MAPPING SOIL MOISTURE IN THE CENTRAL EBRO RIVER VALLEY ( NE SPAIN ) WITH LANDSAT AND NOAA SATELLITE IMAGERY: A COMPARISON WITH METEOROLOGICAL DATA.
Sergio M. Vicente Serrano*, Xavier Pons Fernández** and José Ma Cuadrat Prats*
* Department of Geography. University of Zaragoza. C/ Pedro Cerbuna 12. Zaragoza. 50009. Spain. svicen@posta.u n izar.es,
** Department of Geography and Centre for Ecological Research and Forestry Applications (CREAF), Autonomous University of Barcelona, Bellaterra 08193, Spain.
ABSTRACT:
This paper analyses the spatial distribution of soil moisture using remote sensing: NOAA-AVHRR and Landsat-ETM+ images. The study was carried out in the central Ebro river valley (NE Spain), and examines the spatial relationships between the distribution of soil moisture and several meteorological and geographical variables following a long, intense dry period (winter 2000). Soil moisture estimates were obtained using thermal, visible and near-infrared data and applying the “triangle method”, which describes relationships between surface temperature (TS) and fractional vegetation cover (FR). Significant differences were found between the soil moisture estimates obtained using AVHRR and ETM+ sensors. However, in both cases the spatial distribution of soil moisture was largely accounted for by meteorological variables.
KEY-WORDS: Soil moisture, Spatial scales, Climate-soil moisture relationships, Surface temperature, NDVI, Ebro river valley, Spain.
1- INTRODUCTION:
Natural and agricultural dryland ecosystems depend heavily on soil moisture. In semi-arid regions, where evapotranspiration rates are high, plant growth is limited by low levels of soil moisture (Downing, 1996; Müller-Edzards et al., 1997; Alexandrov and Hoogenboom, 2000; Sheperd et al., 2002). In these areas, knowledge of the spatial distribution of soil moisture is of considerable importance for hydrological, agricultural, economic and social planning.
In areas marked by high climatic variability, soil moisture and vegetation growth are determined by the amount and spatial distribution of precipitation. In periods of abundant rainfall, soil moisture does not vary greatly thereby guaranteeing the normal growth of natural vegetation and crops. Under these conditions, the topographical, edaphic and lithological characteristics have little influence (Western et al., 1999). However, in dry periods, soil moisture becomes a highly limited resource, and this moisture is heterogeneously distributed according to specific meteorological events, the topography, soil type, lithology, land uses and green cover (Western et al., 1999). Therefore, knowledge of the spatial interrelations between soil moisture, climatic factors and the geographical characteristics of the environment is of great applied interest, since in dry periods water availability is a determining factor, and crop success is more likely in those areas that are able to store water for longer periods.
Remote sensing techniques have been extensively used for the analysis of soil moisture and plant water availability. Estimates of soil moisture using satellite data have been conducted using various methods (Tucker, 1980; Crist and Cicone, 1984; Levit et al., 1990; Seguin et al., 1991; Nemani et al., 1993; Dupigny-Giroux and Lewis, 1999; etc) and can lead to significant time and cost savings in environmental and agricultural management. Remote sensing allows continuous estimates to be made, which represents a considerable improvement on the more limited measurements of ground data, and so ensures a virtually continuous temporal register thanks to the high temporal resolution of satellite images.
Soil moisture estimates obtained by remote sensing typically provide relative values in that they allow different areas on the image to be compared; they do not provide physical measurements. Basically, five types of method are available for estimating soil water content by means of remote sensing.
The first group uses microwave images (Wang et al., 1989; Schumugge et al., 1988 and 2002; Fran?ois, 2002) and the method is based upon the high absorption of the electromagnetic long wave radiation caused by water. The second is based on thermal inertia models, which make use of thermal, visible and infrared information when calculating soil moisture (Cracknell and Xue, 1996; Sobrino and Raissounni, 2000). The radiometric behavior of the different surfaces in the middle infrared has been widely used as a third method in estimating soil and vegetation moisture levels (Crist and Cicone, 1984; Musick and Pelletier, 1988; Levit et al., 1990), since the middle infrared reflectance of the soil and the green cover is highly conditioned by water content (Tucker, 1980). Various studies have estimated soil moisture using band combinations and a series of ground measurements to provide ground truth. Such empirical regression models constitute the fourth approach (Carlson, 1986; Levit et al., 1990; Cocero et al., 2000). Finally, the relationship between surface temperature (TS) and fractional vegetation cover (FR) has been widely used for soil moisture estimates (Nemani et al., 1993; Gillies and Carlson, 1995; Gillies et al., 1997; Dupigny-Giroux and Lewis, 1999). This last relationship, which is described in further detail below, was used to estimate soil moisture in this study.
However, although numerous studies have analyzed the spatial distribution of soil moisture using remote sensing, its relationship with climatic conditions and other geographical characteristics has not been always been examined in depth. Similarly, while many studies have compared the parameters recorded using a range of sensors at several spatial scales (mostly NDVI), the relationship with soil moisture estimates made at various spatial platforms has been the subject of little analysis (Carlson et al., 1995; Moran et al., 2002). This issue is clearly important because together with the spectral, radiometrical and orbital satellite characteristics, the different spatial resolutions of each sensor must be taken into consideration since they condition the spatial analysis of soil moisture. Frequently, in remote sensing the scale is the factor that is most restrictive. Indeed, processes seen to be operating at certain scales cannot be analyzed at others (Foody and Curran, 1994), while certain studies performed at a given scale cannot be conducted at another (Woodcock and Strahler, 1987). In the case of soil moisture analysis, soil and vegetation moisture usually present a marked spatial variability, though this might vary from scale to scale. Foody (1991) reported significant changes in crop moisture in relation with microtopography; this indicates that is more difficult to find significant soil moisture patterns at the small scale (Choudhury, 1991).
This paper analyzes the spatial distribution of soil moisture estimated indirectly using thermal, visible and infrared data from satellite images, employing a method similar to that adopted elsewhere (Nemani et al., 1993; Carlson et al., 1994; Lambin and Ehrlich, 1996; Sandholt et al., 2002) but revealing some marked differences between the sensors and in the role ascribed to certain meteorological and geographical variables. The area selected for the analysis was the central Ebro river valley (NE Spain), a semi-arid region of considerable climatic, lithological, edaphic and landscape diversity.
This paper has four objectives:
1- To determine the spatial distribution of soil moisture using satellite images. The analysis was conducted on 17 March 2000 following a period of severe drought in which scarce rainfall presented an uneven and anomalous spatial distribution.
2- To analyze the influence of “static” environmental variables (lithology, soil type, land use and topography) on the spatial distribution of soil moisture.
3- To determine the influence of meteorological factors (precipitation and temperature) prior to image acquisition on the spatial distribution of soil moisture, and to establish whether the severe drought significantly affected the spatial distribution of the soil moisture predicted by remote sensing.
4- To identify whether there are any differences in the climate-soil moisture relations obtained with the two satellites (NOAA and Landsat).
2- STUDY AREA
The location of the study area is shown in Figure 1. This section of the Ebro river basin is an excellent example of a relatively flat topographical area in which the climatic elements present a considerable spatial complexity. The landscape is dominated by horizontal structural platforms that overlie tertiary deposits, with altitudes below 800 m. Terraces and quaternary glacis mark the boundary between alluvial basins and plain bottoms (Pellicer and Echeverría, 1990).
[Insert Figure 1 about here]
The study area is located inside the general circulation of the Temperate Zone, close to the subtropical domain. The relief features isolate the valley from any maritime influences. The principal characteristic of the area is its aridity (Ascaso and Casals, 1981; Creus, 2001). Pluviometric variability is high. In some years the precipitation greatly exceeds (e.g. 646 mm in Zaragoza in 1959) the average (Zaragoza: 322 mm), whereas in other years the study area receives almost half that quantity (e.g. 182 mm in 1995). Furthermore, the lack of rain and the uncertainty of rainfall events combine with a high potential evaporation (1100 mm in the center of the valley) (Martínez-Cob et al., 1997).
The vegetation of the area is thermally influenced steppe (Suárez et al., 1992), determined largely by the lithology, soil-type and, in particular, the aridity. Forests are scarce in the bottom of the valley due to human activity dating back centuries; only some small forests of Juniperus thurifera, Quercus ilex and Pinus halepensis remain on the slopes of the tabular relief. The most common land use is dryland agriculture (wheat and barley). In this agricultural system the climate plays an especially important role as harvests are strongly conditioned by the rainfall, so financial losses can be severe in dry years.
3- METHODS
3.1- PRE-PROCESSING
Two satellite images (Landsat7-ETM+ and NOAA14-AVHRR), taken on 17 March 2000, were used. There was a five-hour lag between the two recordings (Landsat passed at 10:35 GMT and NOAA14 at 15:25 GMT). This date was selected because the Landsat image was the last to be taken before the spring rains. The Landsat image was orthorectified (Palà and Pons, 1995) using a Digital Elevation Model (DEM), while the NOAA-AVHRR image was geometrically corrected using a second-order polynomial adjustment (Richards, 1993). The images were radiometrically corrected in order to account for the atmospheric and solar illumination factors (Pons and Solé, 1994; Chávez, 1988). NDVI was obtained for both images (Tucker, 1979). The NDVI obtained from the Landsat image was resampled to 1 km, using a mean criterion, in order to match the AVHRR resolution.
Thermal bands were transformed to brightness temperatures (Markham and Barker, 1986). NOAA-AVHRR TS was obtained by means of a split-window algorithm (Sobrino and Raussoni, 2000). The Landsat thermal band was not corrected atmospherically given the fact that it is difficult to obtain reliable results in standard cases with only one thermal band without atmospheric data (Vidal et al., 1994; Coll et al., 1994). This does not mean that the use of the Landsat image is limited because the same atmospheric perturbation can be assumed for the whole area in the image. Finally, the effect of surface emissivity, in both images, was also corrected (Sobrino et al., 2001). The Landsat surface temperature was resampled to 1000m resolution using a mean criterion. The NDVI and the TS are shown in Figure 2.
[Insert Figure 2 about here]
Figure 3 shows the relationship between the TS and the NDVI obtained with NOAA-AVHRR and Landsat-ETM+ images at a cell size of 1000m. For NDVI images, the Pearson correlation coefficient was 0.74. The correlation for TS images was 0.71.
[Insert Figure 3 about here]
There are two principal restrictions to the effective comparison of both images. The TS map obtained by means of the AVHRR image is, in general, three degrees higher than that obtained using the ETM+ image. The absence of atmospheric correction in the latter case, and the different recording times of both satellites (Landsat-ETM+ at 10:35 GTM and NOAA-AVHRR at 15:25 GTM) might account for these differences. Gutman (1999) stresses the fact that the time lag between the passing of one satellite and the other causes major divergences in the TS measured. This has been observed in temporal series of TS images from NOAA-AVHRR satellites, where the observations were made progressively later than the launch because of the drift in the equator-crossing time (Price, 1991).
There are also significant differences between NDVI images, with higher values being obtained with the AVHRR image than with the ETM+ image. These differences may be produced by the different spectral configurations of both sensors: AVHRR red and near-infrared bands: 0.58-0.68mm and 0.72-1.10mm, respectively; ETM+ red and infrared bands: 0.63-0.69mm and 0.76-0.90mm, respectively. Nevertheless, a number of studies demonstrate the comparability of AVHRR and ETM+ NDVI images and point out that spectral configurations have little influence on results (Teillet et al., 1997). More important may be the effects of shifts in the photosynthetic vegetal activity during the day, caused by changes in incoming solar radiation, atmospheric moisture and air temperature (Justice et al., 1991). Differences between NDVI images obtained at different times of day have been analyzed in several studies (Holben et al., 1990; Gutman, 1991; Che and Price, 1992; Schultz and Halpert, 1995), showing that satellite orbit drift in the NOAA-AVHRR satellites results in considerable differences in the NDVI results between the moment of launching and the final satellite life because the time pass significantly changes (Teillet and Holben, 1994; Gutman and Ignatov, 1995). This implies discontinuities and inhomogeneities in temporal NDVI series (Kogan and Zhu, 2001) as well as problems in the calibration of images (Rao and Chen, 1996). These problems illustrate that certain limitations are encountered when comparing the information obtained by both satellites.
3.2- SOIL MOISTURE ESTIMATION
Soil moisture estimation using thermal data is based on the relationship between the water content of different surfaces and its temperature. The latent and sensible heat fluxes are conditioned by the surface water content (Eltahir, 1998). On unvegetated soils and in full vegetation areas, evaporation and transpiration increase as the water content rises. When soils are moist, the latent heat fluxes increase because of the greater absorption of water. This process causes sensible heat to decrease. In dry soils the process is the inverse of this. The radiative energy is not consumed in the evapo-transpiration process, and the sensible heat increases, raising the TS. In theory, this is a simple method of soil moisture estimation. However, the environment is particularly heterogeneous (soil, lithology, vegetation cover, topography), and it cannot be assumed that the coldest areas actually coincide with the areas of greatest soil moisture. Indeed, there are a number of elements that interfere in this relationship, the most significant being vegetation (Lambin and Ehrlich, 1996). The hydric vegetation conditions are hardly recognized in the visible region of the electromagnetic spectrum, but changes can be significant in the surface vegetation temperature (Seiler et al., 2000; Kogan, 2001). TS and fractional vegetation cover (FR) can provide information about vegetation and moisture conditions at the surface. Lambin and Ehrlich (1996) summarized the relationship between both variables in a theoretical space that indicates the moist limits and the soil moisture status in relation to different vegetation cover percentages (Figure 4).