Observed and simulated soil moisture variability over the Lower Mississippi Delta Region
Georgy V. Mostovoy and Valentine G. Anantharaj
Geosystems Research Institute, Mississippi State University, MS 39762
Journal of Hydrometeorology
(In press)
Geosystems Research Institute, Mississippi State University
PO Box 9652, Mississippi State, MS, 39762, USA
E-mail:
Abstract
In order to better understand error and spatial variability sources of soil moisture simulated with land surface models, observed and simulated values of soil moisture (offline simulations with the Noah land surface model with four soil layers and approximately 1x1 km² horizontal resolution were used) were compared. This comparison between observed and modeled daily values of soil moisture was performed over the Lower Mississippi Delta region during summer/fall months spanning years 2004 to 2006. The Noah simulations covered 2.5º×2.5º latitude-longitude domain and were forced by the NLDAS atmospheric forcing fields. Hourly soil moisture measurements and other data, including local meteorological and soil physical properties data from twelve SCAN sites, were used. It was shown that both the observed and simulated level of soil moisture depend critically on the specified/sampled soil texture. Soil types with a relatively high observed clay content (more than 50% of weight) retain more water due to the low water diffusivity in comparison with silty/sandy soils having 20% or less of clay, provided that other conditions are the same. This fact is in agreement with previous studies and implies a strong soil texture control (through related hydraulic parameters, e.g. Richter et al. 2004; Braun and Schädler 2005) on the accuracy of simulated soil moisture. Sensitivity tests using the Noah model were performed to assess the impact of using the hydraulic parameters related to the site-specific soil texture on soil moisture quality. Indeed, at some SCAN sites, the errors (root mean square difference and bias) were reduced. Simulated soil moisture showed at least 50% reduction when the site-specific soil texture was used in Noah simulations instead of that derived from the STATSGO data. The most significant improvement of simulated soil moisture was observed within the top 0-10 cm layer where an original positive bias (an excessive wetness) was almost eliminated. At the same time, excessive dryness (negative soil moisture bias), which was dominant within second and third model layers was also reduced. These improvements are expected to be valid at sites/regions with low (less than 0.3) vegetation fraction.
1. Introduction
Soil moisture (SM) is a key variable of the land-atmosphere system, which along with other environmental variables controls the rate of evaporation from the surface and the partitioning of the moisture and energy fluxes across land-atmosphere interface. Hence, SM represents an important input for various environmental and hydrometeorological models. For that reason, accurate specification and prediction of the SM fields at different spatial scales (ranging from local to continental) is a practical need for hydrometeorological applications. The spatial density of available in-situ soil moisture measurements is not adequate to derive reliable area-averaged SM estimates in the range from 1 km to 10 km grid resolution used for initialization of mesoscale atmospheric models at regional or continental scales. Therefore, Land Surface Models (LSM) are commonly used to simulate initial soil moisture states (Crawford et al. 2000; Chen and Dudhia 2001).
The first LSMs (also referred to as land surface parameterization schemes) were originated from rather simple but effective parameterization schemes of the land surface and vegetation processes (e.g. Deardorff 1978). They were developed to describe the lower boundary condition for the atmosphere in weather prediction and climate models. Based on the complexity of the physical processes, the various LSMs can be classified into single soil layer, force-restore (generally having two soil layers, e.g. Bosilovich and Sun, 1995), or multilayer diffusion type models, as suggested by Shao and Henderson-Sellers (1996).
In virtually all current multilayer LSMs of moderate complexity, the vertical flow of water within the soil layer of a finite depth is described as a diffusion process (e.g. Capehart and Carlson 1994; Vitebro and Beljaars 1995; Liang et al. 1996; Smirnova et al. 1997; Irannejad and Shao 1998; Walko et al. 2000; Chen and Dudhia 2001). The vertical distribution of soil moisture is obtained by a numerical integration of the one-dimensional diffusion-type equation, known as Richard’s equation in soil (Hillel 2004) and hydrology (Smith 2002) sciences, with specified water diffusivity (Dw) and hydraulic conductivity (Kw) coefficients and given boundary conditions at upper and lower boundaries of the soil column. These coefficients control the rate of the flow of water in the soils, considered as a vertically homogeneous porous media (typically non-saturated with water), so that the coefficients Dw and Kw depend on the SM content and the soil texture, which in turn is characterized by the distribution of soil particles sizes. Twelve major soil classes represented by the USDA soil texture triangle (Hillel 2004) are typically used in LSM schemes to describe a natural continuum of soil types observed across the globe. A power law having few empirically-derived parameters is usually assumed to describe the relationship between Dw and Kw coefficients and the SM (Smith 2002). This manner of analytical representation of Dw and Kw with constant parameters related to a particular soil texture class provides a universal and efficient approach for modeling the SM at any geographical location where the soil type is known. This surrogate methodology is widely applied although the parameterization of soil hydraulic properties based on a dominant soil texture observed within the model gridbox suffers from obvious uncertainties. It is evident that one of the restrictions of this approach is related to the use of a limited number of soil classes to describe a continuum of soil hydraulic properties. It can be expected that soil texture variations within boundaries of a particular texture class, observed at the local/field scale, may explain a substantial part of simulated SM errors. On the other hand, Baker (1978), studying variability of soil hydraulic properties, reasoned that soils with different textures could be grouped together based on similarities of their hydraulic properties.
Ek and Cuenca (1994) studied the sensitivity of surface fluxes to sample variations of the hydraulic parameters having a range of one standard deviation within a particular soil texture class (sandy loam was considered). The largest impact of these variations on surface fluxes was observed over the bare soil for dry and moderate soil moisture states. Cuenca et al. (1996) investigated the sensitivity of the surface layer temperature and fluxes with a single-column atmospheric model integrated for six hours to different shapes of functions describing soil hydraulic properties. Predictions using functions suggested by Clapp and Hornberger (1978) were compared with those simulated with hydraulic functions proposed by van Genuchten (1980), and results of this comparison showed moderate sensitivity of atmospheric surface layer variables to changes in hydraulic functions.
The research mentioned above and similar sensitivity and validation studies with LSMs were focused on assessment of the atmospheric surface layer response (Pan and Mahrt 1987; Sridhar et al. 2002; Marshall et al. 2003) using in-situ measurements. Point SM simulations performed over a catchment in southeast Australia using the ECMWF land surface scheme (Richter et al. 2004) revealed that SM bias of this scheme could be reduced substantially by using point-specific soil hydraulic parameters, including wilting point, field capacity, and saturated soil conductivity. Simulated SM values may differ from the observed for a number of reasons. But one of these reasons may be due to the differences between hydraulic properties of the actual soil texture at the point location and those assigned in the LSM to the corresponding dominant texture class within the grid cell or element. The importance of soil hydraulic properties adjustment for better accuracy of simulated SM were demonstrated by Shao and Irannejad (1999) and Braun and Schädler (2005). These results imply that observed SM biases are often modulated by the uncertainties in the selected soil hydraulic parameters (Richter et al. 2004).
Only a few studies have been devoted toward the direct validation of simulated SM (Crawford et al. 2000; Robock et al. 2003; Richter et al. 2004). Little is known about the impact of hydraulic properties on spatial and temporal features of simulated SM although it is well established that soil porosity is an important factor, which controls the accuracy of the simulated values (Tischler et al. 2007) and spatial variations of observed (Yoo et al. 1998) soil moisture. Therefore, the main objective of this study is to examine the impact of the differences between the locally-sampled and the spatially-aggregated soil texture prescribed from the Soil Survey data on the accuracy of simulated soil moisture characteristics. It is also anticipated that this study will also be useful in the interpretation of observed spatial and temporal variability of SM fields; and thereby contributing toward a better understanding of the role of various factors influencing the spatial distribution of SM at a regional scale.
Hourly SM observations and physical soil property measurements from twelve Soil Climate Analysis Network (SCAN) sites located mainly over the Lower Mississippi Delta Region (SCAN 2007), or simply the Mississippi Delta, spanning the period from January 2004 to December 2006 were used in this study. The Mississippi Delta was chosen primarily because of the higher spatial density of SCAN sites located in the region, across the states of both Arkansas and Mississippi. Retrospective SM simulations were performed using the Noah LSM (Ek et al. 2003) for these sites to assess the impact of using local soil hydraulic properties related to a site-specific soil texture on the quality of simulated SM. The Noah model, having a moderate level of complexity, is one of the widely used LSMs by the hydrometeorological community.
The rest of this paper is organized as follows. Section two describes SM and other relevant data available from in-situ measurements at the SCAN sites in our domain. The salient aspects of SM estimation using the Noah model and validation of simulated SM against SCAN observations are presented in Section 3. Analysis of observed SM dynamics during dry-down periods is described in Section 4. Performance of the Noah model with different specifications of soil hydraulic properties is discussed in Section 5. Section 6 provides summary of the results and conclusions. It should be noted that a volumetric fraction of the water content expressed in percents [(water vol.)×100%/(soil vol.)] will be used as measurement units for the SM in this study.
2. Data description
Soil moisture measurements from twelve SCAN sites (SCAN 2007) located across the Lower Mississippi Delta Region (including five in the state of Arkansas and seven in the state of Mississippi) were used for analysis and comparison with Noah LSM simulations. The geographical distribution of these sites within the study area is depicted in Fig. 1. The SM volumetric fraction is retrieved from water dielectric constant measured by the Hydra Probe II SM sensor (Stevens 2007) every hour at depths of 5 cm, 10 cm, 20 cm, 51 cm, and 102 cm. These depth levels and the layer thicknesses are depicted in Fig. 2b. This current study covers the 3-year period spanning from January 2004 to December 2006.
In addition to soil moisture measurements, the USDA SCAN network also provides site-specific data about physical parameters of the soils including texture, particle size distribution, water retention, and others. These data represent mean values for soil layers with a different thickness in the range from about 10 cm to 30 cm, which depends on the location and depth.
3. Soil moisture simulations with Noah LSM
a. Soil moisture prediction
In the Noah model, the SM volumetric content (θ) is predicted by numerical integration of the following diffusion-type equation (Chen and Dudhia 2001):
, (1)
where S is a SM sink due to water uptake by plant roots; and z is a vertical coordinate oriented downward from the surface. Equation (1) describes a vertical flow of a fluid with constant density through soil pores within a rigid, isotropic, and homogeneous soil column (Brutsaert 2006) with Po and E specified at the upper boundary of this column. Po and E represent fluxes of precipitation (P) minus runoff (R) and water vapor (total evaporation) at the surface, respectively. They are typically expressed in kg/(m²s) units, or in equivalent units of mm/s. The water density in (1) is assumed to be 1 g/cm³. According to Eq. (1), both the diffusion flux and the gravity flux Kw act to redistribute the water within the soil column. It is assumed in (1) that the force of gravity is directed along the z-axis, so in case of rough terrain with a moderate slope, the term involving Kw should be multiplied by cos (δ), where δ is the terrain slope angle (Capehart and Carlson 1994).
A standard configuration of the Noah model with four soil layers (Chen and Dudhia 2001), with progressively increasing thickness from the top/surface layer, was used in the current study. Figure 2 shows the location and the thickness of these layers. The three top layers in the model may contain plant roots (only in these layers S ≠ 0), and the lowest 1-m layer serves as a water reservoir with a free gravitational drainage condition at its bottom. Soil water diffusion flux () is assumed to be zero across the lower boundary of the lowest layer. It should be noted that before a numerical discretization of Eq. (1), it is integrated vertically within each layer, leading to the equation for layer-averaged tendencies of SM content. Further in the text, layer-averaged values of SM will be denoted by an over bar symbol. To evaluate a vertical derivative of θ, a linear z-dependence for values is usually adopted (Mahrt and Pan 1984).
The total evaporation rate E in the Noah model is described as a sum of (a) direct evaporation from the top soil layer, (b) evaporation of the water intercepted by the canopy, and (c) evapotranspiration from the canopy. The first two components are linearly proportional to the potential evaporation (Ep) and (1-f), where f is the green vegetation fraction, and the evapotranspiration is proportional to fEp. The Ep term, representing an atmospheric evaporative demand, is evaluated from the surface energy balance using the approach suggested by Penman (e.g. Mahrt and Ek 1984). Only a fraction β =/of the rate (1- f)Ep is available for the direct evaporation from the ground surface in the model, where θw and θf are soil moisture content at wilting point and field capacity, respectively. Their values depend on the soil texture class. The β coefficient, aka the soil moisture availability (e.g. Smirnova et al. 1997), provides a simple soil control on the evaporation rate associated with the atmospheric demand and was initially proposed by Deardorff (1978). The evapotranspiration may vary linearly between (1- f)Ep (when the soil moisture content () within the top model layer is equal to θf ) and zero if the surface layer soil moisture content is θw . In other words, evaporation from the soil stops if = θw.