1. Introduction

An outstanding problem in hydrometeorology is determining soil moisture content at high resolution over continental scales. Direct measurements of soil moisture are limited to specific sites and may be unrepresentative of larger areas. For this reason, there have been many efforts to develop remote sensing techniques to estimate surface moisture. Some success has been achieved in quantifying surface moisture where the vegetation cover is not too dense using passive microwave measurements (e.g., Wang et al. 1980; Wilke and McFarland 1986; Choudhury et al. 1987; Neale et al. 1990; Kustas et al. 1993). Most recently, Basist et al. (2001) have shown that a surface wetness product derived from the Special Sensor Microwave Imager (SSM/I) has a strong correspondence with the upper-layer soil moisture in cultivated areas of the world. A similar product has been used to identify regional areas following excessive rainfall from SSM/I measurements.

Remote thermal infrared measurements of the land surface (radiative temperatures or their temporal change) have been used since approximately 1980 to estimate the surface energy budget and soil moisture. The reported accuracy of such estimates has varied for different regimes and approaches (Hall et al. 1992; Brutsaert et al. 1993). In addition, remote measurements in the visible and near-infrared have been combined with those in the thermal infrared to estimate evapotranspiration and characteristics of the vegetation cover (Carlson et al. 1994). Diak et al. (1995) considered relationships of satellite and in situ observations to basic properties of the surface energy balance, including thermal inertia, Bowen ratio, and roughness height. The relationships were tested over much larger scales than previously considered. The relation of observed daytime rise in surface radiative temperature (DTs) versus the normalized difference vegetation index (NDVI) was similar to published results for surface radiative temperature measurements versus NDVI over smaller areas. Namely, the correlation between DTs and NDVI was high (low) in regions where the surface was dry (wet). In regions where NDVI values were small, there was a good correlation of DTs with an antecedent precipitation index (API). This suggested that some combination of NDVI and API could be used for a better description of the surface temperature and energy balance than using the NDVI measurements alone.

In this note, the relationship between DTs derived from the Geostationary Operational Environmental Satellites (GOES) sounder clear-sky radiative temperature and modeled soil moisture is presented. The motivation here is to provide an infrared (IR) satellite–based index for soil moisture, which has a higher resolution than possible with the microwave satellite data. An application of this index is to determine the amount of surface dryness relevant to the issuance of fire weather outlooks at the National Oceanic and Atmospheric Administration (NOAA) Storm Prediction Center. In section 2, the data and processing will be described. Comparisons of the GOES satellite data and modeled soil moisture appear in section 3. Conclusions are found in section 4.

2. Data and techniques

The IR satellite data used are from the GOES-8, -10, and -12 sounders. The characteristics of these instruments are similar to those described in Menzel et al. (1998). The horizontal resolution of the sounder is 10 km × 10 km at nadir and the time resolution is hourly. The surface radiative (or skin) temperature Ts is an output product (Hayden et al. 1996) of the nonlinear physical retrieval process described in Ma et al. (1999). It should be noted that variable surface emissivity is not fully accounted for in the retrieval process. This could lead to a relative error in Ts between wet and dry areas where surface emissivity is expected to vary.

An essential part of the processing includes the determination of clear-sky field of views (FOVs). The Ts is only available in clear-sky regions. This limits the frequency of usable satellite data for any given location, especially as compared to microwave data. For this reason, the use of GOES data may be most applicable to monitoring developing drought conditions when clouds are relatively sparse.

Individual retrievals have been routinely processed at the Cooperative Institute for Meteorological Satellite Studies (CIMSS) at the University of Wisconsin—Madison. The processing utilized 5 × 5 FOV for individual retrievals, yielding a 50 km × 50 km effective resolution at the subsatellite location. The use of single FOV retrievals has recently been implemented for GOES-10 and -12 data, thereby increasing the effective horizontal resolution of the soundings and reducing the overestimation of areas affected by clouds.

Using the Man Computer Interactive Data Access System (McIDAS) software system (Lazzara et al. 1999), individual images of retrieved Ts from the two GOES satellites are combined into a single image covering the entire continental United States (CONUS), where the closest satellite (with the smallest zenith angle) is used at any location. As a result of this constraint, the GOES-10 is used west of 105°W longitude for all years processed. To the east, GOES-8 was used until it was replaced by GOES-12 in April of 2003. Next, data from images are differenced to obtain the daytime rise in DTs at each clear location. As a simplification, the values of DTs were derived by subtracting values of Ts from images at 2100 UTC minus those at 1200 UTC. The results presented here appear to be relatively insensitive to varying the beginning and ending times by as much as 2–3 h. Monthly means of DTs were obtained by averaging the daily values for each clear observation. Satellite data have been processed from archived data for the summer months of June, July, and August 1999–2001. In addition, included in this study are data processed in near–real time from the summer of 2004.

Estimates of soil moisture were obtained from the National Oceanic and Atmospheric Administration (NOAA) Climate Prediction Center. These estimates are based on monthly precipitation and temperature data input into a one-layer water balance model on a 0.5° × 0.5° (50 km × 50 km) grid (Fan and van den Dool 2004). Precipitation measurements are from the cooperative surface network and other observations. In addition to soil moisture, the model diagnoses evaporation and runoff. The potential evaporation is based on the observed temperature. Parameters used to calculate runoff for all areas are based on observations from several small river basins in eastern Oklahoma. Soil-holding capacity is assumed to be 760 mm of water. Given a typical soil porosity of 0.47, this implies a soil column of 1.6-m depth. The soil moisture anomalies for a particular month are based on departures from the monthly mean for the 30-yr period of 1971–2000.

The accuracy of the modeled soil moisture may be limited by the validity of input precipitation measurements, and by complexities of runoff and infiltration associated with varying soil, terrain, and vegetation characteristics. These limitations are most acute in portions of the western United States where the density of rain gauges is most sparse and mountainous terrain exists.

To examine short-term changes in soil moisture, measurements of fractional soil moisture were obtained from Mesonet sites in Oklahoma during July 2004 (Brock et al. 1995). The fractional soil moisture is measured through various layers of soil including 5, 25, 60, and 75 cm (Schneider et al. 2003). These data were obtained from the Oklahoma Agricultural Weather project at the Oklahoma State University (more information available online at http://agweather.mesonet.org).

Maps of vegetation health for the periods of interest have been produced by Kogan (1997). They are based on the vegetation and temperature condition index (VT), which combines information on the chlorophyll and moisture content of vegetation and thermal conditions at the surface using the NDVI and infrared brightness temperature from the Advanced Very High Resolution Radiometer (AVHRR) instrument. The maps classify the vegetation conditions from favorable to stressed relative to mean conditions from 1985 to 2001 at weekly intervals for each NCDC climate division.

Biweekly NDVI data derived from the AVHRR were obtained from the United States Geological Survey Earth Resources Observation and Science (EROS) Data Center. These data are corrected for atmospheric absorption by water vapor and are smoothed using a weighted least squares approach to eliminate spuriously low values (Swets et al. 1999). They were remapped to common grids of the other datasets using the McIDAS software.

3. Analysis

a. Monthly means

Monthly mean values of DTs and modeled soil moisture (SM) are shown in Figs. 1 –4 for June of each year processed. Data are unavailable in areas of persistent cloud cover, and where NDVI values exceeded 0.75 (as discussed later in this section). An inverse relationship of DTs with soil moisture is evident in these figures. Most areas where soil moisture is greater than 300 mm are collocated with regions of DTs less than 3°C. For each year, the highest soil moisture (>300 mm) cover these geographical regions: 1) about the eastern half of the CONUS, 2) mountainous areas of the west, and 3) the northern Pacific coast. Only subtle differences in the general pattern are apparent among the years.

The analysis of DTs contains much more spatial detail than the soil moisture maps. In the west, the patterns of DTs appear to relate more closely to the location of specific topographical features such as mountain ranges and valleys than the SM analyzes. The data used in the SM analysis may lack resolution to resolve these features.

Additional insight into the relations between DTs and SM can be gained from performing a regression analysis between the two variables. For this analysis, CONUS was divided into two regions roughly separated by the Continental Divide. For simplicity, 105°W longitude was used as the boundary between regions. The area to the east has more uniform vegetation coverage and contains the majority of cultivated land, while the west contains more significant variation in altitude and annual precipitation (arid to very wet).

A scatterplot of DTs versus SM, typical of the other months analyzed, is shown in Fig. 5 for June 2001. The lines shown in these plots represent the best-fit linear regressions for this particular month. The results of the regressions for all the months appear in Table 1. The slope and intercept and standard deviation in the table are for Eq. (1):

View Expanded

The variables are more highly correlated for the eastern area. Correlations are typically −0.8 in the east and only −0.6 in the west. The magnitude of the correlations in 1999 are the smallest while all the other years are higher. The difference is most noticeable in the east where the negative correlation is 0.60–0.65 in 1999 and 0.80–0.90 in the other years. The reason for the relatively low correlation in 1999 is unclear. The standard error in estimating SM from DTs is generally 50–60 mm in the east and 70–80 mm in the west. In terms of mean SM, this uncertainty is roughly 10% in the east and 30% in the west. Using the linear regression [Eq. (1)], the explained variance in SM in the east ranges from 64% in June 1999 to 79% in 2004. In the west, the values range from 37% to 51%.

The correlations of DTs and SM with NDVI were also computed for the east and west areas. In general, the correlation of NDVI and DTs ranges from −0.7 to −0.5 in the east and west areas, respectively, while the correlation of NDVI and SM ranges from 0.7 to 0.5. The correlation of SM and DTs is compared for different ranges of NDVI in the eastern domain (Table 2). For areas in the east where NDVI is greater than 0.75, the correlation between SM and DTs is relatively weak (from −0.1 to −0.2). This suggests that Eq. (1) should only be applied in areas of modest NDVI. For this reason, data are only shown in the figures of DTs and SM where NDVI is less than 0.75.

Given the correlation of SM with NDVI in the east, the following regression equation was also evaluated:

View Expanded

Results of the regression using Eq. (2) in the eastern and western domain are given in Table 3. The standard errors in estimating SM and the explained variance in SM did not change significantly by accounting for both DTs and NDVI instead of only DTs. This can be explained by the relatively high correlation between DTs and NDVI within both east and west areas.

Given the relatively low correlations in the west and the extremely varied topography there, the local elevation was examined as a possible predictor of SM. The correlations of elevation to SM and DTs were near zero (absolute value less than 0.1). The amount of variance in SM explained by elevation was near zero.

4. Discussion

Many of the modeled soil moisture anomalies appear to coincide with anomalies of temperature rise of the opposite sign from satellite on a monthly time scale. However, a significant number of discrepancies exist, particularly in the western states. There are several possible causes, which may explain the lack of the expected relationship.

1. Areas with high frequency of clouds during a month may have an inadequate number of observations that are representative of the monthly mean.

2. Spatial separation between rain gauges may be insufficient to capture the local distribution of precipitation, and hence soil moisture, especially in some of the western states where rain gauge measurements are sparse.

3. In addition to soil moisture, the capacity of vegetation to generate evapotranspiration influences the amount of sensible heat flux and temperature rise as sensed from satellite. Root zone depth, vegetation type, health, and growth stage are also factors. For example, previous studies have indicated that the antecedent precipitation index (API) becomes uncorrelated with temperature rise (and sensible heat flux) in areas where green vegetation cover is high (e.g., Diak et al. 1995; Rabin et al. 2000). In this case, the DTs gives a better indication of vegetation state or soil moisture within the root zone active for that vegetation rather than soil moisture within a fixed depth as modeled in SM. In these situations, the values of DTs correspond much more closely to patterns of vegetation health than those of SM. Examples of this occurred in the southern states during June 1999 and 2001.

4. Another unique problem may occur with succulent-type vegetation cover found in arid climates. This vegetation has the capacity to regulate evapotranspiration to conserve moisture under dry conditions. In such cases, the temperature changes in the canopy may not reflect differences in available soil moisture. Other types of vegetation, such as crops in the Midwest and eastern United States, have more limited capacity to regulate moisture loss. The temperature of their canopy responds more directly to available soil moisture. The native vegetation adapted to dry regions of the west may account for decreased correlation between DTs and SM there. Further analysis including vegetation type is required to refine the relationships shown here on a regional basis.

The analyses of monthly mean and anomalous DTs may be carried out on shorter time scales, such as weekly or even daily periods, as demonstrated in section 3b. The occurrence of cloud cover will limit the usefulness of products over these time scales. (Microwave techniques have an advantage in such situations because of their ability to sense surface moisture through most clouds.) However, shorter time-scale analyses using IR data should still be helpful in monitoring the change in surface wetness during clear regimes. In principle, surface radiative temperatures can also be estimated from the current series of GOES imagers with about twice the resolution (4 km) of the sounder products examined in this paper. Hillger and Kidder (2005) have shown that the higher resolution of the imager data has a better chance of resolving clear areas between clouds. In addition, the higher resolution can be useful in identifying the effects of highly variable precipitation, especially on shorter time scales such as individual days.

These relationships between surface skin temperature and the NDVI will be further explored with the next-generation geostationary satellite data. The imager, working in conjunction with the hyperspectral sounder, will have spectral channels to allow the estimation of not only the surface skin temperature, and an improved surface emissivity estimate, but also the NDVI (Schmit et al. 2005; Knuteson et al. 2003). The imager, referred to as the Advanced Baseline Imager (ABI), will also be able to better monitor precipitation patterns and amounts. Data from the GOES-R system will allow a clearer picture of the relationship between the surface temperature and state of the vegetation.