategy may introduce uncertainty in the atmospheric simulations, as the variation in soil moisture with depth has an effect on water and energy partitioning at the land surface. An alternative approach would be to carry out a short spin-up period to allow for the dynamic adjustments of the soil moisture profile. Li et al. (2007), for example, found that a 10-day spin up was sufficient to reach equilibrium during a NAM period. This strategy, however, minimizes differences in initial soil moisture among the simulations, as these would converge under the high summer evapotranspiration demands. Our objective in using the WRF model was to reasonably reproduce the sequence of convective cells leading to rainfall in the basin. Simulated rainfall accumulations were compared against NEXRAD data to assess model performance and to select the physical parameterizations, boundary conditions, and domain representation for the control run (a 5 1). We recognize the selection of different parameterizations would yield differences in the WRF simulations. However, previous research results in the NAM region by Gochis et al. (2002) support the selection of the physical parameterizations applied in this study.

c. Distributed hydrologic model The Triangulated Irregular Network (TIN)-based Real-time Integrated Basin Simulator (tRIBS) is a distributed hydrologic model designed to operate with forcing from rain gauges, weather radar, or meteorological models (Ivanov et al. 2004a,b; Vivoni et al. 2005, 2007a). Simulation of the coupled surface and groundwater response to the forcing is performed by tracking infiltration fronts, water table fluctuations, and lateral moisture fluxes in the vadose and saturated zones. Surface runoff is simulated via four mechanisms: infiltration excess and saturation excess runoff, perched return flow, and groundwater exfiltration. Runoff routing is performed through hydraulic channel routing and hydrologic overland flow. Evapotranspiration is computed from soil evaporation, plant transpiration, and evaporation of intercepted rainfall, which require radi-

VOLUME 10

ation and energy balance calculations based on the surface meteorological fields. The tRIBS domain is derived from a 30-m DEM using techniques described by Vivoni et al. (2004). For the URP, the TIN consists of ;127 000 nodes that preserve terrain characteristics, the channel network structure, and basin boundary. Approximately 8.9% of the original DEM nodes were preserved through the sampling of points in the TIN model. A constant upstream area threshold method was employed to classify DEM points as stream cells and retain these as the channel network in the model domain. We also delineated a floodplain region from the DEM and represented it within the TIN. This allows the relatively flat floodplain topography to be retained at high resolution in the TIN domain to capture near-stream saturation. The setup of the tRIBS model requires specification of spatially distributed data, including topography, soils, vegetation, bedrock depth, and the initial groundwater table position. We obtained these data from different geospatial sources, including the National Elevation Dataset (NED), the National Resource Conservation Service State Soil Geographic (STATSGO) database, and National Land Cover Data (NLCD). Model parameterization is carried out by specifying soil texture and vegetation properties for each classification derived from available data (note that soil and vegetation distributions used in tRIBS do not match the coarser fields assumed in the Noah LSM used in WRF). To limit the potential for overparameterization, within-class parameter variations are not allowed for soil and vegetation properties. Tables 1 and 2 present the parameter values for the vegetation and soil classes in the URP basin obtained from an extensive model testing exercise for the storm period of interest using NEXRAD estimates (Wyckoff 2007). Figure 5 presents the spatial distribution of surface soil texture, land cover, initial depth to groundwater, and the initial surface soil moisture (top 10 cm) used for the URP simulations. The initial soil moisture and depth to groundwater are related in the model physics through the assumption of hydrostatic equilibrium and depend on soil hydraulic properties (Vivoni et al. 2007a). This assumption is made in a range of coupled surface–subsurface hydrologic models to specify the initial soil moisture profile (e.g., Sivapalan et al. 1987; Troch et al. 1993; Salvucci and Entekhabi 1994). tRIBS simulations were conducted for an 8-day period (8–16 September 2003) to span the meteorological forecast and to allow for sufficient time for the flood pulse to exit the watershed. Model simulations contained identical parameterizations, except for the varying meteorological forcings from the WRF simulations and the specification of the initial soil moisture.

JUNE 2009

649

VIVONI ET AL.

TABLE 1. Land cover parameter values for the URP in the calibrated tRIBS distributed hydrologic model. Here, A is the percentage of basin area of this particular land cover type; P is the free throughfall coefficient (unitless); S is the canopy field capacity (mm); K is the drainage coefficient (mm h21); g is the drainage exponential parameter (mm21); Al is the albedo (unitless); h is the vegetation height (m); Kt is the optical transmission coefficient (unitless); Rs is the canopy-average stomatal resistance (s m21); and V is the vegetation fraction. Land cover parameters Land cover type

A (%)

P

S (mm)

K (mm h21)

G (mm21)

Al

H (m)

Kt

Rs (s m21)

V

Water Urban Bare soil Forest Shrubland Grassland Agriculture Wetlands

0.0 0.1 0.4 39.3 29.4 29.2 1.6 0.0

1.00 1.00 1.00 0.48 0.70 0.85 0.75 0.90

1.0 1.0 1.0 2.4 1.5 0.8 0.8 1.0

0.01 0.01 0.01 0.12 0.20 0.10 0.10 0.10

3.7 3.7 3.7 3.7 3.9 4.2 3.6 3.6

0.04 0.20 0.30 0.15 0.16 0.13 0.20 0.14

0.01 8.00 0.01 12.0 0.75 0.18 0.40 0.50

1.00 1.00 1.00 0.45 0.55 0.70 0.65 0.50

0.0 0.0 0.0 200 100 050 075 100

0.95 0.95 1.00 0.80 0.45 0.45 0.65 0.60

Because of the coupled surface–subsurface nature of the tRIBS model, the depth to the water table and soil hydraulic properties define the initial soil moisture profile (Ivanov et al. 2004a). Because of this relation, we ran a ‘‘drainage experiment’’ to create the initial groundwater table (Fig. 1). A drainage experiment is useful to initialize distributed surface and groundwater models when water table data are unavailable (e.g., Troch et al. 1993; VanderKwaak and Loague 2001; Marani et al. 2001). Drainage was conducted from a fully saturated catchment for a long period (;10 yr) in the absence of rainfall and evapotranspiration. This procedure leads to gravity drainage and the readjustment of the subsurface head field in the context of the basin geomorphology. Baseflow at the basin outlet is then related to the coincident water table distribution (Vivoni et al. 2007a). We selected the initial groundwater field (Fig. 5c) corresponding to the baseflow prior to the flood event, which subsequently determined the initial surface soil moisture (Fig. 5d). We performed a first set of tRIBS simulations (termed fixed initial conditions) using the same initial soil moisture field from the drainage experiment. In this scenario, the initial soil moisture in the hydrologic model remains fixed, despite the varying initial condi-

tions in WRF (Fig. 1). Subsequently, we adjusted the depth to groundwater in the hydrologic model to match the initial soil moisture between WRF and tRIBS. This was performed in an aggregated manner as a result of the coarse NARR soil moisture data. In this fashion, we matched the basin-averaged tRIBS and WRF soil moistures for each a (Table 3; further discussed in section 3c). As with the WRF initial soil moisture, we would expect that drier initializations (a , 1) are more realistic for the NAM region. We then performed a second set of tRIBS simulations (termed adjusted initial conditions) using the same meteorological fields as in the fixed initialization. The two sets of simulations were compared to explore the effect of initial soil moisture in the hydrologic model (Fig. 1). This comparison will also allow for assessing the need for adjusting the soil moisture conditions between meteorological and hydrologic models in offline simulations.

d. Numerical experiments A total of six spatially distributed meteorological variables from WRF are used to force the tRIBS model. Hourly meteorological forcing fields, including rainfall (mm h21), atmospheric pressure (mb), air temperature

TABLE 2. Soil parameter values for the URP in the calibrated tRIBS distributed hydrologic model. Here, A is the percentage of basin area of this particular soil type; Ks is the saturated hydraulic conductivity (mm h21); Qs is the soil moisture at saturation (unitless); Qr is the residual soil moisture (unitless); m is the pore distribution index (unitless); cb is the air entry bubbling pressure (mm); f is the conductivity decay parameter (mm21); As is the saturated anisotropy ratio (unitless); Au is the unsaturated anisotropy ratio (unitless); and n is the porosity (unitless). Soil parameters Soil type

A (%)

Ks (mm h21)

Qs

Qr

m

cb (mm)

f (mm21)

As

Au

n

Sandy loam Loam Unweathered bedrock Silty loam Clay loam

16.4 43.3 6.7 0.8 32.8

26.8 20.50 19.95 6.20 0.45

0.412 0.434 0.085 0.486 0.390

0.041 0.030 0.015 0.015 0.077

0.378 0.252 0.165 0.234 0.242

2146.2 2112.2 2373.3 2207.9 2258.9

0.00520 0.00625 0.00157 0.00650 0.00700

65 25 25 65 50

140 125 125 140 140

0.453 0.463 0.150 0.501 0.464

650


VOLUME 10

FIG. 5. Spatial distribution of the (a) surface soil texture classification (STATSGO), (b) land use or vegetation type (NLCD), (c) initial depth to the groundwater table (mm), and (d) initial volumetric soil moisture (m3 m23) in the surface layer (top 10 cm) for the URP basin as represented in the tRIBS model.

(8C), relative humidity (%), wind speed (m s21), and sky cover, are postprocessed from WRF output in the finest domain. Each field is a standard meteorological output obtained at or above the land surface at a height of 2 m with the exception of sky cover. Sky cover (XC) ranges from 0 (no clouds) to 10 (overcast) and was approximated according to Benjamin and Carlson (1986) as

XC 5 12.5(3.2RH

2.4),

(1)

where RH is the relative humidity (%). Sky cover is set to 10 if rainfall occurs and to 0 if RH is less than 75%. Alternative representations of sky cover are possible by analysis of the simulated solar radiation or cloud fields (e.g., Calbo´ et al. 2001; Otkin and Greenwald 2008).

JUNE 2009

651

VIVONI ET AL.

TABLE 3. The mean and standard deviation of the depth to initial groundwater table (mm) and the basin-averaged initial volumetric soil moisture (m3 m23) in the surface layer (top 10 cm) for each adjusted initialization case (a). Depth to initial groundwater table

Basin-averaged initial volumetric soil moisture

a

Mean (mm)

Standard deviation (mm)

Mean (m3 m23)

Standard deviation (m3 m23)

0.75 0.80 0.90 0.95 1.00 1.05 1.10 1.20 1.25 1.50 1.75 2.00 2.25

7522.3 5522.3 2522.3 1772.3 1522.3 1322.3 1022.3 772.3 622.3 221.8 0.0 0.0 0.0

293.7 293.7 293.7 293.7 293.7 293.7 293.7 293.7 293.7 233.1 2.6 2.6 2.6

0.20 0.21 0.24 0.26 0.27 0.28 0.30 0.32 0.34 0.40 0.44 0.44 0.44

0.06 0.05 0.06 0.06 0.06 0.06 0.07 0.08 0.08 0.09 0.07 0.07 0.07

Spatial output from WRF (3.3-km resolution) is resampled onto the URP domain. Model elements in tRIBS are Voronoi polygons (the neighborhood around TIN nodes), which form a spatially continuous field known as the Voronoi polygon network (VPN). WRF output is resampled to the VPN using a nearest neighborhood approach, as performed for NEXRAD rainfall data (see Ivanov et al. 2004a). Ensemble members of the hydrologic simulations were run for eight days. The first four days cover the primary runoff response to the storm. To ensure the streamflow exits the basin outlet, we extended the hydrologic simulations for four days, using zero rainfall padding and replicating the meteorological data from the first day, which exhibited clear conditions. Although this may yield uncertainty in evapotranspiration estimates, it permitted a large computational savings.

3. Results and discussion a. Effects of initial soil moisture on basin-scale rainfall properties We evaluated the meteorological fields over the simulation period to understand the differences in basinscale rainfall among the 16 cases with different initial soil moisture (a from 0 to 2.25). Our analysis focuses on rainfall forcing at the scale of the basin. We do not attempt here to provide a detailed assessment of the synoptic storm patterns, the dynamical mechanisms leading to rainfall, or the pathways through which soil moisture enhances precipitation, as these are beyond the scope of the current study. Instead, we highlight meteorological model results relevant to forcing the hydrologic simulations to generate streamflow forecasts with extended lead times.

We first compared the spatiotemporal distributions of basin rainfall obtained from the NEXRAD Stage III product (4-km, 1-h resolutions; Xie et al. 2005) to the WRF control simulation (a 5 1; 3.3-km, 1-h resolutions). Figure 6a presents the spatial map of total rainfall over the simulation (8–12 September 2003). Note the rainfall ranges of the WRF control run (15–82 mm) and NEXRAD (14–63 mm) are similar. Nevertheless, there are clear differences in the spatial pattern of total rainfall. Precipitation in the WRF simulation is organized with basin topography, with higher amounts along the Nacimiento Mountains and lower rainfall in the lowlands. NEXRAD exhibits lower precipitation along the mountain front. This inverse relation between WRF and NEXRAD rainfall in the mountain front is potentially a result of beam blockage effects from the Albuquerque radar site. Note that the maximum WRF rainfall accumulation occurs to the east of the maximum initial soil moisture in tRIBS (Fig. 5d), suggesting that a local interaction may subsequently occur between the storm rainfall and the initial conditions. The agreement between the two estimates is improved in the desert lowlands near the basin outlet. Figure 6b compares the hourly basin-averaged rainfall for the NEXRAD estimates and for the WRF simulations. Note the rainfall timing is similar during the peak storm activity on 10 September 2003, although the peak magnitudes for NEXRAD are larger than WRF by a factor of 1.5. Furthermore, the WRF simulation reproduces the distinct storm cells traversing the basin from west to east, with approximately the correct timings and interstorm spacing. Using these similarities, the WRF control simulation (a 5 1) is considered to perform reasonably well for the storm period of interest. As a result, the WRF parameterizations of the control run were adopted for the ensemble simulations.

652


VOLUME 10

FIG. 6. Rainfall comparison between NEXRAD product (4 km, 1 h) and WRF control simulation (a 5 1, 3.3 km, 1 h) for the (a) spatial distribution of total storm rainfall (mm), and (b) temporal distribution of basin-averaged rainfall rate (mm h21).

WRF simulations with different initial soil moisture were compared using a range of metrics. The mean areal rainfall over the URP was inspected (not shown) and found to increase considerably with initial soil moisture, with peaks of ;2 mm h21 for a 5 0 and ;6 mm h21 for a 5 2.25. This suggests that regional soil moisture influences rainfall and its intensification for this large storm event. Since the timing of rainfall peaks for individual storm cells were coincident, a stronger soil moisture effect on rainfall intensity was found. Temporal variations in the rainfall coverage (AR) in the URP basin were also inspected (not shown). This metric identifies the percentage of the basin area with rainfall greater than zero (R . 0). Relatively longer and more

pronounced rainfall coverages in the basin are associated with higher initial soil moisture. In addition, the percentage of time with full rainfall coverage in the basin (AR 5 100%) also increases with the initial soil moisture. This metric has important implications for the solar radiation at the land surface and thus the evapotranspiration response to the storm event. Figure 7 presents the spatial pattern of rainfall accumulation over the storm period (mm). Higher rainfall accumulations in most pixels are associated with higher initial soil moisture. Furthermore, a noticeable increase in the areal extent of the storm maximum is observed over the high terrain of the Nacimiento Mountains. This is consistent with the more pronounced rainfall coverage

JUNE 2009

VIVONI ET AL.

653

FIG. 7. Comparison of the spatial distributions of total rainfall accumulation during the storm period (mm), 8–12 September 2003, for the different initial soil moisture (a).

and indicates an important interaction between initial soil moisture and orographically induced rainfall. Note the spatial location of maximum rainfall accumulation tends to vary among the simulations but remains near the northeast portion of the basin. Interestingly, the rainfall accumulations in the northwestern section match the unobstructed NEXRAD observations quite well for a 5 1.5 or higher. Rainfall statistics in the URP basin were also computed as a function of initial soil moisture. Figure 8a shows the increase of the basin rainfall volume (m3)

with increasing a, suggesting a positive relation between rainfall generation and initial soil moisture. Note, however, that limited increases in rainfall volume occur for a , 0.25 and a . 1.5. This indicates that, for this particular event, initial soil moisture linearly controls simulated rainfall volume over the range a 5 0.25–1.5. Further gains in initial soil moisture do not contribute to enhancing the rainfall generation. Figures 8b,c present an overall increase in the maximum basin-averaged and pixel-scale hourly rainfall rates for higher soil moisture. Variations around the linear trends in these statistics

654


VOLUME 10

FIG. 8. Effect of the initial soil moisture multiplier on rainfall field statistics over the URP. (a) Total rainfall volume (m3). (b) Maximum of basin-averaged hourly rainfall (mm h21). (c) Maximum of pixel-scale rainfall rate (mm h21). (d) Total length of time (h) with rainfall coverage in 100% of the basin area.

result from the complex nature of the rainfall generation process and its sensitivity to initial conditions. Figure 8d shows the increase in total length (h) of full rainfall coverage in the basin with wetter initial conditions. In summary, the rainfall analysis indicates the enhancement effect of initial soil moisture on simulated rainfall duration, intensity, and spatial coverage. The observed threshold behavior in the soil moisture control on rainfall for a . 1.5 is likely the result of other limits on the convection processes. Kim and Wang (2007) found that increases of soil moisture beyond a certain level no longer promoted evapotranspiration as a result of radiation limitations. Note that the threshold does not occur for fully wet soils but rather at a soil moisture of ;0.3 m3 m23 (a 5 1.5). This indicates that soil moisture conditions within plausible ranges in the region affect rainfall generation within this modeling framework. Moreover, similarities in storm timing suggest the major differences in the simulations are in rainfall intensity, location, and coverage.

b. Effect of rainfall forcing on hydrologic response under fixed initializations We used the WRF simulations to force the distributed hydrologic model under fixed initial conditions (16 cases). We first inspect the cumulative rainfall volume,

basin-averaged surface soil moisture, and the cumulative discharge in Fig. 9. For comparison, we show the results using the NEXRAD data as forcing as well as the ensemble mean of the 16 simulations. Cumulative rainfall volume in each simulation share similar temporal patterns but exhibit large differences in total volume (a factor of 3.7 between a 5 0 and 2.25) (Fig. 9a). This results in large variations in the basin-averaged relative soil saturation (s), also referred to as soil moisture (Fig. 9b). Soil moisture differs in temporal distribution and magnitude, with a general increase for higher rainfall. However, low a (a 5 0, 0.2) or high a (a . 1.5) cases have similar soil moisture as a result of small changes in rainfall and the fixed initial conditions. The range of soil moisture at the end of the simulations is small (s 5 0.5–0.6) as compared to the range during the peak storm activity (s 5 0.63–0.83). This suggests the drydown period homogenizes soil moisture across the different cases as a result of evapotranspiration and lateral redistribution, consistent with low soil moisture variability for dry conditions in semiarid areas (MartinezFernandez and Ceballos 2003; Vivoni et al. 2008a). Figure 9c shows the cumulative discharge at the URP outlet, the NEXRAD-based predictions, and the observed USGS discharge. Cumulative discharge generally increases for cases of higher initial soil moisture and

JUNE 2009

VIVONI ET AL.

FIG. 9. Comparison of basin response for simulations with different WRF initial soil moisture (a 5 0–2.25) and fixed initial soil moisture in tRIBS. (a) Cumulative basin-averaged rainfall volume (m3). (b) Basin-averaged soil moisture (top 10 cm), expressed as relative saturation (s 5 u/n), where u is the volumetric soil moisture content (m3 m23) and n is the soil porosity. (c) Cumulative discharge (m3) at URP outlet. Note that the NEXRAD forcing and its hydrologic response are included for comparison. The ensemble mean over the 16 different cases is shown (solid black lines). Streamflow observations at the USGS gauge are also shown in (c) for comparison (thick gray line).

655

has a large range (;9 3 106 m3). Clearly, the differences in rainfall propagate to the discharge predictions and are more sensitive than the basin-averaged soil moisture. This is due to the effect of variations in pixel-scale rainfall intensity on the infiltration excess runoff, the primary mechanism for overland flow in the model parameterization (Wyckoff 2007). Rainfall exceeding the infiltration capacity of the soil results in runoff production and streamflow generation. For low a most of the rainfall infiltrates into the soil, leaving a small amount available for runoff; however, for high a a larger fraction of the rainfall exceeds the soil infiltration capacity and produces runoff. This is consistent with the soil properties and rainfall characteristics observed in other semiarid environments (e.g., Beven 2002; Castillo et al. 2003). Figure 10 presents spatial maps of the percent of time with infiltration excess runoff occurrence (%). More frequent infiltration excess runoff is associated with higher a because of the greater rainfall intensity simulated at the pixel and basin scales. Note the locus of runoff production is in the north-central part of the basin characterized by clay loam soils (Fig. 5a). The spatial distribution of infiltration excess runoff matches well with the soil texture pattern. There is no infiltration excess runoff in areas of high saturated hydraulic conductivity (i.e., loam and sandy loam), whereas in the clay loam region, the infiltration excess frequency increases with higher rainfall (Fig. 7). This highlights the dual effect of rainfall intensity and soil hydraulic properties on the site of maximum infiltration excess runoff occurring in the foothills of the Sierra Nacimiento. Although rainfall intensities are higher in the mountain front, these areas generated reduced runoff because of more permeable soils. Note that infiltration excess runoff inside the clay loam region also exhibits slight effects of underlying differences in vegetation and topography. In summary, precipitation variations induced by the initial soil moisture in WRF propagates to the soil moisture and discharge predictions issued by tRIBS. Under fixed initial conditions, infiltration excess runoff is the primary mechanism and is governed by rainfall intensity and soil hydraulic properties. High conductivity soils are less sensitive to changes in rainfall as compared to the low conductivity soils over the tested range of rainfall intensities. As a result, low conductivity soils are responsible for the infiltration excess runoff induced by rainfall changes in the fixed initialization cases.

c. Effect of rainfall forcing on hydrologic response under adjusted initializations In the adjusted initialization cases, we matched the basin-averaged surface soil moisture in the tRIBS and

656


VOLUME 10

FIG. 10. Spatial distributions of the percent of time with infiltration excess runoff occurrence (%) during 8–16 September 2003 for the fixed initial conditions.

WRF models. This was achieved by uniformly increasing or decreasing the groundwater table until a matching surface soil moisture was obtained. This procedure allows a more realistic representation of the effect of initial soil moisture in the hydrologic response, while preserving the spatial patterns in each model. Clearly, this method yields differences in the soil moisture profile as compared to the uniform application of a with depth in WRF. The major difference would be an increase in soil moisture deficit in the unsaturated zone in

tRIBS, as the soil moisture profile above the water table has an exponential shape rather than a uniform value. As shown in Table 3, we were able to match 13 of the simulation cases (except the drier cases, where a 5 0–0.5). For the dry cases, the initial soil moisture in tRIBS could not be decreased under the current physical parameterization. For the wet cases (a 5 1.75–2.25), the basin is close to saturation (the water table is near the surface) in order to match the initial conditions imposed in WRF.

JUNE 2009

VIVONI ET AL.

FIG. 11. Comparison of basin hydrologic response for simulations with adjusted initial soil moisture conditions in tRIBS and WRF (a 5 0.75–2.25). (a) Cumulative basin-averaged rainfall volume (m3). (b) Basin-averaged soil moisture (top 10 cm), expressed as s. (c) Cumulative discharge (m3) at the URP outlet.

Figure 11 shows the cumulative rainfall, basin-averaged surface soil moisture, and cumulative discharge for the adjusted initializations. Basin-averaged soil moisture differs in temporal distribution and magnitude among the cases, indicating a general increase with higher a. Some cases with low a (a 5 0.75, 0.85) or high a (a . 1.75) have similar patterns as a result of similar rainfall and initial conditions. By design, soil moisture at the beginning of

657

the simulation varies to match the WRF initialization (s 5 0.48–1). Soil moisture traces converge during the peak storm period. The range of final soil moisture among the cases is larger (s 5 0.53–0.77) than for the fixed initialization (Fig. 9b), suggesting that the initial condition affects the entire simulation period. Interestingly, the soil moisture in the last four days decreases continuously as the surface dries and the simulations once again diverge. This divergence is because of the differences in drainage for different soil moisture contents. Soil moisture also affects evapotranspiration rates, but this in general would lead to a convergence of simulations during the interstorm period. Figure 11c presents the cumulative outlet discharge. This can be compared to Fig. 9c by the common USGS observations and NEXRAD-based simulations. As in the fixed initialization, cumulative discharge increases with the initial soil moisture in WRF. A major difference between the two scenarios is the magnitude of the discharge response. The adjusted initializations lead to a more pronounced streamflow, including a larger range of cumulative discharge (;3.5 3 107 m3). This suggests that matching the soil moisture initialization has a significant effect on soil moisture and runoff predictions. It is important to understand the reasons in the hydrologic model for the observed increase in discharge. Recall the fixed initialization cases only produce infiltration excess runoff. Next, we explore whether the increase in discharge can be attributed to changes in the runoff mechanisms induced by higher antecedent wetness. Figure 12 presents spatial maps of the percent of time with saturation excess runoff occurrence (%) for the adjusted cases. Saturation excess runoff arises when precipitation falls on soil columns that are saturated from below as a result of a rise in the water table. For dry cases (a 5 0.75–0.90), there is no saturation excess runoff and infiltration excess is the primary mechanism. For a . 0.95, however, saturation excess runoff increases as a result of the wet conditions in the hydrologic model. Saturation excess runoff becomes the major mechanism for a . 1.75. As the surface approaches saturation, there is limited infiltration capacity and runoff is produced. In contrast to Fig. 10, the spatial map of saturation excess runoff for a 5 1.50–2.25 does not match the soil texture field, since the surface saturation reduces the role of soil hydraulic properties. As a result, saturation excess runoff is sensitive to rainfall patterns for wet initial conditions (a 5 1.50–2.25). The spatial variability in saturation excess runoff is also influenced by vegetation patterns (Fig. 5b), due to the role of plants in intercepting rainfall prior to it reaching the saturated surface. In addition, topographic redistribution of initial soil moisture is apparent in the spatial

658


FIG. 12. Spatial distributions of the percent of time with saturation excess runoff occurrence (%) during 8–16 September 2003 for the adjusted initial conditions.

VOLUME 10

JUNE 2009

VIVONI ET AL.

runoff patterns, in particular for intermediate conditions (e.g., a 5 1.20). It is important to summarize how the fixed and adjusted initializations affect the basin hydrologic response. Figure 13a compares the runoff ratio (r 5 Q/P) variations, where Q and P are the total discharge and rainfall volumes, respectively, with rainfall volume for the two scenarios. The fixed cases exhibit a linear increase in runoff ratio over the range of rainfall volume. This increasing trend is less abrupt as compared to the adjusted initialization cases, despite that the same meteorological fields were used. For the fixed initial conditions, once the maximum infiltration capacity is exceeded by the rainfall intensity, a proportional increase in runoff is produced via infiltration excess. In the adjusted initialization, the variation of the water table position leads to a shift toward saturation excess runoff. This transition leads to a nonlinear rise in r. As shown in Fig. 13b for the adjusted cases, infiltration excess runoff is dominant for dry conditions, whereas wetter conditions lead to saturation excess runoff. The behavior in the two scenarios suggests that the mode of runoff production can significantly change with the antecedent wetness. It also shows that the basin response changes from linear (fixed initialization) to nonlinear behaviors (adjusted initialization). Given the identical forcings, we interpret these differences to be due to the effect of soil moisture initialization on the runoff production mode. This is consistent with previous field studies, for example, by Western et al. (2004) and Merz and Bardossy (1998), who found that different initial soil moisture patterns can produce variable runoff responses.

d. Effect of soil moisture initialization on evapotranspiration response In addition to changes in soil moisture and streamflow, we are interested in how the land surface returns water vapor to the atmosphere. While this is not directly captured in one-way simulations, it is possible to assess variations in ET over a range of different initial conditions. Figure 14 compares the basin-averaged ET expressed in total volume (m3) to match the rainfall and streamflow metrics introduced previously. Similar diurnal fluctuations are observed in the two initialization scenarios, as these have the same meteorological forcings and a two-way feedback is not captured. In the fixed initialization, all cases are close to the ensemble mean ET, whereas a wider range is observed among the adjusted cases. This suggests that perturbations in the ET simulations are primarily limited by soil moisture, as opposed to available energy. In general, adjusted initializations exhibit greater ET and higher sensitivity to initial soil moisture. The spread among the adjusted

659

FIG. 13. Effects of rainfall volume on the runoff response. (a) Here, r is for fixed and adjusted initializations. (b) Runoff contributions from individual runoff mechanisms as percentage of total runoff for the adjusted initial conditions.

cases is minimized during the peak runoff and maximized during the early and final periods. This is due to the effect of higher soil moisture on soil evaporation and plant transpiration in the hydrologic model (see Ivanov et al. 2004a). Figure 15 presents a comparison of the total ET and evapotranspiration ratio (ET/P) for the two initialization scenarios as a function of rainfall volume. Total ET for the fixed cases is nearly constant, with increasing rainfall as a result of relatively small differences in antecedent soil moisture. In the adjusted cases, however, an increasing trend of ET is observed with higher rainfall as a result of increased water availability. Note that changes in rainfall volume are greater than changes in ET volume, leading to a decreasing trend of ET/P with rainfall (Fig. 15b). The decreasing ET/P trend is less pronounced for the adjusted initializations, as the higher soil moisture promotes greater ET. Ratios of ET/P greater than 100% imply that ET consumes available soil water storage (e.g., unsaturated zone moisture or groundwater) beyond the available rainfall during the storm event.

660


FIG. 14. Comparison of basin-averaged evapotranspiration volume (m3) for (a) simulations with different WRF initial soil moisture (a 5 0–2.25) and fixed initial soil moisture in tRIBS, and (b) simulations with adjusted initial soil moisture conditions in tRIBS and WRF (a 5 0.75–2.25).

Figure 16 presents a comparison of the spatial maps of time-averaged evaporative fraction (EF) for the fixed and adjusted initializations. EF is the ratio of latent heat (lE) to total turbulent heat (lE 1 H) fluxes, where H is sensible heat flux. Spatial distributions of EF are similar for the fixed initializations and match the soil patterns well, suggesting EF is influenced by soil hydraulic properties for dry conditions. In contrast, the adjusted initializations have a gradually changing EF pattern with increasing initial soil moisture, corresponding to the spatial distribution of vegetation. In particular, the western region with grasslands and shrublands has high EF, while forested mountains exhibit lower EF. As a result, vegetation plays an important role in shaping the evapotranspiration response for high initial soil moisture. Clearly, the method of initializing the hydrologic model influences the partitioning of surface heat fluxes in the one-way offline experiments.

VOLUME 10

FIG. 15. Effects of rainfall volume on basin ET. (a) Total ET volume for the fixed and adjusted initializations. (b) ET/P (%) as a function of rainfall volume.

4. Summary and conclusions The primary objectives of this study were to understand how and to what extent initial soil moisture variations affects rainfall generation and its subsequent hydrologic response in the NAM region through a modeling framework. This region has been identified as having a strong relation between soil moisture and rainfall (e.g., Xu et al. 2004) as a result of the seasonal transition in water availability during the summer (e.g., Vivoni et al. 2007c, 2008b). We generated an ensemble of meteorological fields over the NAM region by varying soil moisture in the WRF model. We then focused on analyzing the rainfall characteristics originating from each WRF initialization at the scale of a regional basin experiencing a large monsoon storm and flood event. To study the effect of hydrologic initialization, we tested two scenarios: (i) a fixed soil moisture initialization in tRIBS and (ii) adjusted initializations in tRIBS to match the WRF initial soil moisture. Subsequently, soil moisture, streamflow, runoff generation, and evapotranspiration responses were compared using basin-averaged

JUNE 2009

VIVONI ET AL.

661

statistics and spatial distributions of simulated variables. Model interpretations take advantage of the distributed modeling approach that produces spatiotemporal predictions, which can be related to the underlying land surface properties. Study results support the following conclusions:

FIG. 16. Effect of initial soil moisture conditions on EF for (left) fixed and (right) adjusted initializations. Only a selected number of a cases are shown for clarity.

1) The uniform application of a soil moisture multiplier (a) in WRF generates a nearly linear increase in the initial soil moisture over all soil depths of the three domains, up to a threshold value of a 5 1.5. Beyond this range, a nonlinear (asymptotic) rise is observed in soil moisture because of the upper limit determined by soil porosity. 2) The increase in initial soil moisture over a large domain generally increases the total simulated rainfall in this warm-season case study, consistent with previous work in the NAM region (e.g., Small 2001; Xu et al. 2004). Given our focus on a single storm event, we identified that the meteorological simulation cases have similar storm timings, but they varied in terms of the rainfall magnitude, storm location, and spatial coverage. 3) The fixed initialization cases exhibited a linearly increasing runoff ratio because of the dominance of infiltration excess runoff. The occurrence of infiltration excess runoff agreed well with the soil texture map. For soils of low permeability, rainfall intensities determined runoff. As a result, both rainfall and soil hydraulic properties control the spatial distribution of infiltration excess runoff and streamflow at the outlet. 4) The adjusted initialization cases exhibited a nonlinearly increasing runoff ratio because of a transition in runoff mechanisms. In particular, runoff production was driven by saturation excess runoff in the wetter conditions. As a result, rainfall and initial soil moisture play dominant roles in shaping the basin runoff response, with soil hydraulic properties playing a role in dry antecedent wetness. 5) Higher antecedent soil moisture induced by rainfall generally increases the evapotranspiration response, as quantified by the ratio ET/P and the evaporative fraction (EF). For low soil moisture, soil hydraulic properties play a significant role in the spatial distribution of EF. In contrast, for saturated initial conditions, vegetation patterns are more influential in evapotranspiration. A transition in the surface flux partitioning was identified in the basin when matching initial conditions were specified in the two models. The results of this study are based on using two physically based models applied to a large storm event and its flood response during the NAM. Although the

662


use of a single event is somewhat limiting, we focus on the atmospheric and hydrologic responses across a range of different soil moisture initializations that should be applicable in the broader NAM region. The general nature of these results for other monsoon storms, seasons, and watershed characteristics in NAM region requires further study, for example, within the hydroclimatological areas identified by Gochis et al. (2006). Our work also depends on the models’ capabilities to simulate the relevant physical processes and our strategies to initialize the soil moisture conditions adequately in each model. Although simplified, our approach enables testing the effect of adjusting the hydrologic initialization to account for imposed soil moisture changes in the atmospheric model. Results clearly indicate that matching the model initial conditions in one-way offline experiments is important when studying the hydrologic responses to the soil moisture– rainfall feedback. The underlying physical reason is that antecedent moisture in the watershed can significantly modulate the dominant runoff generation mechanism and the partitioning of surface heat fluxes. Through the use of the two different initialization scenarios, we demonstrate that streamflow and evapotranspiration responses vary significantly because of the interplay of competing land surface processes as a function of initial soil moisture (i.e., infiltration versus saturation excess runoff and sensible versus latent heat flux). Our results are particularly encouraging, as they point to the importance of representing the soil moisture– rainfall feedback for improved hydrologic predictability, an effort that has not been previously undertaken in the NAM region. In fact, few studies have linked the effect of enhanced precipitation as a result of wet soil moisture to its hydrologic consequences. In this respect, this study contributes to our understanding of how the soil moisture–rainfall feedback affects the soil moisture, runoff generation, streamflow, and evapotranspiration response at the basin scale. Clearly, this should be explored in more detail with a two-way coupled modeling system (e.g., Seuffert et al. 2002; Maxwell et al. 2007) that can represent the hydrologic response in detail and its direct effects on the atmospheric states by naturally matching the soil moisture conditions in the two models. Acknowledgments. This research was sponsored by the U.S. Army Research Office (Contract 47334-EVYIP). We thank the National Center for Atmospheric Research for providing computing and technical support. We also appreciate the helpful comments from three anonymous reviewers. Discussions with Joseph Galewsky, Giuseppe Mascaro, Ricardo Mantilla, David

VOLUME 10

J. Raymond, John L. Wilson, and Wei Yu are greatly appreciated. REFERENCES Aligo, E. A., W. A. Gallus, and M. Segal, 2007: Summer rainfall forecast spread in an ensemble initialized with different soil moisture analyses. Wea. Forecasting, 22, 299–314. Avissar, R., and Y. Liu, 1996: Three-dimensional numerical study of shallow convective clouds and precipitation induced by land surface forcing. J. Geophys. Res., 101, 7499–7518. Beljaars, A. C. M., P. Viterbo, M. J. Miller, and A. K. Betts, 1996: The anomalous rainfall over the United States during July 1993: Sensitivity to land surface parameterization and soil moisture. Mon. Wea. Rev., 124, 362–383. Benjamin, S. G., and T. N. Carlson, 1986: Some effects of surface heating and topography on the regional severe storm environment. Part I: Three-dimensional simulations. Mon. Wea. Rev., 114, 307–329. Beven, K. J., 2002: Runoff generation in semi-arid areas. Dryland Rivers: Hydrology and Geomorphology of Semi-Arid Channels, L. J. Bull and M. J. Kirkby, Eds., John Wiley and Sons, 57–105. Bosilovich, M. G., and W. Y. Sun, 1999: Numerical simulation of the 1993 Midwestern flood: Local and remote sources of water. J. Geophys. Res., 104, 19 415–19 423. Calbo´, J., J.-A. Gonza´lez, and D. Pages, 2001: A method for skycondition classification from ground-based solar radiation measurements. J. Appl. Meteor., 40, 2193–2199. Castillo, V. M., A. Gomez-Plaza, and M. Martinez-Mena, 2003: The role of antecedent soil water content in the runoff response of semiarid catchments: A simulation approach. J. Hydrol., 284, 114–130. Chancibault, K., S. Anquetin, V. Ducrocq, and G.-M. Saulnier, 2006: Hydrological evaluation of high-resolution precipitation forecasts of the Gard flash-flood event (8-9 September 2002). Quart. J. Roy. Meteor. Soc., 132, 1091–1117. Chen, F., and J. Dudhia, 2001: Coupling an advanced land surface– hydrology model with the Penn State–NCAR MM5 modeling system. Part I: Model implementation and sensitivity. Mon. Wea. Rev., 129, 569–585. Decharme, B., and H. Douville, 2006: Uncertainties in the GSWP-2 precipitation forcing and their impacts on regional and global hydrological simulations. Climate Dyn., 27, 695–713. Dudhia, J., 1989: Numerical study of convection observed during the winter monsoon experiment using a mesoscale twodimensional model. J. Atmos. Sci., 46, 3077–3107. Ek, M. B., and A. A. M. Holstag, 2004: Influence of soil moisture on boundary layer cloud development. J. Hydrometeor., 5, 86–99. Eltahir, E. A. B., 1998: A soil moisture rainfall feedback mechanism 1. Theory and observations. Water Resour. Res., 34, 765–776. Gochis, D. J., W. J. Shuttleworth, and Z. L. Yang, 2002: Sensitivity of the modeled North American monsoon regional climate to convective parameterization. Mon. Wea. Rev., 130, 1282–1298. ——, ——, and ——, 2003: Hydrometeorological response of the modeled North American monsoon to convective parameterization. J. Hydrometeor., 4, 235–250. ——, L. Brito-Castillo, and W. J. Shuttleworth, 2006: Hydroclimatology of the North American monsoon region in northwest Mexico. J. Hydrol., 316, 53–70. Goodrich, D. C., T. J. Schmugge, T. J. Jackson, C. L. Unkrich, T. O. Keefer, R. Parry, L. B. Bach, and S. A. Amer, 1994: Runoff simulation sensitivity to remotely sensed initial soil water content. Water Resour. Res., 30, 1393–1405.

JUNE 2009

VIVONI ET AL.

Hay, L. E., M. P. Clark, M. Pagowski, G. H. Leavesley, and W. J. Gutowski, 2006: One-way coupling of an atmospheric and a hydrologic model in Colorado. J. Hydrometeor., 7, 569–589. Hong, S.-Y., and H.-L. Pan, 2000: Impact of soil moisture anomalies on seasonal, summertime circulation over North America in a regional climate model. J. Geophys. Res., 105, 29 625– 29 634. Ivanov, V. Y., E. R. Vivoni, R. L. Bras, and D. Entekhabi, 2004a: Catchment hydrologic response with a fully distributed triangulated irregular network model. Water Resour. Res., 40, W11102, doi:10.1029/2004WR003218. ——, ——, ——, and ——, 2004b: Preserving high-resolution surface and rainfall data in operational-scale basin hydrology: A fullydistributed physically-based approach. J. Hydrol., 298, 80–111. Janjic, Z. I., 2001: Nonsingular implementation of the MellorYamada level 2.5 scheme in the NCEP Meso model. NCEP Office Note 437, 61 pp. Kain, J. S., and J. M. Fritsch, 1990: A one-dimensional entraining detraining plume model and its application in convective parameterization. J. Atmos. Sci., 47, 2784–2802. Kemp, P. R., J. F. Reynolds, Y. Pachepsky, and J. L. Chen, 1997: A comparative modeling study of soil water dynamics in a desert ecosystem. Water Resour. Res., 33, 73–90. Kim, Y., and G. Wang, 2007: Impact of initial soil moisture anomalies on subsequent precipitation over North America in the coupled land–atmosphere model CAM3–CLM3. J. Hydrometeor., 8, 513–533. Koster, R. D., and Coauthors, 2004a: Realistic initialization of land surface states: Impacts on subseasonal forecast skill. J. Hydrometeor., 5, 1049–1063. ——, and Coauthors, 2004b: Regions of strong coupling between soil moisture and precipitation. Science, 305, 1138–1140. Kurc, S. A., and E. E. Small, 2007: Soil moisture variations and ecosystem-scale fluxes of water and carbon in semiarid grassland and shrubland. Water Resour. Res., 43, W06416, doi:10.1029/2006WR005011. Li, J., X. Gao, and S. Sorooshian, 2007: Modeling and analysis of the variability of the water cycle in the upper Rio Grande basin at high resolution. J. Hydrometeor., 8, 805–824. Lin, C. A., L. Wen, G. H. Lu, Z. Y. Wu, J. Y. Zhang, Y. Yang, Y. F. Zhu, and L. Y. Tong, 2006: Atmospheric-hydrological modeling of severe precipitation and floods in the Huaihe River Basin, China. J. Hydrol., 330, 249–259. Mahfouf, J.-F., 1991: Analysis of soil-moisture from near-surface parameters—A feasibility study. J. Appl. Meteor., 30, 1534–1547. ——, E. Richard, and P. Mascart, 1987: The influence of soil and vegetation on the development of mesoscale circulations. J. Climate Appl. Meteor., 26, 1483–1495. Marani, M., E. Eltahir, and A. Rinaldo, 2001: Geomorphic controls on regional base flow. Water Resour. Res., 37, 2619–2630. Martinez-Fernandez, J., and A. Ceballos, 2003: Temporal stability of soil moisture in a large-field experiment in Spain. Soil Sci. Amer. J., 67, 1647–1656. Maxwell, R. M., F. K. Chow, and S. J. Kollet, 2007: The groundwater-land-surface-atmosphere connection: Soil moisture effects on the atmospheric boundary layer in fully-coupled simulations. Adv. Water Resour., 30, 2447–2466. Merz, B., and A. Bardossy, 1998: Effects of spatial variability on the rainfall runoff process in a small loess catchment. J. Hydrol., 213, 304–317. Mesinger, F., and Coauthors, 2006: North American regional reanalysis. Bull. Amer. Meteor. Soc., 87, 343–360.

663

Mintz, Y., 1984: The sensitivity of numerically simulated climates to land-surface boundary conditions. The Global Climate, J. T. Houghton, Ed., Cambridge University Press, 79–105. Miyakoda, K., G. D. Hembree, and R. F. Strickler, 1979: Cumulative results of extended forecast experiments. Part II: Model performance for summer cases. Mon. Wea. Rev., 107, 395–420. Mlawer, E. J., S. J. Taubman, P. D. Brown, M. J. Iacono, and S. A. Clough, 1997: Radiative transfer for inhomogeneous atmospheres: RRTM, a validated correlated-k model for the longwave. J. Geophys. Res., 102, 16 663–16 682. Mo, K. C., J. E. Schemm, H. Kim, and W. R. Higgins, 2006: Influence of initial conditions on summer precipitation simulations over the United States and Mexico. J. Climate, 19, 3640–3658. Oglesby, R. J., S. Marshall, and D. J. Erickson III, J. O. Roads, and F. R. Robertson, 2002: Thresholds in atmosphere–soil moisture interactions: Results from climate model studies. J. Geophys. Res., 107, 4224, doi:10.1029/2001JD001045. Otkin, J. A., and T. J. Greenwald, 2008: Comparison of WRF model-simulated and MODIS-derived cloud data. Mon. Wea. Rev., 136, 1957–1970. Pal, J. S., and E. A. B. Eltahir, 2001: Pathways relating soil moisture conditions to future summer rainfall within a model of the land–atmosphere system. J. Climate, 14, 1227–1242. Ratnam, J. V., and K. K. Kumar, 2005: Sensitivity of the simulated monsoons of 1987 and 1988 to convective parameterization schemes in MM5. J. Climate, 18, 2724–2743. Salvucci, G. D., and D. Entekhabi, 1994: Equivalent steady state soil moisture profile and the time compression approximation in water balance modeling. Water Resour. Res., 30, 2737–2749. Scha¨r, C., D. Lu¨thi, U. Beyerle, and E. Heise, 1999: The soil– precipitation feedback: A process study with a regional climate model. J. Climate, 12, 722–741. Schubert, S. D., M. J. Suarez, P. J. Pegion, R. D. Koster, and J. T. Bacmeister, 2004: On the cause of the 1930s Dust Bowl. Science, 303, 1855–1859. Seuffert, G., P. Gross, C. Simmer, and E. F. Wood, 2002: The influence of hydrologic modeling on the predicted local weather: Two-way coupling of a mesoscale weather prediction model and a land surface hydrologic model. J. Hydrometeor., 3, 505–523. Sivapalan, M. K., K. J. Beven, and E. F. Wood, 1987: On hydrologic similarity. 2. A scaled model of storm runoff production. Water Resour. Res., 23, 2266–2278. 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-4681STR, 88 pp. Small, E. E., 2001: The influence of soil moisture anomalies on variability of the North American monsoon system. Geophys. Res. Lett., 28, 139–142. Taylor, C. M., D. J. Parker, and P. P. Harris, 2007: An observational case study of mesoscale atmospheric circulations induced by soil moisture. Geophys. Res. Lett., 34, L15801, doi:10.1029/2007GL030572. Thompson, G., R. M. Rasmussen, and K. Manning, 2004: Explicit forecasts of winter precipitation using an improved bulk microphysics scheme. Part I: Description and sensitivity analysis. Mon. Wea. Rev., 132, 519–542. Troch, P. A., F. P. de Troch, and W. Brutsaert, 1993: Effective water table depth to describe initial conditions prior to storm rainfall in humid regions. Water Resour. Res., 29, 427–434. VanderKwaak, J. E., and K. Loague, 2001: Hydrologic-response simulations for the R-5 catchment with a comprehensive physics-based model. Water Resour. Res., 37, 999–1013.

664


Vivoni, E. R., V. Y. Ivanov, R. L. Bras, and D. Entekhabi, 2004: Generation of triangulated irregular networks based on hydrological similarity. J. Hydrol. Eng., 9, 288–302. ——, ——, ——, and ——, 2005: On the effects of triangulated terrain resolution on distributed hydrologic model response. Hydrol. Processes, 19, 2101–2122. ——, R. S. Bowman, R. L. Wyckoff, R. T. Jakubowski, and K. E. Richards, 2006a: Analysis of a monsoon flood event in an ephemeral tributary and its downstream hydrologic effects. Water Resour. Res., 42, W03404, doi:10.1029/2005WR004036. ——, D. Entekhabi, R. L. Bras, V. Y. Ivanov, M. P. Van Horne, C. Grassotti, and R. N. Hoffman, 2006b: Extending the predictability of hydrometeorological flood events using radar rainfall nowcasting. J. Hydrometeor., 7, 660–677. ——, ——, ——, and ——, 2007a: Controls on runoff generation and scale-dependence in a distributed hydrologic model. Hydrol. Earth Syst. Sci., 11, 1683–1701. ——, ——, and R. N. Hoffman, 2007b: Error propagation of radar rainfall nowcasting fields through a fully distributed flood forecasting model. J. Appl. Meteor. Climatol., 46, 932–940. ——, and Coauthors, 2007c: Variation of hydrometeorological conditions along a topographic transect in northwestern Mexico during the North American monsoon. J. Climate, 20, 1792–1809. ——, M. Gebremichael, C. J. Watts, R. Bindlish, and T. J. Jackson, 2008a: Comparison of ground-based and remotely-sensed surface soil moisture estimates over complex terrain during SMEX04. Remote Sens. Environ., 112, 314–325. ——, H. A. Moreno, G. Mascaro, J. C. Rodriguez, C. J. Watts, J. Garatuza-Payan, and R. L. Scott, 2008b: Observed relation between evapotranspiration and soil moisture in the North American monsoon region. Geophys. Res. Lett., 35, L22403, doi:10.1029/2008GL036001. Wang, G. L., Y. Kim, and D. G. Wang, 2007: Quantifying the strength of soil moisture–precipitation coupling and its sen-

VOLUME 10

sitivity to changes in surface water budget. J. Hydrometeor., 8, 551–570. Weaver, C. P., 2004: Coupling between large-scale atmospheric processes and mesoscale land–atmosphere interactions in the U.S. Southern Great Plains during summer. Part I: Case studies. J. Hydrometeor., 5, 1223–1246. Western, A. W., S. L. Zhou, R. B. Grayson, T. A. McMahon, G. Bloschl, and D. J. Wilson, 2004: Spatial correlation of soil moisture in small catchments and its relationship to dominant spatial hydrological processes. J. Hydrol., 286, 113–134. Westrick, K. J., and C. F. Mass, 2001: An evaluation of a highresolution hydrometeorological modeling system for prediction of a cool-season flood event in a coastal mountainous watershed. J. Hydrometeor., 2, 161–180. Wyckoff, R. L., 2007: Sensitivity to arroyo development scenarios: Insights from a distributed hydrologic model. M.S. thesis, Dept. of Hydrology, New Mexico Institute of Mining and Technology, 193 pp. Xie, H. J., X. B. Zhou, E. R. Vivoni, J. M. H. Hendrickx, and E. E. Small, 2005: GIS-based NEXRAD Stage III precipitation database: Automated approaches for data processing and visualization. Comput. Geosci., 31, 65–76. Xu, J. J., and E. E. Small, 2002: Simulating summertime rainfall variability in the North American monsoon region: The influence of convection and radiation parameterizations. J. Geophys. Res., 107, 4727, doi:10.1029/2001JD002047. ——, W. J. Shuttleworth, X. Gao, S. Sorooshian, and E. E. Small, 2004: Soil moisture-precipitation feedback on the North American monsoon system in the MM5-OSU model. Quart. J. Roy. Meteor. Soc., 130, 2873–2890. Zheng, X. Y., and E. A. B. Eltahir, 1998: A soil moisture–rainfall feedback mechanism. 2. Numerical experiments. Water Resour. Res., 34, 777–785.