Sensitivity of model-simulated summertime precipitation over the Mississippi River Basin to the spatial distribution of initial soil moisture



[1] Using a numerical model, the Regional Atmospheric Modeling System (RAMS), we simulate July precipitation over parts of the Mississippi River Basin and surroundings for each of three years, 1995–1997, with six different initial soil moisture patterns: three (control, dry, and wet) with a realistic (observationally based) spatial distribution, and three (control, dry, and wet) with a horizontally homogeneous distribution. Our goal is to determine the impact on future simulated precipitation of changing the initial soil moisture spatial distribution. The spatially homogeneous initial soil moisture pattern represents, in effect, a “wet west/dry east” anomaly imposed on the realistic soil moisture pattern (that reflects the west-to-east climatological gradient). The impact of this anomaly, i.e., increasing soil moisture in the western half and decreasing it in the eastern half of the simulation domain, is most pronounced for the dry experiments and weakens nonlinearly with increasing domain-average initial soil moisture. In the dry regime, the impact is to enhance the total monthly precipitation in both the west and east. We examine the various terms in the atmospheric moisture budget to interpret these results. The changes in precipitation in the runs with a homogeneous compared to realistic initial soil moisture spatial pattern are consistent with enhanced evaporation in the western half of the model domain accompanied by enhanced west-to-east horizontal moisture transport that helps restore the initially depleted soil moisture in the east. In this manner, the zonal moisture flux acts toward re-establishing the initial climatological soil moisture pattern of the region, thus acting as a negative feedback mechanism. In addition, the soil moisture anomaly generally produces diminished meridional moisture transport into the simulation domain from the south through a decrease in the low-level meridional wind speed. This decrease in meridional flux acts in the same direction as the zonal flux change in the west, and in the opposite direction to the zonal flux change in the east. Since this change is most pronounced in the west, it therefore also contributes to the overall negative feedback of the atmospheric dynamics on the initial soil moisture. The persistence timescale of the impact of this particular soil moisture anomaly pattern on precipitation is on the order of 3 months in the dry regime. Sensitivity of the results to a change in convection scheme is also explored.

1. Introduction

[2] Improving hydrometeorological predictability on seasonal and longer timescales is a key goal of the scientific community and an organizing principle for research initiatives such as the GEWEX Continental Scale International Project (GCIP) and the GEWEX Americas Prediction Project (GAPP).

[3] An important potential aid to improving predictability of hydrological and related meteorological variables on these timescales is land memory. Among the chief parameters influencing land memory is soil moisture: Globally averaged soil moisture has been shown to have a persistence timescale of the order of months, longer than any other land-surface parameter [e.g., Robock et al., 2000]. Because of this persistence, soil moisture can have significant, long-lasting effects on the atmosphere [e.g., Beljaars et al., 1996; Liu and Avissar, 1999a, 1999b; Koster and Suarez, 2001]. It has been suggested, for example, that knowledge of the correct soil moisture state leads to improvements in precipitation predictability in many areas including midlatitude regions [Koster and Suarez, 2001; Dirmeyer, 2000]. Therefore, understanding the impact of soil moisture, and specifically, soil moisture anomalies, on the atmosphere may increase our skill when it comes to longer term (i.e., seasonal and longer timescale) weather prediction. Better understanding of the evolving soil moisture state and its influence requires an improved knowledge of the mechanisms, feedbacks, and sensitivities linking the land and the atmosphere.

[4] The land-surface is linked to the overlying atmosphere by means of energy and moisture exchange, and this linkage can be described by the water and energy budgets. The modification of these budgets is responsible for adjustments in the vertical fluxes of heat and moisture, acting as a main source of energy for convective precipitation [Pielke, 2001]. Modification of the local thermodynamics, causing changes in the lifting condensation level (LCL), and convective available potential energy (CAPE), as well as vertical and horizontal moisture transport, can have important consequences regarding the location and timing of convection initiation. The interaction and subsequent feedbacks between the local thermodynamics and large-scale dynamics may provide an added source for thunderstorm fuel. Changes in, for example, surface sensible and latent heat fluxes as a result of changes in soil moisture can drive such changes in convection. Therefore an appropriate representation of the land surface and its interaction with the overlying atmosphere becomes a necessary component for the proper simulation of convective precipitation.

[5] It has been shown that rainfall, and related atmospheric variables, are sensitive to soil moisture, especially during summer months when convective activity is most pronounced and when precipitation efficiency is at its highest [e.g., Shukla and Mintz, 1982; Rind, 1982; Pan et al., 1996; Baker et al., 2001]. Since long-term records of soil moisture do not exist in most parts of the world, much of the work regarding the influence of soil moisture on the atmosphere has been accomplished through the use of numerical models. For example, Pal and Eltahir [2001] have used a regional climate model forced by different magnitudes of spatially homogeneous initial soil moisture to examine the feedbacks between soil moisture conditions and future summer rainfall in the Great Plains of the United States. They propose that anomalously high soil moisture increases the frequency and magnitude of convective rainfall by decreasing the boundary layer depth and thus increasing moist static energy per unit mass of air, by decreasing the entrainment of low moist static energy air from above the boundary layer, and by increasing moist static energy flux from the surface via an increase in surface radiation. They find an asymmetric response to initial soil moisture conditions whereby the positive soil moisture-rainfall feedback is more pronounced for dry than for wet conditions, and they suggest that there exists a threshold where a further increase in soil wetness reverses the positive feedback and decreases both the frequency and magnitude of convective events.

[6] Similarly, Schar et al. [1999] have investigated the soil moisture-rainfall feedback through a number of continental-scale numerical experiments over Europe. Again making use of a homogeneously initialized soil moisture, they find a positive feedback between soil moisture and rainfall. They also find that local recycling alone does not explain the surplus of precipitation occurring over wet rather than dry soils. They suggest that a large fraction of the precipitation occurring over wet soils is derived from atmospheric advection, but that the same advection occurring over dry soils would not result in precipitation; in other words, precipitation dynamics seem to operate differently over dry versus wet soils.

[7] Although previous numerical modeling work, such as the studies cited above, has explored the role of feedbacks with soil moisture in warm season continental precipitation, the focus has largely been on how domain average initial soil moisture amount conditions future precipitation. While these investigations have greatly increased our understanding of the relevant feedback pathways, the role of initial spatial distribution has received less attention. Nevertheless, there have been some investigators who have looked into this issue from both a regional and global modeling perspective. Pal and Eltahir [2002] have used a regional climate model to investigate teleconnections between soil moisture in the southwest United States and midwest summer precipitation during the 1993 North America summer floods. They conclude that soil moisture anomalies can be important both locally and remotely via large-scale dynamics, depending on their location. Small [2001] also found that soil moisture forcing relies heavily on the location of the anomalous surface conditions in their investigation of the influence of soil moisture anomalies on the variability of the North American Monsoon System. Fennessy and Shukla [1999] have used a GCM to examine the importance of initial soil wetness by using numerical integrations consisting of four ensembles, each initialized with a different soil wetness. They find several factors that determine the strength of the impact of initial soil wetness difference on precipitation (among other fields) including the magnitude of initial soil wetness difference and soil wetness persistence, strength of solar forcing, accessibility of (other) moisture sources, and dynamical circulation. Douville and Chauvin [2000] produced a global soil moisture climatology for 1987–1988 to determine that soil moisture has a significant impact on planetary-scale circulation variability. Their results indicate that realistic initial soil moisture conditions lead to improved summer climate predictions in the Northern Hemisphere.

[8] We wish to extend some of this work by focusing on the Mississippi River basin, particularly during non-extreme years. For example, what effect would a drier than normal spring in the eastern part of this region that persisted into the early summer months, accompanied by wetter than normal conditions in the west, have on summertime basin-wide rainfall, and how might this anomaly evolve (persist) into the upcoming summer months? Such knowledge, relating to specific spatial patterns of soil moisture, would be of practical benefit, for example, to the agricultural and water management communities.

[9] In this paper, we use the Regional Atmospheric Modeling System (RAMS) to study the impact of soil moisture spatial distribution on precipitation in the Mississippi River Basin during the GCIP Enhanced Seasonal Observing Periods of 1995, 1996, and 1997 (ESOP-95, ESOP-96, ESOP-97) (see Coughlan and Avissar [1996] for an introduction to the GCIP initiative). Convectively generated precipitation is important in this region and season: the Mississippi River Basin, covering nearly half the area of the contiguous 48 states, receives as much as 70–90% of its annual rainfall from thunderstorms [Changnon, 2001]. Mesoscale Convective Systems (MCSs), organized mesoscale/synoptic-scale clusters of thunderstorms, make up 30–70% of the warm season precipitation in the central United States [Fritsch et al., 1986; Bernardet et al., 2000]. A study of shorter timescale processes (i.e., synoptic to monthly) linking soil moisture to convection and precipitation is necessary for an improved understanding of the longer timescale (seasonal and longer) evolution of the system. Our primary focus lies in furthering our understanding of the 3-D dynamical mechanisms involved in the soil moisture-rainfall feedback. We expect our results to have implications for future studies of soil moisture persistence and hydrometeorological predictability.

[10] In section 2 we discuss the model setup and methods used in our analysis. Section 3 presents an evaluation of model performance and the results of the soil moisture sensitivity experiments. Our conclusions are given in section 4.

2. Methods

2.1. Model

[11] All simulations are performed using RAMS version 4.3 [Walko and Tremback, 2000]. RAMS is a non-hydrostatic model that solves the full nonlinear equations of motion, and includes a comprehensive soil and vegetation model [Walko et al., 2000b] and parameterizations of subgrid convection (see below), turbulence [Mellor and Yamada, 1974, 1982], radiation [Harrington, 1997], and cloud and precipitation microphysics that is prognostic in both mixing ratio and number concentration for individual microphysical species [Walko et al., 2000a; Meyers et al., 1997].

[12] The soil and vegetation component of RAMS, known as the Land Ecosystem-Atmosphere Feedback Model (LEAF-2 [Walko et al., 2000b]) is prognostic in energy and water exchange between the land surface and the overlying atmosphere. LEAF-2 allows for the division of each RAMS grid cell into multiple patches. Each patch interacts individually with the overlying atmosphere independent of adjacent patches. For coarse resolution simulations, this enhanced capability allows the user to include a more realistic representation of some aspects of the interaction between the atmosphere and the heterogeneous land surface without compromising the computing time gained by using a fewer number of atmospheric grid points.

[13] The resolution of the simulations makes use of a sub-grid-scale convective scheme necessary. The standard RAMS convection parameterization is a modified Kuo scheme [Kuo, 1974]. More recently, a new option has been implemented into RAMS allowing the use of the Kain-Fritsch scheme [Kain and Fritsch, 1992; C. Castro, personal communication, 2002].

[14] Our simulations consist of one coarse grid with a horizontal grid spacing of 40 km. Our model domain, centered over Oklahoma, encompasses a 2200 km × 2200 km area (see Figure 1), has 38 vertical atmospheric levels up to 22 km, with 19 levels in the first 3 km, and includes 11 soil layers down to 2 m, with the first eight soil layers in the topmost meter of soil. The four outermost points of the lateral boundaries are nudged toward the NCEP reanalysis (see below) every time step with a RAMS nudging timescale of 7200 s. These points are not included in the subsequent analysis. It is worth noting that the results of regional modeling studies have been shown to be sensitive to the user's choice of domain size [Seth and Georgi, 1998]. In general, larger modeling domains may be desirable for many applications (given the same resolution). We have performed two additional simulations using a larger domain size (5200 km × 2800 km) to examine this sensitivity in our results, and we discuss them briefly in section 3. We note that the domain size used in most of our experiments is roughly equivalent to recent previous regional modeling studies of soil moisture-rainfall interactions.

Figure 1.

RAMS domain with topography (in m) overlaid.

2.2. Data

[15] Our model requires initial surface and boundary conditions. Each of our experiments is initialized with standard pressure level data (geopotential height, relative humidity, vorticity, temperature, and winds) available from the National Center for Environmental Prediction (NCEP) National Center for Atmospheric research (NCAR) reanalysis project [Kalnay et al., 1996]. The surface is initialized using the appropriate volumetric soil moisture and soil temperature for the initial starting model time, also obtained from the NCEP/NCAR reanalysis. We make use of Land Data Assimilation System (LDAS) soil texture, a data set re-sampled to 1/8° horizontal grid spacing with 11 vertical layers from the surface to 2 m and 16 soil texture classes [Reynolds et al., 2000]. Among the important parameters included in the LDAS soil texture data set are the relative percentages of sand, silt, and clay, and the porosity of a particular soil type. More information about this and similar data sets can be found at

[16] To aid in the interpretation of the numerical integrations, we make use of the Arkansas Basin Red-River Forecast Center (ABRFC) hourly precipitation data (more information about this and other similar data sets can be found at ABRFC gridded precipitation data is a blended rain gauge and radar measurement product with a horizontal grid spacing of 4 km, and encompasses an area roughly one half the size of our analysis domain.

2.3. Experiments

[17] This study aims to improve our understanding of the link between initial soil moisture conditions and their subsequent impact on simulated summertime precipitation. Specifically, we focus on the impact of the spatial distribution of initial soil moisture relative to the impact of the magnitude of initial soil water content. We perform 12 monthly simulations each for three different Julys: 1995, 1996, and 1997. These case study months coincide with the GCIP ESOP-95, ESOP-96, and ESOP-97 periods. The simulations consist of a control simulation (CONTROL) and five sensitivity tests (TEST1 through TEST5). The total of 12 results from our carrying out each simulation using both the Kuo and Kain-Fritsch convective schemes to examine the sensitivity of our results to this aspect of model physics. These simulations are described below and summarized in Table 1. CONTROL for each of the case study years initializes the land component of RAMS with the unmodified NCEP reanalysis volumetric soil moisture. The other simulations examine the physical aspect of the soil moisture-atmosphere relationship from two angles: the effects of varying the spatial distribution of the initial soil moisture and the effects of varying the magnitude of the initial soil moisture. First, we perform a set of two simulations (TEST1 and TEST2) wherein we vary the volumetric soil moisture of CONTROL by decreasing and increasing, by 30%, respectively, at every grid point in the model domain. In this way we conserve the initial soil moisture variability from the CONTROL simulation but change the total water content in each grid cell (as a function of the soil texture's water holding capacity) (shown in Figure 2, for 1995). For TEST2, in the very small number of grid cells where local soil moisture exceeded saturation, the excess water is discarded; we do not expect this to significantly impact our results. The second series of simulations (TEST3, TEST4, and TEST5) are configured by averaging the volumetric soil moisture at each soil level of each of CONTROL, TEST1, and TEST2 across the entire model domain; this is to say that soil moisture is initially homogeneous with the same mean value as CONTROL, TEST1, and TEST2. By smoothing out the heterogeneity in initial soil water content, we hope to discern the relative importance of initial soil moisture spatial variability and amount on simulated summertime precipitation.

Figure 2.

RAMS initial, vertically averaged (top 30 cm), volumetric soil moisture for July 1995 for (a) CONTROL experiment, (b) TEST1, and (c) TEST2.

Table 1. Summary of 1-Month Experiments Performed, Using Both the Kain-Fritsch and Kuo Convective Schemes
 1995 Kuo1995 Kain-Fritsch1996 Kuo1996 Kain-Fritsch1997 Kuo1997 Kain-Fritsch
−30% of NCEP Reanalysis at every grid point of domainTEST1TEST1TEST1TEST1TEST1TEST1
+30% of NCEP Reanalysis at every grid point of domainTEST2TEST2TEST2TEST2TEST2TEST2

[18] We perform two additional 5-month-long simulations lasting from the beginning of May through the end of September 1995. Even though each simulation begins in May, we use the TEST1 and TEST4 July 1995 initial soil moisture patterns. The motivation for the longer simulations is to help us in answering the following questions: Is the dynamical behavior in May in response to an altered pattern of initial soil moisture similar to that in July, and what is the persistence of the soil moisture anomaly and its impact on precipitation? The reason for choosing the drier regime (TEST1 and TEST4 as opposed to another simulation pair) for these longer simulations is because, as will be shown below, this is where the response seems to be the greatest.

2.4. Atmospheric Moisture Budget Analysis

[19] A large part of our diagnosis involves an analysis of each simulation's atmospheric moisture budget. We briefly review our procedure for these calculations below. For a more detailed analysis of this topic the interested reader is referred to Peixoto and Oort [1992].

[20] The balance equation for the atmospheric branch of the hydrologic cycle may be simplified to

equation image

The overbar, in all cases, represents a time average. The first term represents the tendency of vertically-integrated atmospheric water vapor, commonly referred to as “precipitable water,” i.e., the storage term, W. Precipitable water is the total water mass contained in a column of air,

equation image

[21] The rearrangement of the terms in the hydrostatic equation yields units of kg m−2. For computational savings we limit the top pressure level to correspond to the model level at roughly 8400 m. Vertical fluxes of water vapor through this interface are minimal.

[22] The second term in equation (1) represents the time-averaged, vertically integrated horizontal divergence of water vapor. This is simply the time-averaged horizontal divergence of precipitable water. Thus, temporarily omitting the time averaging notation, the horizontal transport of water vapor may be separated into its two components [e.g., see Peixoto and Oort, 1992],

equation image
equation image

where equations (3a) and (3b) are the zonal and meridional components of Q (i.e., the moisture fluxes), respectively.

[23] The net divergence of water vapor may be calculated using the Gauss (or Divergence) theorem from vector calculus [e.g., see Stewart, 1995] as a line integral of the flux around the border enclosing the region for which divergence is desired, divided by the area of the region, A,

equation image

The term inside the integral is the dot product of the moisture flux and the unit vector normal to the line around which we integrate. This convention allows us to assign a positive value to outgoing fluxes and a negative value to incoming fluxes to obtain the proper sign for divergence.

[24] For a rectangular domain, the contribution to the net divergence from the fluxes at each of the four faces can be calculated individually; therefore in the paper we will also refer to incoming and outgoing zonal and meridional fluxes for the given region, in addition to the divergence resulting from them. The remaining two terms on the right side of Equation (1) are the time and domain averaged evaporation (source) and precipitation (sink).

[25] Though our calculations account for most of the water vapor in the domain, because we are calculating the various integrals numerically (along with other small approximations), the budget is not entirely closed. The budget equation is therefore more correctly written with a residual term on the right-hand side representing the degree of error in our approach,

equation image

[26] The residual from our calculations for each case is in general about 1 order of magnitude or less than the key terms in the balance equation. Therefore we feel justified in using the atmospheric water budget as a tool in our analysis.

3. Results and Discussion

3.1. Model Validation

[27] As the primary focus of this paper is on the impact of soil moisture on future precipitation, we briefly evaluate the model's precipitation performance. We compare the model simulated monthly precipitation against ABRFC observations over the subset of the model domain covered by the ABRFC network (roughly one half of the total analysis domain).

[28] Figure 3 shows the time-averaged ABRFC observed (top figure in each pair) and RAMS simulated (bottom figure in each pair) precipitation for each of July 1995 (Figure 3a), 1996 (Figure 3b), and 1997 (Figure 3c), in units of mm/hr. The RAMS output shown here is from the CONTROL experiment that makes use of the Kain-Fritsch scheme. The local precipitation maxima and minima are generally reproduced in the same geographical locations as in the observations. Some of the interannual variability in the observations is also represented: The model captures the generally northwest-southeast orientation of the band of maximum rainfall in 1995 and 1996 as well as the opposite (southwest-northeast) orientation in 1997, and it correctly projects 1996 as the wettest of the three Julys.

Figure 3.

ABRFC observed (top figure of each pair) and RAMS CONTROL simulation (bottom figure of each pair) monthly averaged total precipitation (mm hr−1) for (a) July 1995, (b) July 1996, and (c) July 1997, using the Kain-Fritsch convective scheme.

[29] On the other hand, notable shortcomings of the CONTROL simulations include the following. RAMS simulates high precipitation in the lee of the Rockies that does not appear in the observations (particularly in 1995 and 1997). Additionally, the Kain-Fritsch scheme overestimates monthly averaged precipitation for 1995 while it underpredicts precipitation for 1996 and 1997. For both 1996 and 1997, somewhat of a northward shift of the main axis of simulated precipitation serves to decrease the simulated versus observed precipitation rates and degrades the observational intercomparison.

[30] Figure 4 shows the hourly, domain-averaged RAMS simulated and ABRFC observed precipitation for each of July 1995 (Figure 4a), 1996 (Figure 4b), and 1997 (Figure 4c), also in units of mm/hr. The model is usually able to correctly simulate the observed episodic precipitation events and intervening dry periods: More often than not, when precipitation occurred over the ABRFC domain, our CONTROL experiments were able to simulate its onset and magnitude. However, a deficiency of all simulations was the inability of the model to accurately predict the high magnitude of a few extreme events. Especially noticeable are two key features of the 1996 CONTROL run toward the middle and end of the month when domain averages were in excess of 1 mm/hr. This is a common shortcoming of most mesoscale model simulations of convective precipitation [e.g., see Bernardet et al., 2000; Stensrud et al., 2000]. In spite of this, the hour-to-hour variability in domain-averaged precipitation seemed reasonably well reproduced. The mean 1996 CONTROL variance was underestimated by 45% (Table 2). However, the mean simulated variance was underestimated by less than 5% and 25% for 1995 CONTROL, and 1997 CONTROL, respectively, months without such extreme precipitation events (e.g., in excess of 1 mm/hr). When comparing the total monthly domain averaged precipitation for each of the 3 years with respect to ABRFC observations, RAMS overestimates by 18.4% in 1995, underestimates by 22.4% in 1996, and underestimates by 26.7% in 1997 (Table 2). As noted above, some of this disagreement can be attributed to the slight latitudinal shift of the model-simulated precipitation relative to the observations.

Figure 4.

ABRFC observed and RAMS CONTROL hourly domain averaged total precipitation (mm hr−1) for (a) July 1995, (b) July 1996, and (c) July 1997, using the Kain-Fritsch convective scheme.

Table 2. Summary of Precipitation Statistics for the July 1995–1997 CONTROL Simulations Using the Kain-Fritsch Convective Scheme Over the ABRFC Domain
July 1995
Hourly domain averaged precipitation sum, mm62.3073.74 
Hourly domain averaged precipitation mean, mm/hr0.090.10 
Hourly domain averaged precipitation variance, (mm/hr)20.020.02 
RMSE, mm/hr  0.16
July 1996
Hourly domain averaged precipitation sum, mm113.4588.01 
Hourly domain averaged precipitation mean, mm/hr0.160.12 
Hourly domain averaged precipitation variance, (mm/hr)20.040.02 
RMSE, mm/hr  0.21
July 1997
Hourly domain averaged precipitation sum, mm74.1954.42 
Hourly domain averaged precipitation mean, mm/hr0.100.076 
Hourly domain averaged precipitation variance, (mm/hr)20.020.012 
RMSE, mm/hr  0.15

[31] In addition, we calculate the root-mean square error (RMSE) between the three different monthly averaged CONTROL simulations and precipitation observations (see Table 2). The best agreement, as indicated by RMSE, occurs for 1995 and 1997, and the worst for 1996. This emphasizes, again, the deteriorating impact that a small number of isolated, extreme precipitation events may have on monthly mean statistics.

[32] The agreement with observations was generally not as good for the simulations using the Kuo convective scheme (not shown), though the basic features were still represented. For this reason, we mostly discuss our results obtained with the Kain-Fritsch scheme. At the end of the paper, we compare the results of the sensitivity experiments using the two schemes, and we discuss the differences introduced and their implications for our conclusions.

3.2. Results of the Soil Moisture Sensitivity Experiments

3.2.1. 1995: Domain-Averaged Precipitation

[33] Figure 5 depicts the RAMS simulated monthly averaged total precipitation for the entire analysis domain for all July 1995 experiments (using the Kain-Fritsch scheme). In general, increasing the initial soil moisture amount tends to produce greater precipitation (e.g., compare TEST1 to TEST2). This is in agreement with Pal and Eltahir [2001] and Pan et al. [1996], among others. Also in agreement with these studies, the sensitivity of future precipitation to initial soil moisture seems to be greatest for the driest initial soil moistures (for example, compare TEST1 to CONTROL and TEST2 to CONTROL). In other words, although the initial soil moisture magnitude changes by an equivalent percentage from CONTROL to TEST1 and from CONTROL to TEST2 (and actually, by a greater absolute magnitude in the wetter experiments; see Figure 6) precipitation differences by the end of the month are greatest between CONTROL and TEST1.

Figure 5.

RAMS simulated July 1995 precipitation (mm hr−1) for (a) TEST1, (b) TEST4, (c) CONTROL, (d) TEST3, (e) TEST2, and (f) TEST5, using the Kain-Fritsch convective scheme.

Figure 6.

RAMS, July 1995, initial soil moisture difference: (a) TEST1 minus TEST4, (b) CONTROL minus TEST3, and (c) TEST2 minus TEST5.

[34] Our main focus, however, is on the impact of initial soil moisture spatial distribution on future rainfall, rather than the effect of initial soil moisture amount. Therefore the key comparisons we will make will be between each of the two members of the realistic/homogeneous pairs: TEST1 and TEST4, CONTROL and TEST3, and TEST2 and TEST5. Many of the results discussed in this section are summarized in Table 3.

Table 3. Summary of Atmospheric Water Budget Components for all 1-Month Simulations Using Kain-Fritsch Convective Schemea
  • a

    All units are in mm day−1.

July 1995
July 1995 (W)
July 1995 (E)
July 1996
July 1996 (W)
July 1996 (E)
July 1997
July 1997 (W)
July 1997 (E)

[35] Figure 6 displays the initial soil moisture difference between each of the July 1995 simulation pairs. Because the climatological soil moisture pattern is a generally drier west and a generally moister east, we can think of the initial soil moisture pattern in the homogeneous simulation as essentially a removal of water from the eastern half of the domain and a relocation to the west. Since each of the three heterogeneous simulations reflect the dry west/wet east climatology, the resulting homogeneous simulations can be thought of as anomalous with respect to them, i.e., a “wet west/dry east” initial soil moisture anomaly pattern. What influence does this shift in soil moisture pattern have on the resulting hydrometeorology for the given month?

[36] Figure 7 shows the RAMS simulated monthly averaged total precipitation difference, and convective precipitation difference, between the various experiment pairs for July 1995. On each map we draw a line of demarcation across our domain, thus dividing it into western and eastern halves (also roughly corresponding to the location of the greatest west/east gradient in initial soil moisture). Below we will examine both subdomains individually. Figure 7 shows that, in the dry case, in the homogeneous initial soil moisture simulations, precipitation increases both in the west (e.g., the increase from TEST1 to TEST4 is from 1.16 mm/day to 1.41 mm/day), where soil moisture was initially greater (recall Figure 6), and in the east (e.g., the increase from TEST1 to TEST4 is from 2.79 mm/day to 3.25 mm/day), where soil moisture was initially lower (since we are subtracting TEST4 from TEST1, light shades indicate greater precipitation in the simulation with homogeneous initial soil moisture). A similar result holds for CONTROL and TEST3, although the difference in precipitation is much less. TEST5, however, produces slightly less precipitation in the east than its initially heterogeneous counterpart (3.78 mm/day for TEST5 as compared to 3.85 mm/day for TEST2), though again, the differences between TEST2 and TEST5 are relatively small compared to those between TEST1 and TEST4. This shift from a positive to negative feedback at high values of soil moisture is also consistent with Pal and Eltahir [2001].

Figure 7.

RAMS simulated total precipitation (mm hr−1) difference (left plot) and convective precipitation difference (right plot): (a, b) TEST1 minus TEST4, (c, d) CONTROL minus TEST3, and (e, f) TEST2 minus TEST5, averaged for July 1995.

[37] We note that most of the differences in precipitation between simulation pairs are due to differences in precipitation yielded by the convective scheme (the three rightmost panels of Figure 7). This is not surprising, as the convective scheme generates nearly 80% of all precipitation during each of the three Julys studied.

3.2.2. Water Budget Analysis: 1995

[38] In our analysis of the atmospheric water budget for July 1995, we focus first on the dry regime (TEST1 and TEST4) because the sensitivity is greatest there. Figure 8 shows the water budget calculated separately for the western and eastern halves of our analysis domain for TEST1 and TEST4, following the methodology outlined in section 2.4 (see also Table 3). Initializing TEST4 with a “wet west/dry east” soil moisture anomaly (recall Figure 6) produces a precipitation increase of 22% in the west (W) box, from 1.16 mm/day to 1.41 mm/day. However, evaporation in the west increases by 31% from 2.64 mm/day (TEST1) to 3.45 mm/day (TEST4): the increased precipitation in the western portion of TEST4 does not account for all the extra water that has evaporated into the atmosphere.

Figure 8.

Atmospheric water budget diagram for (top) TEST1 and (bottom) TEST4. Each horizontal box corresponds to the RAMS analysis domain, which is further subdivided into a West and East box. E and P stand for evaporation and precipitation, respectively. Numerals and arrows indicate incoming and outgoing zonal and meridional moisture fluxes. Units are mm day−1.

[39] Much of the discrepancy is made up by an enhanced zonal moisture transport out of the W box and into the east (E) box. There is also a decrease in incoming meridional moisture flux at the southern boundary of the W box, and this is also a contributing factor that will be discussed further below. The incoming zonal and outgoing meridional moisture fluxes do not exhibit much change; thus the net moisture flux divergence from the W box goes up. It is important to note that this change in zonal flux is due primarily to changes in atmospheric water vapor content, and not to changes in wind speed.

[40] Conversely, since the zonal flux is the primary conduit of moisture between the western and eastern halves of the domain, the marked increase in the zonal transport of water from W to E, in TEST4 compared to TEST1, produces a significant decrease in moisture flux divergence (i.e., increase in convergence) in the E box. Similarly to the W box, the other flux components exhibit much less change. This increase in incoming zonal moisture flux and decrease in moisture divergence is accompanied by a 16% increase in E box precipitation, from 2.79 mm/day to 3.25 mm/day. This is a smaller percentage but larger absolute change when compared with the W box, as is apparent from Figure 7.

[41] In this manner, the zonal moisture flux seems to act toward re-establishing the more climatological soil moisture distribution of the region; over the course of the month it dampens the effect of the imposed soil moisture perturbation. Thus, the zonal transport is a negative feedback mechanism on this particular (“wet west/dry east”) soil moisture anomaly pattern.

[42] For additional insight, we examine the time evolution over the course of the month of the differences, between TEST1 and TEST4, in key water budget components in each box (Figure 9). The time series of these differences are shown separately for the W (Figure 9a) and E (Figure 9b) boxes. Starting with Figure 9a, we see that the initial addition of extra water in the W box results in a significant increase of evaporation in TEST4 compared to TEST1 through the middle of the month (i.e., large negative values of the TEST1–TEST4 evaporation difference curve). A considerable portion of the excess water does not precipitate locally (note the magnitude of the TEST1–TEST4 precipitation difference in the W box is generally smaller than the magnitude of the evaporative difference there) but instead migrates into the E box; the TEST1–TEST4 difference in zonal flux out of the W box (and therefore into the E box) strongly resembles the evaporative difference curve in the W box after the mid-month evaporation peak. In Figure 9b, we see that this enhanced zonal moisture transport into the E box in TEST4 results in reduced divergence there, which is followed by enhanced E-box precipitation in TEST4 compared to TEST1 (and hence a negative TEST1–TEST4 precipitation difference curve in Figure 9b).

Figure 9.

Evolution of the 5-day running mean between TEST1 and TEST4 of water budget components over the (a) W and (b) E box, separately, for 1995. In both plots, the moisture flux difference indicates transport from W (outgoing) box to E (incoming) box.

[43] In both boxes, changes in precipitation recycling from TEST1 to TEST4 seem to play a significant role. For example, the mid-month peaks in TEST4 relative to TEST1 precipitation in both the W and E boxes are followed by increases in evaporation in TEST4 compared to TEST1. In the W box, this secondary evaporation peak in TEST4 is followed by a secondary peak in horizontal export to the E box, and this additional transport into the E box is quickly followed by a temporary rise in TEST4 precipitation there. Over the course of the month, precipitation moistens the soil and causes a subsequent increase in evaporation, which then either re-precipitates in the box (positive local feedback) or is transported away. In other words, changes in the local recycling of water seem to operate in conjunction with the changes in atmospheric moisture transport to produce the observed domain-wide response to the initial soil moisture anomaly pattern.

[44] Figure 10 shows the water budget calculated separately for the western and eastern parts of the RAMS domain for TEST2 for July 1995. As in TEST4, initializing TEST5 with a “wet west/dry east” soil moisture anomaly produces an increase in evaporation in the W box, though not nearly as large. This smaller evaporation change does not force a significant response of the zonal moisture flux, and, robbed of this extra source of atmospheric water vapor from the W, the E-box precipitation actually decreases slightly, rather than increasing significantly as in TEST4. In the W, the precipitation goes up somewhat, consistent with the evaporation increase. In percentage terms, however, the changes in precipitation from TEST2 to TEST5 in both boxes are much smaller than those from TEST1 to TEST4. Similar to the findings from earlier investigations on the sensitivity to soil moisture amount, soil moisture distribution seems to have little effect on future precipitation for wet soils. Both TEST2 and TEST5 initialize the surface boundary with a relatively high magnitude of soil moisture, and the response to an initial horizontal redistribution of this water seems to be small; that is, some soil moisture is moved from W to E, but in both experiments both boxes remain relatively wet. The major exception to the small moisture flux changes between TEST2 and TEST5 is the incoming meridional flux in the W, where a decrease of 0.18 mm/day in TEST5 compared to TEST2 occurs. We discuss this further shortly.

Figure 10.

Atmospheric water budget diagram for (top) TEST2 and (bottom) TEST5. The convention for arrows is the same as in the previous water budget diagram (Figure 8).

[45] The response of the system in the intermediate case, i.e., CONTROL compared to TEST3, lies between the dry and wet cases (see Table 3). There is some enhanced W-to-E zonal moisture transport in TEST3 (an increase from 8.12 to 8.31 mm/day) resulting from the enhanced W-box evaporation, but, similarly to the TEST2/TEST5 comparison, the changes in precipitation are relatively small, again illustrating the nonlinearly weaker response with progressively wetter soils.

[46] As mentioned above, in addition to the changes in zonal moisture transport from the heterogeneous to homogeneous initial soil moisture runs, there are also differences in meridional moisture transport. Let us examine these differences further for both the dry and wet regimes. In the dry experiment (TEST4 compared to TEST1), though the change in meridional moisture transport is less significant compared to the change in zonal transport, it still plays a role. In both the western and eastern parts of our subdivided domain, the incoming meridional moisture flux at the southern boundary decreases in the simulations with the homogeneous compared to the heterogeneous soil moisture distribution. This difference is most pronounced in the W. Distinct from the changes in the zonal flux that we have already discussed, the main cause of this change in meridional flux is a change in meridional wind speed at low levels, and not a change in atmospheric water vapor content. Figure 11 shows a vertical cross-section of the monthly averaged meridional wind speed at the southern extent of our analysis boundary (i.e., ∼28°N) (Figures 11a and 11b for TEST1 and TEST4). In all simulations, this meridional low-level jet is centered somewhere below about 1 km above the surface, and the corresponding meridional moisture transport thus occurs primarily in the lower levels. Decreasing the initial east-west gradient in soil water correspondingly decreases the surface heating difference between west and east, as more of the incoming solar radiation in the west and less in the east goes into evaporation rather than heating the surface. This decreases the strength of the low-level jet via the thermal wind relation. Furthermore, changes in the relative west-east distribution of cloud cover that accompany changes in convection may also be important. We note that these changes in the surface heating gradient, and hence meridional wind speed, occur not just between the W and E boxes, but within each box as well (as the relevant gradients are continuous across the entire domain). Therefore both the W and E incoming meridional moisture transport decrease from the heterogeneous to homogeneous initial soil moisture case.

Figure 11.

Monthly (July 1995) averaged cross-section of meridional wind (m s−1) at 28°N for (a) TEST1, (b) TEST4, (c) TEST2, and (d) TEST5.

[47] There is a similar decrease in the meridional wind for TEST5, although the difference is of smaller magnitude than for TEST1–TEST4. Consistent with this change in wind speed, the moisture fluxes also decrease for TEST5 compared to TEST2.

[48] This decrease in incoming meridional flux in the homogeneously initialized simulations contributes to drying of both the W and E boxes (i.e., increase in net moisture divergence). By contrast, the zonal flux from the W to the E box contributes to drying in the W and moistening in the E. Therefore the meridional flux change acts in the same direction as the zonal flux change in the W, and in the opposite direction to the zonal flux change in the E. In general across all simulations, however, the changes in meridional flux in the W box are more pronounced than those in the E box (e.g., see Table 3). As a result, we can consider meridional flux changes, in net, as contributing to the overall negative feedback on the initial “wet west/dry east” soil moisture anomaly pattern.

[49] These dynamical adjustments progressively modify the initial soil moisture anomaly. Figure 12 demonstrates that the July 1995 initial soil moisture differences between the homogeneous and heterogeneous runs are diminished by the end of the month, considerably so for the dry case (compare to Figure 6). This emphasizes again that the overall effect of the changes in evaporation, transport, and precipitation is to generally drive the system back toward a more climatological distribution of soil moisture.

Figure 12.

RAMS, July 1995, final soil moisture difference: (a) TEST1 minus TEST4, (b) TEST2 minus TEST5, and (c) CONTROL minus TEST3.

3.2.3. Interactions Between Dynamics and Thermodynamics: 1995

[50] While it seems reasonable based on the above findings that changes in 3-D atmospheric dynamical transport can play a critical role in redistributing soil moisture anomalies, thereby significantly influencing precipitation on synoptic-to-monthly timescales (and possibly longer), some additional discussion is warranted. Specifically, the atmospheric dynamics and horizontal moisture flows mostly act indirectly on precipitation, at least during the warm season, by altering the vertical thermodynamic structure of the atmosphere and changing the frequency and intensity of convection. This is suggested by Figure 7, for example, which shows that most of the change in precipitation between simulations occurs in the convective component. Therefore the precipitation response to a specific soil moisture anomaly pattern occurs as a result of interactions between the 3-D atmospheric dynamics and the 1-D (vertical) thermodynamics. We explore this briefly here.

[51] Previous investigations have provided considerable insight into the vertical processes that contribute to feedbacks between soil moisture and precipitation [e.g., Betts et al., 1996; Findell and Eltahir, 1997; Pal and Eltahir, 2001; Findell and Eltahir, 2003a, 2003b]. For example, higher soil moisture might be followed by an increase in evaporation, accompanied by a decrease in both the boundary layer depth and entrainment of air from above the boundary layer, thus leading to an increase in moist static energy per unit mass of boundary layer air and hence an increase in convective precipitation. Alternatively, under the right atmospheric conditions, increased convective precipitation might be expected over drier soils (a negative feedback). This would come about if the enhanced sensible heat flux, and hence deeper and more actively growing daytime convective boundary layer, lifted parcels to a level of free convection that they otherwise would not have been able to reach. In other words, this mechanism might be most effective in atmospheres where the limiting factor for convection is instability [e.g., see Bernardet et al., 2000]. Recent work by Findell and Eltahir [2003a, 2003b] suggests that, over most of the region coinciding with our simulation domain, summertime convective precipitation is probably expected to increase with a wetter, rather than drier, initial soil moisture; i.e., where a positive feedback between soil moisture and rainfall should operate. Nevertheless, more work is needed in this area.

[52] Figure 13 shows profiles of monthly mean RAMS simulated θe, separately for the W and E boxes, for all six July 1995 simulations. Large differences between the numerical experiments are apparent in the W: The dry simulations (TEST1 and TEST4) present the most stable profiles, while the rest of the simulations are grouped more closely together. In addition, the difference between TEST1 and TEST4 is relatively large. This again illustrates the nonlinear behavior of the system across dry to wet regimes. By contrast, in the E, the differences between simulations are much smaller, though again TEST1 and TEST4 are the most stable, and the largest differences in the profiles are between TEST1 and TEST4, and between TEST4 and the other cases. We note also that the E box is characterized by less stable profiles than the W box for all cases.

Figure 13.

Vertical profiles of time (July 1995) and domain averaged equivalent potential temperature for simulations using Kain-Fritsch scheme over (a) W and (b) East box.

[53] Considering the changes between TEST1 and TEST4 as presented in Figures 8, 12, and 13 as our example, we can now articulate a simple conceptual model for the response of the system to the soil moisture anomaly pattern. In the W, enhanced evaporation in TEST4 compared to TEST1 increases the boundary layer moist static energy and the overall vertical instability. Enhanced outgoing zonal moisture transport to the E, combined with diminished incoming meridional moisture transport at the southern boundary, result in an overall increase in atmospheric moisture divergence. This increase in moisture divergence mitigates some of the potential effects of the decrease in stability, and the result is only a small increase in W-box convective rainfall. Conversely, in the E, the enhanced zonal moisture transport into the box drives an overall decrease in atmospheric moisture divergence (increase in convergence). This results in an increase in convective rainfall, which replenishes the soil moisture, significantly reducing its initial deficit. This combination of enhanced transport and local recycling is efficient enough that both monthly mean evaporation (recall Figure 8) and the average low-level instability (recall Figure 13b) in the E box are even slightly greater in TEST4 than in TEST1, in spite of TEST4 being initialized with substantially less soil moisture.

[54] This is the general picture for July 1995. We need to test whether similar dynamical responses to the imposed soil moisture anomaly patterns occur consistently in the sensitivity experiments for the remaining 2 years.

3.2.4. Water Budget Analysis: 1996 and 1997

[55] Figure 14 shows the atmospheric water budget, calculated separately for the western and eastern parts of the analysis domain for TEST1 and TEST4, for both 1996 and 1997. As was the case for 1995, the effect of initializing the western part of the domain with a wet anomaly is to increase the W-box evaporation, the W-to-E zonal moisture transport, and the precipitation in both W and E. In fact, the enhancement in both evaporation in the W and zonal flux from the W to the E box are both greater in 1996 and 1997 than in 1995. Also consistent with the 1995 results, the incoming meridional moisture flux at the southern boundary also decreases in TEST4 compared to TEST1, with a slightly greater decrease in the W. The differences in other flux components, particularly in 1997, are not as small as in 1995, though they are still significantly smaller than the changes in W-to-E zonal flux.

Figure 14.

Atmospheric water budget diagram for 1996: (top left) TEST1 and (bottom left) TEST4. Atmospheric water budget diagram for 1997: (top right) TEST1 and (bottom right) TEST4. The convention is the same as in Figure 8.

[56] Similarly, Figure 15 shows the water budget for TEST2 and TEST5, also for both 1996 and 1997. Again, these results are consistent with those for 1995. The addition of initial soil water in the west to TEST5 increases evaporation only slightly, with correspondingly small changes in zonal flux and precipitation.

Figure 15.

Atmospheric water budget diagram for 1996: (top left) TEST2 and (bottom left) TEST5. Atmospheric water budget diagram for 1997: (top right) TEST2 and (bottom right) TEST5. The convention is the same as in Figure 8.

3.2.5. Extension to a Longer Simulation Period

[57] Given that we have an improved understanding of the response of the large-scale dynamics to initial soil moisture distributions during a particular month, July, we may ask if this behavior holds during any other part of the warm season when the large scale flow and regional convective regimes can be different. As described in section 2.3, we use two 5-month simulations to address this question. Figure 16 shows the water budget calculated separately for the western and eastern parts of the analysis domain for TEST1 and TEST4 experiments for May 1995 (but initialized using the July 1995 TEST1 and TEST4 soil moistures). Changes in atmospheric moisture transport appear to be playing the same role during May as in July. Enhanced evaporation occurring in TEST4 compared to TEST1 (an increase of 37%) translates into a 12% increase in W box precipitation, accompanied by a significant increase in moisture transport into E box. The increase in moisture transport acts to redistribute the water across the domain such that by the end of May, the TEST4 E-box evaporation is less than 2% lower than in TEST1, and the precipitation greater by 0.29 mm/day (5%).

Figure 16.

Atmospheric water budget diagram, for (top) TEST1 and (bottom) TEST4 for May 1995. The convention is the same as in Figure 8.

[58] Unlike the July (monthly) simulations, however, the incoming meridional moisture flux at the southern boundary changes very little between the simulations (and actually increases slightly in the W box in TEST4 compared to TEST1). This may be due to the generally more meridional orientation of the soil moisture gradients by the end of May (as shown below) compared to the July only runs. However, as these meridional flux changes for May are negligible compared to those found previously for July, the net impact of the atmospheric dynamics appears to be the same: i.e., a negative feedback on the initial soil moisture anomaly pattern.

[59] Figure 17 shows the differences in RAMS simulated total and convective precipitation between TEST4 and TEST1 averaged over the months of May (Figure 17a), June (Figure 17b), and July (Figure 17c). Most of the relatively large differences that are apparent at the end of the first month have been eliminated after 3 months. This gives us some idea of the persistence of the impact of this particular soil moisture anomaly pattern on regional hydrometeorology. We emphasize that this specific timescale holds for the dry regime only. Because of the generally weaker response under wetter initial soil moisture conditions, we might expect the soil moisture anomaly pattern in the TEST3-and TEST5-type runs to persist longer (recall Figure 12), thus allowing more time to influence the atmosphere, but at the same time producing a weaker atmospheric response.

Figure 17.

RAMS (left) simulated total precipitation (mm hr−1) difference and (right) convective precipitation difference between TEST1 and TEST4 averaged over the months of (a, b) May, (c, d) June, and (e, f) July 1995.

3.2.6. Sensitivity to a Larger Model Domain

[60] Last, we briefly mention the results of the simulations using a much larger model domain. An important caveat is that, because of this larger area, the perturbed soil moisture values also extend beyond the boundaries of our original sensitivity runs, thereby changing the overall initial soil moisture pattern in both runs. We note also that the larger domain chosen has an approximately 2.5-times greater zonal extent than that of the smaller domain, and an intermediate choice between the two might yield results that are closer. In spite of this, we see some qualitative similarities in behavior between the results for the larger and smaller domains. Figure 18 shows the atmospheric water budget diagrams for the large-domain TEST1 and TEST4 experiments. We emphasize that the boundaries of the W and E boxes used to calculate the budget terms shown in Figure 18 are identical to those used in the previous budget diagrams. In other words, the budget domain is embedded within the much larger simulation domain. This means that a few of the flux components that remained relatively constant in our earlier budget calculations, for example the incoming zonal flux at the western boundary of the W box, now change significantly in response to the change in soil moisture outside the box. This renders the interpretation of the atmospheric moisture budget much less clear-cut than for the smaller-domain simulations. The differences in monthly mean precipitation and evaporation in each box compared to the smaller-domain runs (e.g., compare with Figure 8) also reflect both the different soil moisture boundary conditions from areas outside the W and E boxes, as well as the somewhat different amount of initial soil moisture redistribution between the heterogeneous and homogeneous cases. Nevertheless, the increases in W-box evaporation, W-box precipitation, W-to-E zonal moisture transport, and decrease in W-box incoming meridional moisture transport are all qualitatively similar to the results obtained with the smaller domain. This provides some confidence that certain features of the atmospheric response to our changes in soil moisture distribution might consistently appear, given reasonable choices for the simulation domain.

Figure 18.

Atmospheric water budget diagram for (top) TEST1 and (bottom) TEST4, using the Kain-Fritsch convective scheme with a larger domain size. The convention is the same as in Figure 8.

3.3. Impact of Changing Convective Parameterization

[61] Because most of the precipitation produced in our simulations is convective, it is important to ascertain how sensitive our results are to changing how convection is parameterized in the model. This may help give us insight into the robustness of our conclusions. We emphasize that we have made no attempt to tune either scheme in order to optimize agreement with observations or with each other.

[62] Figure 19 shows the differences in RAMS simulated total and convective precipitation between the July 1995 TEST4 and TEST1 (Figure 19a), TEST3 and CONTROL (Figure 19b), and TEST5 and TEST2 (Figure 19c) experiments using the Kuo convective scheme. There is some broad agreement with the results obtained using the Kain-Fritsch scheme (e.g., compare with Figure 7). For example, the largest precipitation differences occur for the dry simulations, with progressively smaller differences occurring under wetter soil moisture conditions (e.g., TEST2 and TEST5 are almost the same; bottom panels of Figure 19). In addition, similar to what is shown in Figure 7, most of the monthly mean precipitation differences (both positive and negative) occur in the eastern half of the domain. Unlike with the Kain-Fritsch scheme, however, the absolute differences are significantly smaller, and not as large a portion of these differences is due to convectively generated precipitation. This is generally consistent with the lower precipitation totals, and smaller amount of convection in general, simulated with the Kuo scheme compared to the Kain-Fritsch scheme.

Figure 19.

RAMS simulated total precipitation (mm hr−1) difference (left plot) and convective precipitation difference (right plot): (a, b) TEST1 minus TEST4, (c, d) CONTROL minus TEST3, and (e, f) TEST2 minus TEST5, averaged for July 1995 using the Kuo convective scheme.

[63] The water budget for the July 1995 Kuo dry experiment is shown in Figure 20. In agreement with the results obtained using the Kain-Fritsch scheme, the primary mechanism responsible for driving the variation in precipitation patterns over this area seems to be the zonal moisture flux from the western to the eastern half of the simulation domain. The extra soil moisture initially placed in the west enhances evaporation there, and some of this is responsible for an 8% increase in W-box precipitation. Much of the rest is transported to the E, raising precipitation there as well. There are some differences in the details of the other flux components compared to the Kain-Fritsch runs, but qualitatively the major features are similar.

Figure 20.

Atmospheric water budget diagram for (top) TEST1 and (bottom) TEST4, using the Kuo convective scheme. The convention is the same as in Figure 8.

[64] An interesting point is that, while, for example, the TEST1–TEST4 differences in W-box evaporation and W-to-E zonal moisture transport are similar for both the Kuo and Kain-Fritsch simulations (compare Figures 8 and 20), the large increase in E-box precipitation and recovery of E-box evaporation that occurs in the Kain-Fritsch runs does not occur in the Kuo runs. In other words, while similar amounts of the initially enhanced soil moisture in the W box are moved by the atmospheric dynamics into and through the E box, less of it precipitates out there when the Kuo scheme is used; that is, the precipitation efficiency is reduced in the Kuo compared to the Kain-Fritsch TEST4 run. This is again more or less consistent with the generally lower amounts of total and convective rainfall in the Kuo simulations. A detailed deconstruction of the aspects of each scheme that might be responsible for these differences in behavior is beyond the scope of this study (see Pan et al. [1996] for such a discussion between the Kuo and Grell schemes). However, there are a few points worth noting. One major difference between the two is that Kain-Fritsch contains a parameterization of convective downdrafts, while Kuo does not [Pielke, 2002]. This is generally thought to allow for better simulation of mesoscale convective processes. Since a large part of the precipitation in this region during this time of year occurs as part of organized mesoscale systems, the Kain-Fritsch scheme may provide a more faithful representation of the interaction between convection and its large-scale environment. Consistent with this reasoning, and as mentioned in section 3.1, the Kain-Fritsch simulations agree much better with observations than the Kuo simulations. Therefore it might be reasonable to assume that the response of the system to the imposed soil moisture anomaly pattern is more realistically captured with the Kain-Fritsch runs.

[65] These results underscore the high sensitivity of model-generated precipitation to choice of convective scheme, and therefore the need for close attention to the role of convective parameterization in the outcome of numerical experiments. Owing to coupling between the various components of the system, differences in precipitation introduced by changing convective schemes have the potential to propagate, for example, by forcing further changes in evaporation and the soil moisture field. Nevertheless, the fact that we find qualitative agreement between the two schemes in several key aspects of the response to the initial soil moisture perturbation increases our confidence in the validity of the overall picture we are attempting to present.

4. Conclusions

[66] We simulate July precipitation for each of 3 years, 1995–1997, with six different initial soil moisture patterns: three (control, dry, and wet) with a realistic (observationally based) spatial distribution, and three (control, dry, and wet) with a horizontally homogeneous distribution. The realistic soil moisture patterns reflect the climatology of the region: relatively drier in the western half of the domain as compared to the eastern half. Therefore the homogeneous runs can be thought of as starting from a “wet west/dry east” soil moisture anomaly pattern. In addition, we investigate the sensitivity to choice of convective parameterization by repeating all experiments for two different schemes: Kain-Fritsch and Kuo.

[67] Even though the initial (climatological) heterogeneity in soil moisture is smoothed out in the homogeneous runs, the impact of increasing soil moisture in the west and decreasing soil moisture in the east is to enhance total domain averaged precipitation in both the west and east. Although this behavior is seen in most of our simulations, it is most pronounced in the dry regime, consistent with earlier work. We examine the various terms in the atmospheric water budget to explain these results. Comparing the homogeneous to the heterogeneous (realistic) runs, we find the following general atmospheric response to our particular imposed soil moisture anomaly pattern (most markedly for the dry experiment pair, TEST1 and TEST4). In the western half of the model domain, higher initial soil moisture in the homogeneous runs leads to enhanced evaporation, which in turn increases the boundary layer moist static energy and the overall vertical instability. Enhanced outgoing zonal moisture transport toward the eastern half, combined with diminished incoming meridional moisture transport at the southern boundary (due primarily to a decrease in the strength of the low-level meridional jet), result in an overall increase in atmospheric moisture divergence here. This increase in moisture divergence counteracts some of the impact of the stability decrease, with the result being only a rather small increase in W-box convective rainfall. By contrast, in the east, the enhanced zonal moisture transport into the box results in an overall decrease in atmospheric moisture divergence (i.e., increase in convergence). This produces an increase in convective rainfall, which, averaged over the month, replenishes the soil moisture, significantly reducing its initial deficit, while allowing evaporation and the moist static energy profile to recover to the levels present in the runs with heterogeneous initial soil moisture. In other words, though there is greater evaporation on the wetter west of the homogeneous runs, a significant fraction of this additional water is transported into the east by the zonal flow. This decreases significantly the extra water available for precipitation in the west, but provides a large additional source of moisture for the east, thus compensating for the loss in water due to the initial negative soil moisture anomaly there.

[68] Furthermore, precipitation recycling seems to play a significant role throughout the domain. For example, at least in the dry regime, enhanced precipitation in the west in the homogeneous initial soil moisture cases can lead to subsequent pulses of enhanced evaporation, west-to-east moisture transport, and east precipitation. In other words, local soil moisture-rainfall feedbacks work with the atmospheric dynamics to help restore depleted rainfall and soil moisture levels in the east. In summary, the large-scale atmospheric dynamics, primarily the west-to-east zonal moisture transport, acts to restore the imposed “wet west/dry east” soil moisture anomaly back toward the climatological distribution, thus acting as a negative feedback mechanism.

[69] Again, we emphasize that this response is greatest in the dry regime and nonlinearly weaker (even slightly reversed in some aspects) with increasing soil moisture. In addition, these findings are specific to the particular soil moisture anomaly pattern we have chosen. A different (perhaps more realistic) anomaly distribution could be expected to yield different results. The longer simulations we performed suggest that, in the dry regime, the persistence of this anomaly pattern is roughly 3 months.

[70] While there are some important sensitivities, significant features of these results are robust with changes in simulation month (May versus July), domain size, and convective scheme (Kuo versus Kain-Fritsch). For the comparison between convective schemes, we speculate that the inclusion of a specific representation of convective downdrafts in the Kain-Fritsch scheme may be responsible for the differences in results and the overall improved performance of this scheme.

[71] Finally, we can say something about the relative impacts on precipitation of the specific changes in the initial soil moisture spatial distribution and amount. In most cases the greatest absolute change in precipitation occurs when comparing the driest to the wettest simulations, for example, TEST1 to TEST2 or TEST4 to TEST5. Only modest changes occur between the CONTROL simulations and the wet cases (TEST2 and TEST5). However, significant changes occur at the dry end of the spectrum, for example, between TEST1 and CONTROL, in both the western and eastern halves of the model domain, for all years. For example, in July 1995, there is an increase of 28% in the western half and of 24% in the eastern half in CONTROL compared to TEST1. A similar comparison between heterogeneous and homogeneous initial soil moisture runs, i.e., between TEST1 and TEST4, yields a 17% and 14% increase for west and east, respectively. Therefore it is possible to conclude that the impact of changing soil moisture distribution is significant, at least for the cases considered here, compared to that of changing soil moisture amount. It should thus be considered as an additional factor in studies of land-atmosphere feedbacks, soil moisture persistence, and extended-range precipitation predictability.


[72] The authors appreciate the efforts of Chris Castro and Adriana Beltran for providing us with the Kain-Fritsch convective parameterization. This research was supported by NOAA GAPP grant NA16GP1618. The reanalysis data were obtained from NCAR, which is supported by NSF. The authors wish to thank two anonymous reviewers whose comments significantly improved the quality of the paper.