Soil Hydrology Model (SHM)

Regional-scale atmospheric models require the initialization of SW (often referred to as soil moisture) at various layers in the soil over the horizontal model domain. Routine SW measurements over large areas (like the ones typically covered by regional-scale model domains) are unfeasible. Therefore, climatological estimates of SW have often been used to provide initial conditions to the regional models. Unfortunately, this approach may result in large errors during periods of excessive rainfall or drought. In the absence of observations, an indirect method is needed to determine SW fields from routine meteorological observations. We have taken two approaches to this: 1) remote sensing (described previously), which we refer to as "top-down", and 2) use of a soil hydrology model (SHM), which we refer to as "bottom up". The original objective of our study was to combine the two approaches, specifically by using the remote measurements to "nudge" the SHM. We now discuss the bottom-up method and the limitations in the nudging approach.

The SHM, developed by Capehart and Carlson (1994), uses a one-dimensional SW diffusion and gravitation scheme to determine the temporal evolution of SW as a function of depth. It is driven entirely by routine meteorological observations at the surface (i.e., the atmospheric forcing: air temperature, moisture and wind speed, as well as cloud cover and precipitation), information on soil and land cover characteristics (soil texture, land cover type, and fractional vegetation cover), and terrain slope. It includes modules for precipitation, interception of rainwater by the vegetation, ponding, infiltration, surface and subsurface runoff, subsurface diffusion, ET, and some rudimentary vegetation mechanics. The soil column depth, the vertical grid spacing within the soil, and the nature of the lower boundary (impermeable or saturated barrier) in the model are specified by the user.

The SHM is initialized with an arbitrary SW profile, usually chosen to be 50% of saturation. The model, driven by the atmospheric forcing, converges toward a common solution regardless of the initial SW value. This convergence takes place over a period ranging from a few weeks to a couple of months, depending on the precipitation amounts. We call this period of adjustment the "balancing" period.

Linkage of the SHM with the "Top-Down" Approach

Use of remote sensing to nudge the SW estimates provided by the SHM has encountered a theoretical block. The "triangle method" (Figure 13) implies the prevalence of very dry soil patches even during periods of relatively abundant rainfall. Conversely, there are also many pixels with high SW even in arid regions. Moreover, validation studies (Perry and Carlson 1988; Gillies et al., 1995b) indicate that the remotely derived SW values are always lower than those measured by in-situ methods or by microwave techniques, even when the correlation between the two data sets is high. Preliminary analyses by Humes and Kustas at USDA indicate that this is a real effect. The lack of correlation between thermal and microwave estimates of SW is, at first glance, perturbing but it does suggest some interesting insights into the manner in which soil dries. Our current modeling work suggests that soil crusting plays a role in the surface drying. When a surface dries out, the hydraulic conductivity decreases rapidly, thereby forming a seal for the water below. This allows the surface to become relatively warm despite relatively elevated SW at deeper layers.

Crusting occurs selectively when there is a range of soil properties. Figure 15 shows that even within a single soil class (silt loam) the normal variance of soil properties will result in parts of the surface drying out very rapidly following a rainstorm, while other areas remain wet for days. The significance of this finding is that surface radiant temperature measurements () are incapable of providing reliable information on the root zone SW. This is the reason why surface radiant temperatures may not be useful for subsurface hydrology, although it is clear that the surface moisture content is closely tied to the surface energy balance. Indeed, it is clear that Fr and Mo are the two dominant variables governing the surface energy balance. As such they are also important parameters for inclusion in the SHM and the MM, although the moisture availability would nevertheless pertain only to a surface layer.

Linkage of the SHM with the Mesoscale Model

In order to use the SHM to provide SW initial conditions for the land-surface component of the MM (i.e., BATS), the SHM must be applied to each grid cell over the entire MM domain. Smith et al. (1994) developed and tested this procedure (called the SHM system) which is illustrated in Figure 16. This system comprises three major components: the preprocessing programs, MIXER, and DRIVER. The preprocessing programs (some of which are part of the MM modeling system) use topographic information to provide gridded terrain heights for the domain of interest, and perform an objective analysis of surface meteorological data over the model grid, which results in the atmospheric forcing dataset. MIXER incorporates the atmospheric forcing dataset, the gridded soil- and land cover-characteristics datasets and the terrain-slope dataset, calculates the necessary radiation and ET variables required by the SHM, and produces the complete gridded dataset that is used by DRIVER. DRIVER then runs the SHM for each grid cell and produces the SW fields required by the MM. The SHM was modified so that it uses the same soil and land cover characteristics as BATS.

Smith et al. (1994) tested the SHM-MM linkage, where the SHM system was driven by a four-and-a-half-month historical meteorological dataset for three nested MM domains centered on the Mahantango Creek Watershed (MCW) in the SRB (Figure 17). Each domain contains a 60 by 60 array of grid cells with cell dimensions of 36-km, 12-km, and 4-km, respectively for the three domains. The MM was then initialized with the SHM-produced SW fields and a 12-hour simulation was made with the MM. A five-minute video has been prepared which demonstrates the time evolution of the SW fields over the 36-km SRBEX domain as generated by the SHM for that same period. This tape is available for demonstration (copy available on loan).

We are currently developing a real-time version of the SHM system, which will soon provide up-to-date SW fields to MM5 (Lakhtakia et al., 1994a,b). This will allow for real-time testing of the SHM-system capabilities when linked to MM5. Testing will include MM5 simulations, using the SHM-simulated SW fields as initial conditions, and comparison of the predicted afternoon temperature and specific humidity close to the surface with observations. We will also use the method of Mahfouf (1991) in which errors in the forecast variables are used to correct the SW used as MM5 initial conditions.

We have also begun (in collaboration with Dr. Jimy Dudhia of NCAR) to gather the information to create a multilayer SW climatology using the SHM on a rectangular 18-km grid covering the 48 conterminous US, for the period 1979 through 1995. Available in-situ measurements of SW will be used to assess the accuracy of the SHM output.

Land Cover and Soils Information

As discussed previously, surface characteristics are an important component of the modeling effort within SRBEX. The SHM and the MM both require the specification of several landcover parameters for each grid cell within the domain. This is achieved by specifying one of 18 land cover types, which, with the help of a lookup table, provide the land cover characteristics re quired by both models (e.g., roughness length, depth of rooting-zone soil layer, vegetation albedo, minimum stomatal resistance, etc.).

Until recently, the only high-resolution land cover dataset routinely used in the MM was the one archived at NCAR, which covers the entire globe at 10 arc-minute resolution (approximately 19-km). In order to initialize the SHM and the BATS module within the MM, this dataset (hereafter referred to as the NCAR dataset) has to be converted to the appropriate BATS land cover type. Figure 18 shows the land cover type distribution over the 4-km SRBEX domain using the NCAR dataset.

The Land Cover Characteristics Database developed at the United States Geological Survey EROS Data Center (Loveland et al., 1991) (hereafter referred to as the EDC database), has also been used in preliminary SRBEX studies. This database provides land cover types for the entire 48 conterminous United States at 1-km resolution. The original 167 land cover classes in the EDC database were reduced to the 18 BATS land cover types, and analyzed to the MM horizontal nested domains using a modal aggregation technique. Land cover information for the areas of Canada that are part of the 36-km and the 12-km MM domains was merged from the NCAR dataset. This resulted in seamless land cover maps for each of the domains. Figure 18 also shows the land cover type distribution over the 4-km domain using the EDC database. The modeling results using the EDC database were presented at the USGS-organized workshop: "Test and Evaluation of the USGS 1-km AVHRR Land-Cover Characteristics Data for the Conterminous United States: Results and Recommendations", that took place in April 1994 at the EROS Data Center, in Sioux Falls, South Dakota.

Apart from information on the land cover type, the SHM and the MM also require the specification of the soil texture (one of the 12 USDA soil texture classes) for each grid cell in the MM domain. The soil texture, with the help of a lookup table, provides the soil parameters required by both models (e.g., SW at saturation, minimum soil suction, saturated hydraulic conductivity, SW at the wilting point, etc.). Traditionally, the lack of reliable information on soil characteristics at the regional scale has been an impediment to SVAT improvement. In fact, until now the SHM and the BATS module within the MM have relied on soil texture derived from land cover. Figure 19 shows the soil texture distribution over the 4-km domain derived from the NCAR land cover and the EDC land cover.

The recent development of the State Soil Geographic Database (STATSGO) by the USDA Natural Resource Conservation Service (NRCS) shows potential for the delivery of much needed realistic soil information to the modeling community (Miller and Lakhtakia, 1994a, b). STATSGO has been developed for river basin, multi-state, state, and multi-county resource planning. The compiled soil maps are created with the USGS 1:250,000-scale topographic quadrangles as base maps. STATSGO contains information on a wide range of soil properties (e.g., texture, particle size distribution, available water capacity, and bulk density). Initial work with STATSGO has produced a soil texture dataset that is compatible with the lookup table approach of the models. Figure 19 shows the STATSGO-derived soil texture distribution over the 4-km MM domain.

Two simulations with the SHM, using the atmospheric forcing dataset from Smith et al. (1994), were performed to test the effect of soil texture information on the SHM results. While in both simulations the SHM used the EDC land cover information, for the first simulation it used the EDC soil texture information and for the second the STATSGO soil texture information. The last ten days of the period over which the SHM simulations were performed (9 July - 18 July 1990) coincide with the NASA Multisensor Airborne Campaign (MAC-HYDRO '90), which was coordinated by Dr. E. T. Engman of NASA Goddard Space Flight Center. This campaign took place over the Mahantango Creek Watershed, which is located in the geographical center of the SRBEX model domains. The campaign area contained 27 sites where SWC measurements were taken with neutron-probe and electromagnetic soil-moisture measuring devices. Instrument readings of SWC were integrated over 15-cm and 30-cm depths from the surface. Most of the probe sites were in a meadow environment, with some probes placed in corn fields as well. The SHM output from both simulations from the 4-km domain central grid cell are compared with the averaged field measurements for the upper 15-cm soil layer (Figure 20). The overall pattern of the SHM results from both simulations is consistent with the field measurements. However, the results from the second simulation show much better agreement with the observations (Figure 21).

Table of Contents Previous Section Next Section Download Report
Graphics and HTML text markup by Phil Kolb
Comments to: R. White, raw@essc.psu.edu