Journal of Geophysical Research: Atmospheres

Soil moisture-vegetation-precipitation feedback over North America: Its sensitivity to soil moisture climatology

Authors

  • Yeonjoo Kim,

    Corresponding author
    1. Department of Civil and Environmental Engineering, University of Connecticut, Storrs, Connecticut, USA
    2. Korea Environment Institute, Seoul, South Korea
      Corresponding author: Y. Kim, Korea Environment Institute, 290 Jinheungno, Eunpyeong-gu, Seoul 122-706, South Korea. (yjkim@kei.re.kr)
    Search for more papers by this author
  • Guiling Wang

    1. Department of Civil and Environmental Engineering, University of Connecticut, Storrs, Connecticut, USA
    Search for more papers by this author

Corresponding author: Y. Kim, Korea Environment Institute, 290 Jinheungno, Eunpyeong-gu, Seoul 122-706, South Korea. (yjkim@kei.re.kr)

Abstract

[1] Our previous studies examined how vegetation feedback at the seasonal time scale influenced the impact of soil moisture anomalies (SMAs) on subsequent summer precipitation with a modified version of the coupled Community Atmosphere Model-Community Land Model 3 that includes a predictive phenology scheme. Here we investigate the climatology sensitivity of soil moisture-vegetation-precipitation feedback using the same model as the baseline model (BASE) and its derivative with modifications to the model runoff parameterization as the experiment model (EXP), in which we eliminate the subsurface lateral drainage to reduce the known dry biases of BASE. With vegetation feedback ignored, precipitation is more sensitive to wet SMAs than dry SMAs in BASE; opposite to BASE, the wetter mean soil moisture in EXP leads to higher sensitivity of precipitation to dry SMAs than to wet SMAs. However, in both BASE and EXP, the impact of dry SMAs on subsequent precipitation persists longer than the impact of wet SMAs. With vegetation feedback included, EXP shows a positive feedback between vegetation and precipitation following both dry and wet SMAs in summer, while BASE shows a positive feedback following wet SMAs only, with no clear signal following dry SMAs due to dry soil biases. In BASE, the magnitude of precipitation changes due to vegetation feedback is comparable to that due to soil moisture feedback when more realistic SMAs are applied. In addition, a major difference is found in spring when the vegetation impact on subsequent precipitation is negative and significant in BASE, but not significant in EXP.

1. Introduction

[2] Seasonal predictability of precipitation largely depends on slowly varying states of land and ocean, and the impact of land surface is particularly significant for summer precipitation in the interior of large continent such as North America [Koster et al., 2004; Jiang et al., 2009]. Namias [1952]speculated that soil moisture could support month-to-month persistence in climatic anomalies over the United States. Since then, numerous studies have examined land-atmosphere interaction, particularly focusing on soil moisture-precipitation coupling using general circulation models (GCMs) [e.g.,Rind, 1982; Koster et al., 2010], regional climate models [e.g., Paegle et al., 1996; Patarčić and Branković, 2012], or observational data [e.g., Entin et al., 2000; Mei and Wang, 2011]. These studies led to the consensus that soil moisture could add to summer precipitation predictability over regions where the land memory is long and the soil moisture-precipitation coupling is strong. The slowly varying soil moisture “remembers” past and present precipitation anomalies, and as the resulting soil moisture anomalies feed back to influence precipitation, this may lead to the persistence of soil moisture and precipitation anomalies.

[3] Numerous studies have tackled the issue of how soil moisture anomalies impact subsequent climate conditions [e.g., Rind, 1982; Shukla and Minz, 1982; Oglesby and Erickson, 1989; Paegle et al., 1996; Bosilovich and Sun, 1999; Pal and Eltahir, 2001; Georgescu et al., 2003; Kim and Wang, 2007a]. Most of these studies agree upon a positive feedback between soil moisture and precipitation. Wetter (drier) than normal soil tends to promote (suppress) precipitation and through the soil moisture-precipitation feedback may lead to persistence of higher (lower) than normal precipitation (Figure 1). Possible pathways for the feedback include local moisture recycling [e.g., Bosilovich and Chern, 2006] as well as changes in moisture convergence from remote sources [e.g., Oglesby and Erickson, 1989]. Using the National Center for Atmospheric Research (NCAR) Community Climate Model (CCM1), Oglesby and Erickson [1989]imposed desert-like initial soil moisture condition over an extensive area of North America. They found that soil moisture reduction prolonged or amplified drought, and suggested that moisture advection from the Gulf of Mexico played an important role in determining the area where the soil moisture reduction persisted.Bosilovich and Chern [2006] analyzed the water vapor tracer in GCM simulations and suggested strong precipitation recycling over the Mississippi River Basin (MRB) during summer.

Figure 1.

Diagram for the impact of vegetation feedback on how precipitation responds to initial soil moisture anomalies. ΔP and ΔP′ indicate change in precipitation due to soil moisture feedback and vegetation feedback, respectively, and solid and dashed lines indicate positive and negative feedbacks, respectively.

[4] Less studied is how the subsequent vegetation response to persistent precipitation anomalies may influence precipitation. More (less) precipitation may result in an increase (decrease) in vegetation, which has two consequences (Figure 1). First, the increase (decrease) in vegetation favors more (less) precipitation through its impact on albedo, Bowen ratio and roughness [Charney, 1975], which enhances the persistence of wet (dry) soil moisture anomalies, leading to a positive feedback [Sud et al., 1988; Los et al., 2000; Buermann et al., 2001]. Second, the enhanced (suppressed) transpiration leads to faster/slower depletion of soil moisture, which may reduce the persistence of wet (dry) anomalies, leading to a negative feedback [e.g., Pielke et al., 1998; Lu et al., 2001; Wang et al., 2006; Kim and Wang, 2007b; Notaro et al., 2008; Liu et al., 2010].

[5] Charney [1975] first suggested a positive biogeophysical feedback mechanism to explain droughts in the Sahara; a decrease in vegetation results in an increase in albedo, leading to an increase in subsidence which suppresses precipitation. Sud et al. [1988]showed that a reduction in surface roughness due to vegetation cover changes produced an increase in the boundary layer wind and a decrease in surface stress, leading to a reduction in the boundary layer water vapor transport convergence and precipitation. Recent studies have showed both positive and negative feedbacks using coupled ecosystem-climate models.Lu et al. [2001]coupled the CENTURY ecosystem model with the Regional Atmospheric Modeling System (RAMS) and performed simulations using both RAMS stand alone and the coupled RAMS-CENTURY model over the United States. Based on spatial averages over the central United States, lower simulated leaf area index (LAI) in the coupled model (relative to that prescribed in RAMS) leads to more precipitation due to larger vegetation transmissivity, resulting in greater radiation at the land surface, and finally more convective precipitation in the coupled model (i.e., a negative feedback). On the other hand, at a grid cell where winter wheat is the dominant vegetation, lower LAI due to harvest in the coupled model (relative to LAI in RAMS) leads to less precipitation via surface warming caused by the decrease of evaporative cooling (i.e., a positive feedback). Investigating the Holocene North Africa climate with a fully coupled climate-global dynamic vegetation model Fast Ocean Atmosphere Model/Lund-Postdam-Jena (FOAM-LPJ) [Gallimore et al., 2005], Liu et al. [2010] demonstrated that an initial anomaly in vegetation can induce a direct, positive vegetation feedback, leading to an increase in subsequent precipitation and an indirect, negative feedback, leading to reductions in soil moisture and precipitation. An increase in vegetation tends to reduce ground evaporation, and the evaporation reduction may offset the increase of plant transpiration, favoring a negative feedback. Furthermore, grasses are more effective than trees in inducing the negative feedback, because of their limited transpiration capacity and shallower roots.

[6] Our previous studies [Kim and Wang, 2007a, hereafter KW07a; Kim and Wang, 2007b, hereafter KW07b] tackled the impact of initial soil moisture anomalies on subsequent precipitation over North America and how vegetation feedback influences such impact, using a modified version of the NCAR Community Atmosphere Model version 3 and the Community Land Model version 3 (CAM3–CLM3). Without vegetation feedbacks (KW07a), the response of precipitation to wet soil anomalies in summer is larger in magnitude than to dry soil anomalies; however, the resulting positive precipitation anomalies quickly dissipate within a month or so, while the resulting negative precipitation anomalies remain substantial for a longer period. Consistently, wet spring soil moisture anomalies dissipate before summer, thus having a smaller impact on summer precipitation than dry spring soil moisture anomalies. In KW07b, when the predictive phenology scheme of Kim and Wang [2005]is included in the model, wet soil moisture anomalies during summer lead to increases in LAI, favoring precipitation through increases in evapotranspiration (ET) as well as increases in surface heating, induced by increases in longwave radiation outcompeting decreases in the shortwave radiation. The response of vegetation to dry soil moisture anomalies in the summer months, however, is not significant, which is attributed to a dry bias in the coupled CAM3-CLM3. During the summer following wet spring soil moisture anomalies, the vegetation feedback on precipitation is negative; i.e., it tends to suppress the response of precipitation through the depletion of soil moisture by vegetation.

[7] Studies in the past several years have made it evident that soil moisture-precipitation coupling depends on model parameterizations and model climatology [Guo et al., 2006; Lawrence et al., 2007; Wang et al., 2007; Kumar et al., 2010]. Lawrence et al. [2007]modified CLM3 to improve its ET partitioning and reduce dry soil biases and investigated its impact on land-atmosphere interaction. They modified a number of vegetation and hydrology parameterizations: Changes in vegetation parameterizations include parameterizations for canopy integration, canopy interception, soil water availability and soil evaporation; changes in soil hydrology include eliminating lateral drainage of soil water, increasing infiltration of water into the soil and increasing the vertical redistribution of soil water. With the modified CLM3,Lawrence et al. [2007]showed diverse influences of improved ET portioning (i.e., increases in transpiration and decreases in soil and canopy evaporation, and thus reduced dry soil biases in the model) on land-atmosphere interaction, including an extended ET response to rain events and a shift in the precipitation distribution, as well as slightly stronger influences of subsurface soil moisture on precipitation.

[8] The soil moisture-vegetation-precipitation feedback processes investigated in our previous studies [KW07a and KW07b] may differ with model parameterization and model climatology. In particular, we hypothesize that dry bias of CAM3-CLM3 over the MRB [Bonan and Levis, 2006; Kim and Wang, 2007a] influences the feedback processes among soil moisture, vegetation and precipitation over North America, and thus we investigate how such feedback processes might change if CAM3-CLM3's dry bias were reduced. Among several components of CAM3-CLM3 that contribute to this dry bias, we chose to modify the subsurface runoff parameterization in CLM3 (seesection 2.2 for details). Runoff processes, including surface and subsurface runoffs, in marcoscale models such as CLM3 are crudely parameterized and is subject to major uncertainties [Niu et al., 2005; Choi and Liang, 2010]. Following Wang et al. [2005], Lawrence et al. [2007], and Sun and Wang [2008], we assume that all subsurface runoff comes from drainage at the bottom of the soil column and saturation excesses, and subsurface lateral runoff is eliminated, as a temporary, simple solution to reduce the known dry bias in soil moisture. Then, we perform a series of numerical experiments with the modified model and compare the results with those from KW07a and KW07b.

2. Model Description

[9] The model used in this study is the NCAR CAM3 [Collins et al., 2006] coupled with the CLM3 [Dai et al., 2003]. Oceanic boundary conditions in this coupled land-atmosphere model are prescribed with the climatological monthly varying sea surface temperature and sea ice coverage [Hurrell et al., 2008]. The concentration of atmospheric CO2 is assigned to be 355 ppm. Among the three dynamics schemes available in CAM (Eulerian spectral, semi-Lagrangian dynamics and Finite Volume [FV] dynamics), we choose the FV dynamical core [Lin, 2004] with a horizontal resolution of 1.9° latitude by 2.5° longitude and 26 vertical levels. The land model CLM3 has 10 unevenly spaced soil layers, up to 5 snow layers, and 1 vegetation layer. The land surface within each grid cell is represented by the fractional coverage of four types of patches (glacier, lake, wetland, and vegetated), and the vegetation portion of the grid cell is represented by the fractional coverage of up to 4 out of 16 different plant functional types (PFTs) available in the model (Figure 2a). There are seven primary PFTs: needleleaf evergreen and deciduous trees, broadleaf evergreen and deciduous trees shrubs, grasses and crops. These are further categorized into 16 physiological variants on the basis of climate rules (arctic, boreal, temperate and tropical) and supplemented by discrimination of C3 and C4 grasses and crops. While the original version of CLM3 supports only C3 crop, the modified version of CLM3 used in this study (described below) discriminates C3 and C4 crops (i.e., wheat and corn).

Figure 2.

(a) Percentage of the grid cells occupied by and (b) maximum LAI of needleleaf trees, broadleaf trees, shrubs, grasses, and crops on a 0.5° × 0.5° resolution.

2.1. Leaf Phenology Scheme

[10] In the original version of CLM3, leaf phenology is prescribed and the seasonal course of leaf area index (LAI) for each PFT is derived through interpolating the climatological monthly PFT-specific LAI. In this study, the default leaf phenology scheme in CLM3 is replaced with a predictive scheme that has been validated against the latest Moderate Resolution Imaging Spectroradiometer (MODIS) observational data over North America [Kim and Wang, 2005]. While the default leaf phenology prescribes the climatological (i.e., seasonally varying, but not interannually varying) LAI, the predictive phenology scheme simulates the response of vegetation at the seasonal time scale to meteorological and environmental conditions, providing an opportunity to investigate seasonal vegetation-climate interactions.

[11] In the predictive phenology scheme, the PFT-specific leaf area index (LAIdaily) is updated daily by scaling down the PFT-specific annual maximum leaf area index (LAImax) with a predictive phenology factor (D):

display math

where LAImax is derived from MODIS LAI at 0.5° × 0.5° resolution [Tian et al., 2004] (Figure 2b). The phenology factor (D), ranging from zero to one, is simulated for cold-deciduous plants and drought-deciduous plants separately. For plants responding to both coldness and drought (e.g., grasses and crops), the phenology factor is determined based on the multiplicative effect of cold and drought stresses. For evergreen trees, their LAI seasonality is prescribed based on the MODIS LAI observations.

[12] The cold deciduous phenology scheme considers the impact of air temperature, soil temperature and photoperiod (i.e., the length of daytime) in predicting leaf green-up, development and senescence. The 10 day average air temperature (T10), accumulated growing degree days (AGDD) from 1 January and the coldest monthly temperature based on climatology (Tc) are used, and the base temperatures for AGDD are 0°C for trees and −5°C for grass. Leaf onset takes place once T10 exceeds its threshold (0°C or (Tc + 5) °C, whichever is larger, for trees; 0 for grasses), and AGDD exceeds its threshold of 100 for trees and 150 for grasses. Leaf senescence occur once T10 drops below its threshold, 0°C or (Tc + 5) °C (whichever is larger) for trees and 0 for grasses. Additionally, for the broadleaf deciduous trees over the North America, leaf senescence is initiated once soil temperature or photoperiod drops below its thresholds. After the criteria for leaf green up or senescence are met, the full leaf display at the beginning of the growing season or complete leaf offset at the end of the growing season is assumed to take 15 days for simplicity.

[13] The drought deciduousness is predicted based on soil water availability, which regulates plant growth in dry regions. The whole-plant water stress factor (W) is calculated by the following equation:

display math

where froot,j is the fraction of the root biomass within soil layer j and ψ is the soil potential. Therefore it depends on soil water potential in different soil layers and the plant rooting profile, and ranges from zero at the permanent wilting point to one at saturation. W is scaled with its threshold of 0.4 to determine the phenology factor, D.

[14] For crops, their climatological planting and harvesting times are prescribed, but their LAI between times of planting and harvesting is predicted in response to hydrometeorological conditions in the same way as done for grasses. The planting and harvesting dates are extracted from MODIS Normalized Difference Vegetation Index (NDVI) at the 1.0° resolution. Further details about the phenology scheme can be found in the study by Kim and Wang [2005].

2.2. Runoff Scheme

[15] The runoff scheme of CLM3 was modified to investigate the impact of model parameterization and model climatology/biases on the feedback processes involving soil moisture, vegetation and precipitation. As mentioned in section 1, we hypothesize that the dry bias of CAM3-CLM3 over the MRB [Bonan and Levis, 2006; Kim and Wang, 2007a] could influence the feedback processes among soil moisture, vegetation and precipitation over North America, and thus we aim to investigate how such feedback processes might differ if CAM3-CLM3's dry bias were reduced. In particular, some of the asymmetric responses of precipitation to wet and dry soil moisture anomalies may be attributed to dry biases of soil moisture in the model. Specifically, the response of precipitation to wet soil anomalies in summer is larger in magnitude, and also persists longer than the response to dry soil anomalies. Also with vegetation feedback included, wet soil moisture anomalies during summer lead to increases in precipitation via increases in LAI, but the response of vegetation to dry soil moisture anomalies in the summer months is not significant, which may be attributed to the dry bias in the coupled CAM3-CLM3.

[16] Among many potential sources of dry bias in soil moisture, we chose to focus on the runoff parameterization. Runoff directly influences soil moisture, and is therefore expected to influence the soil moisture-vegetation-precipitation feedback processes. However, the runoff parameterization in macroscale models, such as CLM3, is subject to major uncertainties [Niu et al., 2005; Choi and Liang, 2010]. Runoff processes are conceptually difficult to represent in a climate model because they vary considerably at the local scale. Also, most model validations are performed with field measurements at small-scale catchment, often smaller than the spatial scale of interest in climate models.

[17] In CLM3, the runoff parameterization is based on the TOPMODEL [Beven and Kirkby, 1979] and Biosphere-Atmosphere Transfer Scheme (BATS) [Dickinson et al., 1993]. The runoff parameterization involves the determination of saturated fractions (fsat) for the grid cell. Precipitation falling over the saturated fraction is immediately converted to surface runoff. Therefore the saturated fraction, which depends on topography and water table depth of groundwater, is a dominant control on surface runoff. The concept of topographic index (λ), also referred to as the wetness index, is introduced in TOPMODEL to represent topographic variations. The topographic index is

display math

where a is the upstream area above a pixel that drains through the unit contour at the pixel and tanβ is the local surface topographic slope. The saturated fraction is

display math

where wfact is a parameter determined by the distribution of the topographic index and set to be 0.3. zw is a dimensionless representation of the mean water table depth, calculated using the sum of soil wetness over the 10 soil layers:

display math

where fz is a water table depth scale parameter, which set to be 1 m−1, zh,10 is the bottom depth of the lowest soil layer, si is the soil wetness for layer i, and Δzi is the soil layer thickness.

[18] Surface runoff includes the runoff directly from the saturated fraction and the infiltration excess surface runoff over unsaturated areas. Subsurface runoff is also calculated for both saturated and unsaturated areas. Subsurface runoff includes 1) lateral drainage from the sixth to ninth soil layer for saturated and unsaturated areas (which is turned off in our experiment model for sensitivity studies as discussed later in this section), 2) drainage out of the bottom of the soil layer, and 3) adjustments needed to keep liquid water content of each layer between maximum (i.e., saturation) and minimum (i.e., zero liquid water content) values. Lateral drainage from the saturated fraction depends on the soil wetness and soil and topographic features, while lateral drainage from the unsaturated fraction depends on the hydraulic conductivity-weighted average of wetness in the sixth to ninth soil layers, hydraulic conductivity, and soil texture.Dai et al. [2003] provide further details on the subsurface drainage scheme in CLM3.

[19] The runoff parameterization in CLM3 is not well constrained with observations [Choi and Liang, 2010]. While the above mentioned tuning parameters vary with topographic variation and soil texture, those are set to be constant based on tuning for an earlier version of CLM. The saturated hydraulic conductivity for the bottom layer (0.04 mm/s) exceeds any likely precipitation rate [Niu et al., 2005]. Furthermore, the parameterization of subsurface lateral drainage is rather arbitrary and uncertain; it is defined as a lateral drainage out from the sixth to ninth soil layers with no scientific basis and oversaturated water in a soil layer is being added to subsurface drainage instead of adding to the overlying soil layer. Such drainage is very efficient in the model, resulting in a rapid flushing of soil water out of the column, which is directed into runoff (not to an adjoining cell) and effectively lost from the soil [Lawrence et al., 2007]. These represent a large source of uncertainty in the model simulation of hydrologic processes, and lead to dry biases. Following Wang et al. [2005], Lawrence et al. [2007], and Sun and Wang [2008], we choose to turn off the subsurface lateral drainage from the sixth to ninth soil layers in the model and only allow subsurface runoff in the form of drainage from the bottom of the soil column and saturation excesses. Although this treatment of subsurface flow is a rather arbitrary solution to reduce the dry soil bias of the model, it does result in reasonable mean soil moisture (see section 2.3) and a contrast with the baseline model used in KW07a and KW07b. It therefore provides a tool for examining the impact of the soil moisture climatology on soil moisture-vegetation-precipitation feedback processes. Note that by the elimination of subsurface lateral drainage, our target is to reduce dry soil biases, rather than to improve runoff prediction. The latter requires more comprehensive, systematic modifications of the hydrological parameterizations at the land surface (e.g., canopy parameterizations for how much water reaches the soil, runoff parameterizations for how much water moves out of the soil column, and water table parameterization for how soil water interacts with groundwater).

2.3. Model Climatology

[20] We perform the 20 year integrations with the interannually varying SST from 1979 to 1998 [Hurrell et al., 2008] at a 1.9° × 2.5° resolution, using two version of the CAM3-CLM3 model, one with the default subsurface runoff scheme (i.e., with the baseline model, referred as to BASE hereafter) and one with the modified subsurface runoff scheme in which no lateral drainage is allowed (i.e., the experiment model, referred as to EXP hereafter). To evaluate the model simulation of key hydrologic variables closely related to soil moisture-precipitation-vegetation feedback processes, we use the North American Regional Reanalysis (NARR) data [Mesinger et al., 2006] as a surrogate for observations, which provides a spatially continuous regional reanalysis data over the North America since 1979. This analysis assimilates observations, including precipitation from rain gauging stations, radiance data from satellite observations, near surface wind and moisture from Global Reanalysis outputs, sea and lake surface temperature, and sea and lake ice cover data. The data are available at a 32 km resolution, and have been regridded to the 1.9° × 2.5°CAM3-CLM3 resolution in this study. The volumetric soil moisture vales in both NARR and CAM3-CLM are averaged throughout the soil depth of 2 m, as the NARR data are available up to 2 m and CAM3-CLM3 data are available up to 3.43 m.

[21] BASE shows dry biases compared to NARR in all fields, including soil moisture, ET and precipitation (Figure 3). These dry biases are reduced in general in EXP relative to BASE, although the spatial patterns in EXP are still similar to those in BASE. We find statistically significant increases in soil moisture in EXP relative to BASE over most of North America. Such increases in soil moisture result in increases in ET, and thus precipitation. The statistically significant increases in ET and precipitation are found over the continental interior only, where soil moisture-precipitation coupling is strong [Koster et al., 2004; Wang et al., 2007] and thus land surface conditions play a key role in the summer climate.

Figure 3.

Averages during JJA of (a) soil moisture, (b) evapotranspiration, and (c) precipitation in NARR (first row), EXP (second row), and BASE (third row), as well as their differences between EXP and BASE (fourth row). Simulations are driven by the interannually varying SST from 1979 to 1998. Only differences exceeding the 90% confidence level are shaded. The numbers in the bottom left of each panel indicate averages over the MRB.

[22] Note that ET in NARR is a simulated variable without data assimilation, and shows significant biases as compared to Fluxnet data [Kumar and Merwade, 2011]. For soil moisture, Vivoni et al. [2008] reported overestimation of soil moisture in NARR compared to observations in Arizona, USA and Sonora, Mexico, while Miguez-Macho et al. [2008] reported underestimation in NARR compared to observations in Oklahoma, Illinois, and Iowa, USA. However, both these studies are subject to uncertainties related to the scale mismatch of soil moisture between NARR (with a 32 km resolution) and observations (which are at point scale).

3. Experimental Design

[23] Using each version of the model (i.e., BASE and EXP), we perform an initial integration and a large number of ensemble simulations that differ in initial soil moisture conditions and vegetation treatment to investigate the impact of mean soil moisture on the sign and strength of the soil moisture-vegetation-precipitation feedback. Driven with climatological SST, the initial integration is carried out for 12 years. Data from the first 2 years are discarded as model spin up, and the last 10 years of data, although it may be short, are used to derive the model climatology of soil moisture on the first day of each month. For example, soil moisture values on 1 June for 10 years are averaged in each soil layer. This soil moisture climatology is used to initialize subsequent ensemble simulations that run from the first day of a given month until the end of the year.

[24] Three different types of ensembles are designed for BASE and EXP, respectively: the Control, SM Anomaly and SM_Veg Anomaly (Table 1). The ensembles with the BASE model are a subset of ensembles carried out in KW07b, and the ensembles with the EXP model are carried out in this study. The Control ensemble is initialized with climatological soil moisture, and the SM Anomaly and SM_Veg Anomaly ensembles are initialized with specified soil moisture anomalies imposed to the soil moisture climatology. Vegetation seasonality in the Control and SM_Veg Anomaly ensembles is predicted by the prognostic phenology scheme, and is prescribed in each SM Anomaly simulation using model output from the corresponding Control simulation. Therefore, climate differences between the SM Anomaly ensemble and the Control ensemble are attributed to the impact of soil moisture initialization through soil moisture-precipitation interactions; climate differences between the SM_Veg Anomaly ensemble and the Control ensemble are attributed to the impact of soil moisture initialization and vegetation feedbacks; and climate differences between the SM_Veg Anomaly ensemble and the SM Anomaly ensemble represent the impact of vegetation feedback alone. Furthermore, climate differences between EXP and BASE represent the impact of soil moisture climatology.

Table 1. List of Ensemble Simulationsa
EnsembleStart DateDirection of AnomalyMagnitude of AnomalybInitial Soil Moistureb
  • a

    Anomaly ensembles include SM Anomaly and SM_Veg Anomaly ensembles. Ensemble simulations are performed with both EXP and BASE models, and each case includes five ensemble members.

  • b

    Presented as percentages of climatological soil moisture.

Control1 AprNA0100
 1 JunNA0100
Anomaly1 Junwet+80180
 1 Junwet+30130
 1 Jundry−8020
 1 Jundry−3070
 1 Aprwet+80180
 1 Aprdry−8020

[25] Each ensemble includes five members, which differ from one another only in the initial soil moisture condition throughout the whole soil depth in the model (∼3.4 m). Specifically, for an ensemble without initial soil moisture anomalies, its five members are initialized with 100%, 99%, 98%, 97% and 96% of the climatological soil moisture; for an ensemble with 80% dry (or wet) anomalies, its five members are initialized with 20%, 19%, 18%, 17% and 16% (or 180%, 179%, 178%, 177% and 176%) of the climatological soil moisture, which represent a decrease (or increase) of initial soil moisture by 80% from the climatological value (Table 1).

[26] Note that in the baseline CAM3–CLM3 over much of North America, it takes more than 80% increase of soil moisture from its climatological value to reach field capacity; about 20% decrease is needed to reach the wilting point [KW07a, Figure 4]. Further, the Illinois State Water Survey observational data [Hollinger and Isard, 1994] shows that soil moisture ranges from about 90% below to about 50% above its mean value at the most variable station, and ranges from about 40% below to about 30% above at the least variable station. This indicates that the 80% increase or decrease of soil moisture from its climatology may be beyond the range of natural variability in some places, although the observed soil moisture is not directly comparable to the model soil moisture due to their discrepancies in the spatial and temporal resolutions [KW07a, section 4b]. Also note that soil moisture anomalies are applied throughout the whole soil depth in the model (∼3.4 m), and the depth of soil moisture anomalies may influence soil moisture-vegetation-precipitation feedback processes. Given that different types of vegetation have different rooting depth therefore respond selectively to soil moisture anomalies at different depth, theoretically the magnitude of the impact of vegetation feedback may vary with the depth of initial soil moisture or with the dominant vegetation type. However, over the MRB, which is the part of our model domain where precipitation and vegetation are most responsive to initial soil moisture anomalies [KW07a and KW07b], the land cover is dominated by grass and crops. Their root system is fairly shallow, mostly residing in the top 0.5 m of the soil, and reducing the depth of soil moisture anomalies to the top seven layers in the model (about 0.83 m) does not significantly influence the precipitation response [KW07a].

[27] Initial soil moisture anomalies are applied across much of North America (inner box in Figure 4), and these soil moisture anomalies are applied on 1 June and 1 April to investigate the impact of summer and spring soil moisture anomalies, respectively. Our results analysis will focus on the MRB (shaded in Figure 4). Over this region, precipitation is highly sensitive to initial soil moisture anomalies because the region falls into the transition zone between dry and wet hydrologic regimes, where ET is highly sensitive to soil moisture and shows large temporal variability [Koster et al., 2004; Guo et al., 2006; Kim and Wang, 2007a]. Furthermore, vegetation-soil moisture-precipitation coupling is strong in this region [Kim and Wang, 2007b], where vegetation is dominated by grass and crops, and water stress is a major limiting factor for their growth.

Figure 4.

Map of the study domain. An inner box with dashed lines represents North America (NA), the domain of initial soil moisture anomalies in the numerical experiments. The shaded area defines the MRB on a 1.9° × 2.5° resolution.

[28] In addition to the 12 year initial simulations with climatological SST used to derive the soil moisture climatology, we have carried out a 20 year simulation driven with interannually varying SST from 1979 to 1998. This simulation is used to get a more realistic estimate of the interannual variability of the climate, in order to examine the statistical significance of simulated climate differences between two different types of ensembles (e.g., difference between SM_Veg and SM ensembles). Note that the simulation with climatological SST underestimates the interannual variability of climate over land, which if used would have caused spuriously high statistical significance. Based on this 20 year integration, we estimated the tstatistics for each grid cell and region. For each grid cell, monthly output from this 20 year simulation is used to derive the 90% confidence interval in the significance tests of monthly results over the 2-D spatial domain. Daily output is used to derive the 90% confidence interval in the significance tests of the daily results and the corresponding 10 day running averages.

4. Result Analysis

4.1. Soil Moisture-Precipitation Feedback

[29] A positive feedback between soil moisture and precipitation has been shown in KW07a using the BASE model. Positive (negative) soil moisture anomalies applied at the beginning of June over North America lead to an increase (decrease) in precipitation in the subsequent 2–3 months over a major portion of North America. Using the EXP model in this study, we also find a positive feedback between soil moisture and precipitation over the continental interior of North America, consistent with the results using the BASE model (Figure 5). Dry soil moisture anomalies lead to significant decreases in subsequent precipitation over the upper MRB (Figure 5b), which is a wet region (Figure 5a); wet soil moisture anomalies lead to significant increase of subsequent precipitation over the lower MRB and the southwestern North America (Figure 5c), which is a dry region. As detailed in KW07a, dry (wet) soil moisture anomalies cause increases (decreases) in surface temperature and consequently decreases (increases) in surface pressure. Such local changes, in turn, modify the large-scale circulations, leading to weaker (stronger) westerlies and northward (southward) shift of the high precipitation belt in the dry (wet) cases. Therefore, the belt of high precipitation moves northward (southward) (Figures 5b and 5c). The precipitation responses mostly disappear by September in the ensemble with wet soil moisture anomalies, while they persist until September in the ensemble with dry soil moisture anomalies.

Figure 5.

(a) Monthly precipitation in the Control ensembles, (b) precipitation anomalies (SM Anomaly – Control) in dry SM Anomaly ensembles relative to the Control ensemble, and (c) precipitation anomalies (SM Anomaly – Control) in wet SM Anomaly ensembles relative to the Control ensemble in June, July, August, and September with the EXP model. The dry/wet SM Anomaly ensembles are initialized with an 80% decrease/increase in climatological soil moisture on 1 June, and the averages of five ensemble members are presented. Only differences exceeding the 90% confidence level are shaded. The numbers in the bottom left of each panel indicate averages over the MRB.

[30] In June, the magnitude of the actual soil moisture anomalies is larger in the wet SM Anomaly ensembles than in the dry ensembles in both EXP and BASE (Figure 6). The 80% decrease from the climatological soil moisture is more than what it takes for soil moisture to reach the wilting point, and the magnitude of the actual soil moisture anomalies is therefore smaller than that of the 80% dry anomaly. However, the precipitation response to dry initial soil moisture anomalies is larger in magnitude than the response to wet initial soil moisture anomalies in EXP, which is opposite to the results in BASE (Figure 7). In EXP (BASE), changes in precipitation, averaged in July over the MRB, are 1.2 (2.1) mm/d and −2.2 (−1.5) mm/d for the 80% wet and dry soil moisture anomalies, respectively. As mentioned in KW07a, a wetter regime/region tends to be more sensitive to dry anomalies and a drier regime/region tends to be more sensitive to wet anomalies. Therefore, compared to BASE, the wetter soil moisture climatology in EXP (see Figure 3) leads to higher sensitivity to dry soil moisture anomalies during June.

Figure 6.

Ten day running averages of soil moisture anomalies (soil saturation level in percentage) (SM Anomaly – Control) as a response to an 80% decrease (solid lines) or an 80% increase (dashed lines) in climatological soil moisture applied on 1 June in the dry/wet SM Anomaly ensembles relative to the Control ensembles with (a) the EXP model and (b) the BASE model, averaged over the MRB in the first, third, and fifth soil layers. The dry/wet SM Anomaly ensembles are initialized with an 80% decrease (solid lines) and increase (dashed lines) in climatological soil moisture on 1 June. The shaded area presents the 90% confidence interval for the 10 day average of soil moisture in each layer. Five ensemble members are presented in gray lines, and their averages are in black lines.

Figure 7.

Ten day running averages of precipitation anomalies (SM Anomaly – Control) as a response to an 80% increase (top row) or an 80% decrease (bottom row) in climatological soil moisture applied on 1 June in the dry/wet SM Anomaly ensembles relative to the Control ensembles with (a) the EXP model and (b) the BASE model, averaged over the MRB. The shaded area presents the 90% confidence interval for the 10 day average of precipitation. Five ensemble members are presented in gray lines, and their averages are in black lines.

[31] The impact of dry anomalies tends to persist longer than the impact of wet anomalies in both BASE and EXP (Figure 6). For example, in the topsoil layer, the ensemble average of 10 day running mean of soil moisture anomalies mostly dissipate after 5 and 3 months (i.e., by October and August) in EXP and BASE, respectively. Although the mean soil moisture in the model differs between BASE and EXP, the soil over the MRB in both models dries up from spring to summer, and thus the dry soil moisture and precipitation anomalies persist longer than the wet soil and precipitation anomalies. Furthermore, initial soil moisture anomalies on 1 June dissipate much slower in EXP than in BASE, which is particularly obvious in the wet SM Anomaly ensemble, since lateral subsurface drainage is turned off in EXP (Figure 6). Such longer persistence of soil moisture anomalies in EXP than in BASE leads to longer persistence of positive precipitation anomalies (Figure 7).

[32] As pointed out earlier, the 80% soil moisture anomalies are to ensure clear signals of climate responses to soil moisture anomalies, but may be beyond the range of natural variability. In GW07a (i.e., BASE), the comparison between the 80% anomaly and the 30% anomaly cases suggests that as the magnitude of soil moisture anomalies decreases, the resulting precipitation anomalies decrease as well. The precipitation differences averaged over the MRB for July and August decrease from 1.0 (−0.6) mm/d for the 80% wet (dry) soil moisture anomalies to 0.8 (−0.3) mm/d for the 30% wet (dry) soil moisture anomalies in BASE (Figure 8b). In EXP (Figures 5, 8a, and 9), decreases in precipitation due to reductions in the imposed soil moisture anomalies from 80% to 30% of climatological soil moisture are also found, but the magnitude of precipitation differences is more substantial than in BASE. The precipitation differences decrease from 1.8 (−1.2) mm/d for the 80% wet (dry) soil moisture anomalies to 0.4 (−0.3) mm/d for the 30% wet (dry) soil moisture anomalies. This model difference is mainly attributed to the different representation of subsurface runoff between EXP and BASE. Specifically, in BASE, part of the wet initial soil moisture anomalies always run off through the subsurface lateral flow in the soil column, but in EXP, the wet soil moisture anomalies persist in the soil column until they drain out of the bottom of soil column. Furthermore, we find both the persistence and the magnitude of precipitation anomalies decrease as the magnitude of soil moisture anomalies decreases.

Figure 8.

Changes in precipitation due to soil moisture feedback (white; SM – Control), vegetation feedback (light gray; SM_Veg – SM), and both soil moisture and vegetation feedback (dark gray; SM_Veg – Control), averaged over the MRB throughout the second and third months (as indicated inside parentheses) following the 80% and 30% wet (top row) and dry (bottom row) soil moisture anomalies applied on 1 June in (a) the EXP model and (b) the BASE model. The error bar presents a standard deviation among five ensemble members.

Figure 9.

Precipitation anomalies (a) in SM Anomaly ensembles relative to Control ensembles (SM Anomaly – Control) and (b) in SM_Veg Anomaly ensembles relative to SM Anomaly precipitation anomalies (SM_Veg Anomaly – SM Anomaly) in June, July, August, and September with the EXP model. The wet SM Anomaly ensembles are initialized with a 30% increase in climatological soil moisture on 1 June, and the averages of five ensemble members are presented. Only differences exceeding the 90% confidence level are shaded. The numbers in the bottom left of each panel indicate averages over the MRB.

[33] For spring, the response of precipitation to soil moisture changes does not become substantial until summer (Figure 10). In the wet EXP ensemble, larger increases in precipitation are found than in BASE, resulting from relatively slow drainage of soil moisture in EXP. The increases in precipitation averaged over the MRB in April are 0.8 mm/d and 0.6 mm/d in EXP and BASE, respectively. This is consistent with results from the ensembles with summertime anomalies. In addition to major responses in precipitation during summer, we observed changes in precipitation in April following the wet soil moisture anomalies. As detailed in KW07a, relatively large decreases in temperature in the southern part of the continent results in the convergence of moisture flow in the northern part and divergence in the southern part, leading to increases in large-scale precipitation in the northern part of the continent in April and decreases in convective precipitation in the northeastern part (not shown here).

Figure 10.

Ten day running averages of precipitation anomalies (SM Anomaly – Control) as a response to an 80% increase (top) or an 80% decrease (bottom) in climatological soil moisture applied on 1 April in the dry/wet SM Anomaly ensembles relative to the Control ensembles with (a) the EXP model and (b) the BASE model, averaged over the MRB. The shaded area presents the 90% confidence interval for the 10 day average of precipitation. Five ensemble members are presented in gray lines, and their averages are in black lines.

4.2. Soil Moisture-Vegetation-Precipitation Feedback

[34] Here we examine the impact of vegetation feedback on the response of precipitation to soil moisture anomalies. Using the BASE model, KW07b shows a positive vegetation feedback under wet summer soil moisture anomalies. Wet soil moisture anomalies lead to increases in LAI, which induce higher ET and ultimately higher precipitation (positive feedback). At the same time, the increase in LAI also causes greater water consumption by vegetation through transpiration, leading to faster soil moisture depletion (negative feedback). Whether soil becomes wetter or drier when the impact of vegetation is included (i.e., in the SM_Veg Anomaly relative to the SM Anomaly) depends on the competition between these two mechanisms. KW07bfound that in BASE the impact of increased vegetation on precipitation dominates over the impact of vegetation-induced soil drying, leading to a positive vegetation feedback following wet summer soil moisture anomalies.

[35] Similar to in the KW07b study, in EXP, we also find increases in LAI and subsequent decreases in soil water in response to the 30% (Figure 11) and 80% (not shown) wet soil moisture anomalies on 1 June in the SM_Veg Anomaly relative to the SM Anomaly. Here, LAI increases are mainly concentrated in the MRB, where the dominant PFTs over the land surface are grasses and crops, the types of vegetation that can take advantage of wet soil moisture anomalies since their growth is limited by both water and cold stresses (see Figure 2). Soil water decreases relatively slowly after July, as it takes time for increased vegetation to further deplete soil water. Surprisingly, the precipitation differences are only clear with the 30% wet soil moisture anomalies, but not with the 80% (Figures 8a and 9). In EXP, the precipitation differences between SM_Veg and SM Anomaly averaged over the MRB for July and August are 0.6 mm/d and 0.0 mm/d with the 30% and 80% wet soil moisture anomalies, respectively. The increases in precipitation attributable to vegetation change (i.e., precipitation difference between the SM_Veg and SM Anomaly) decrease with the increasing magnitude of soil moisture anomalies in EXP. In BASE [KW07b], the increases in precipitation are enhanced as the magnitude of soil moisture anomalies increase; the precipitation differences are 0.3 mm/d and 0.9 mm/d with the 30% and 80% wet soil moisture anomalies, respectively (Figure 8b). One possible explanation for this unexpected results in EXP is that the overwhelmingly large amount of precipitation increases during July and August due to the initial soil moisture anomalies in the 80% wet anomaly case leaves little or no room for vegetation to further increase precipitation. Furthermore, in EXP with the 30% wet soil moisture anomalies (Figure 9), increases in precipitation due to soil moisture feedback (SM Anomaly – Control) are found mostly in June and July, and increases in precipitation due to vegetation feedback (SM_Veg Anomaly – SM Anomaly) are found later in August due to the delayed response of vegetation (as explained earlier), and their magnitudes due to vegetation feedback are comparable to those due to soil moisture feedback; the differences in precipitation averaged over the MRB in SM (SM_Veg) Anomaly relative to Control are 0.9, 0.8 and 0.0 mm/d (−0.2 0.3 and 1.0 mm/d) in June, July and August, respectively.

Figure 11.

Differences (SM_Veg Anomaly – SM Anomaly) in (a) LAI and (b) soil water between the SM_Veg Anomaly ensemble and the SM Anomaly ensemble in June, July, August, and September with the EXP model. The anomaly ensembles are initialized with a 30% increase in climatological soil moisture on 1 June, and the averages of five ensemble members are presented. Only differences exceeding the 90% confidence level are shaded.

[36] In BASE [KW07b], dry soil moisture biases cause underestimation of the LAI response to dry summer soil moisture anomalies, leading to little vegetation-induced changes in subsequent soil moisture and precipitation. In EXP (not shown), however, with the dry soil moisture biases reduced, decreases in LAI in response to the dry soil moisture anomalies persist for at least two months, and their magnitude are comparable to that of increases in the wet ensembles. In response to the 30% dry soil moisture anomalies, the decreases in LAI induce no changes in soil moisture (not shown) and decreases in precipitation (Figure 8a, second row) in SM_Veg Anomaly relative to SM Anomaly. This indicates that a positive vegetation feedback on the subsequent precipitation exists, and the positive impact on soil moisture is offset by the drying impact of vegetation on soil moisture. In the response to the 80% dry soil moisture anomalies, however, vegetation in both SM_Veg and SM Anomaly ensembles is severely water stressed and thus LAI differences between SM_Veg and SM Anomaly cannot play a role to change soil moisture, transpiration and precipitation.

[37] For soil moisture anomalies in spring, a negative feedback from vegetation was found in BASE. In response to wet spring soil moisture anomalies, the precipitation and vegetation responses combined together can lead to decreases in soil moisture later during summer, as detailed in KW07b.

[38] 1. The impact of spring soil moisture anomalies on precipitation is not evident until early summer. This is because in this region, convective precipitation (the type that responds to land condition changes) does not become dominant over large-scale precipitation until May or June [KW07a, Figure 9].

[39] 2. There is a delay in vegetation response. Grasses and crops, the dominant land cover types in MRB, respond to both cold stress and water stress, and during spring, vegetation growth is still limited by low temperature, which is further enhanced by the wet soil moisture anomalies [KW07b, Figure 8]. So vegetation in April cannot take advantage of the increased soil moisture, and the increases in LAI do not occur until May.

[40] Based on (1) precipitation responses and (2) vegetation responses together, KW07b suggested that soil moisture is on its way back to normal in April and May in the SM Anomaly, while the enhanced vegetation in the SM_Veg Anomaly speeds up the processes of soil moisture depletion and may even lead to dry anomalies in the soil. Note that in addition to KW07b, the soil drying effects of increased vegetation has also been found in a few previous studies [Notaro et al., 2008; Liu et al., 2010], although not necessarily due to the same reason. As briefly introduced in section 1, Liu et al. [2010] showed that increases in vegetation tend to reduce soil evaporation, which offsets the increase of plant transpiration, leading to a negative feedback. The resulting decrease of precipitation ultimately leads to drier soil. They also found that grasses are more effective than trees in inducing the negative feedback because the shallow roots of grasses limit their transpiration and effectively change evaporation. Consistently, KW07b find the negative vegetation feedback over the MRB, where grasses and crops dominate the land cover (Figure 2).

[41] In response to the 80% wet anomalies applied on 1 April, vegetation feedback to soil moisture and precipitation is not significant although increases in LAI are present in EXP (Figure 12). There are two parallel processes at work. First, precipitation increases over the southeastern North America during May leads to increases in soil moisture during May. These increases in precipitation are attributed to changes in the large-scale flow due to initial soil moisture anomalies although detailed pathways are not identified in this study. The soil moisture anomalies persist for more than three months. Second, increases in initial soil moisture seem to bring about increases in vegetation during June and July, leading to slight decreases in soil moisture in the Upper MRB. In the spring cases, vegetation feedback is very weak, and persistent soil moisture anomalies brought by precipitation anomalies are obvious.

Figure 12.

Differences (SM_Veg Anomaly – SM Anomaly) in (a) LAI, (b) soil water, and (c) precipitation between the SM_Veg Anomaly ensemble and the SM Anomaly ensemble in April, May, June, and July with the EXP model. The anomaly ensembles are initialized with an 80% increase in climatological soil moisture on 1 April, and the averages of five ensemble members are presented. Only differences exceeding the 90% confidence level are shaded. The numbers in the bottom left of each panel indicate averages over the MRB.

5. Conclusions and Discussions

[42] In this study, we investigated the impact of model climatology on soil moisture-vegetation-precipitation feedback over North America. After shutting down the lateral subsurface drainage to reduce the dry biases in CAM3-CLM3, we performed a series of ensemble simulations with the EXP model to be compared to those with the BASE model used in our previous studies ofKW07a and KW07b. When vegetation feedback is ignored, wetter mean soil moisture in EXP (relative to BASE) leads to higher sensitivity of precipitation to dry anomalies than to wet anomalies, which is the opposite of the comparison in BASE. However, in both BASE and EXP, the impact of dry soil moisture anomalies on subsequent precipitation tends to persist longer than the impact of wet soil moisture anomalies. With vegetation feedback included, EXP demonstrates a positive feedback between vegetation and precipitation following both dry and wet soil moisture anomalies in summer, while in BASE the feedback is positive following wet soil moisture anomalies only (due to a strong dry soil bias). In EXP, the magnitude of precipitation changes due to vegetation feedback is comparable to that due to soil moisture feedback when more realistic soil moisture anomalies (the 30% soil moisture anomalies, compared to the 80%) are applied. Also, a major difference is found in spring when the vegetation impact on subsequent precipitation is negative and significant in BASE, but not significant in EXP.

[43] Our study suggests that soil moisture climatology influences the soil moisture-vegetation-precipitation feedback to some extent. Some of our previous findings, inherent to natural processes, are robust regardless of model parameterizations, but some are not. Such model dependencies mainly limit the studies using one individual model and thus a multimodel intercomparison project will be helpful for a comprehensive evaluation of the soil moisture-vegetation-precipitation feedback processes. To this end, the Global Land-Atmosphere Coupling Experiment (GLACE) project involving 12 GCMs in quantifying the strength of soil moisture-precipitation coupling [Koster et al., 2004, 2006; Guo et al., 2006] offers a good precedent to follow.

[44] With the use of CLM3 in this study, one may question how the soil moisture-vegetation-precipitation feedback processes would differ with the latest version of CLM (i.e., CLM4). The performance of CLM3 is adequate in many aspects, and the main deficiency is with respect to the hydrologic cycle [Oleson et al., 2008]. The global-scale partitioning of ET is unrealistic with soil evaporation and canopy evaporation dominating transpiration, and runoff and soil water storage are deficient [Lawrence et al., 2007]. Compared to CLM3, CLM4 include several modifications directly related to the hydrologic cycle, consisting of new parameterizations for canopy integration, canopy interception, frozen soil, soil water availability, and soil evaporation, a TOPMODEL-based model for surface and subsurface runoff, and a groundwater model for determining water table [Oleson et al., 2008]. Among these modifications, canopy interception and soil water availability are closely linked to the soil moisture-vegetation-precipitation feedback processes. For example, the erroneous ET partitioning has been improved with the increased transpiration, decreased canopy evaporation and decreased soil evaporation in CLM4. Thus, increased fraction of precipitation reach the soil and increased fraction of ET comes from soil moisture, implying that the positive soil moisture-vegetation-precipitation feedback would enhance in principle. Also, the new parameterization for soil water availability lowers the soil moisture levels at which plant water stress begins to occur, leading to increased soil water availability for plants. With the new parameterization, the plants response to dry soil moisture anomalies may also become less sensitive, possibly weakening the positive impact of vegetation on soil moisture-precipitation feedback during the summer. It is hard to speculate which might be dominant, and this is subject to further study.

Acknowledgments

[45] This work was supported by the NOAA GEWEX Americas Prediction Project program (NA03OAR4310080). The authors thank Michael Notaro and two anonymous reviewers for their constructive comments on an earlier version of this manuscript.