Evaluation of Regression Model, Kriging Method and Supervised Classification of LISS-III Sensor Data in Estimating Soil Salinity

Employing recent technological advances in surveying and mapping soil salinity is a step forward in controlling saline soils. The aim of this study was to map the topsoil salinity, the depth of 0-5 cm, using different methods within the environmental context of the area around Tashk & Bakhtegan Lake, with the area of 8062 ha, that in this region soil salinity appears to be a major threat to agriculture production. We used three different methods to produce soil salinity map and then compared the results with the soil salinity data that were measured on the ground. A set of 143 soil salinity sample, electrical conductivity of the water extracted from saturated past (ECe), was systematically sampled on a 750-m grid and was used to assess two mapping methods; regression models (RM) and ordinary kriging (OK). As a third method, supervised classification (Scl) of LISS-III sensor satellite images was employed. We used linear, power and exponential regression models for estimating of salinity values. In these regression models, digital numbers of the satellite images were set as independent variables and ECe values as dependent variable. In order to provide a prediction map of the soil, the salinity data were interpolated using the ordinary kriging method. In case of the satellite images, we classified the training pixels with maximum likelihood algorithm and then the land cover map was prepared. Our results revealed that regression models could not appropriately predict the salinity values and the vegetation indexes had poor correlation with the topsoil salinity values. The salinity percentages obtained from OK and Scl were nearly similar where the salinity was high (≥16dS/m), but differed in other salinity classes. Therefore, in the supervised classification of LISS-III sensor data the bare soil surfaces with high salinity (≥16dS/m) were successfully identified and separated from the rest of the soils. The regression model estimated 100 % of the study area as saline soil. The kriging method predicted 87.6 % of the area to be classified as saline soils (> 4dS/m), while supervised classification predicted that to be 62.5 %. Each of these methods has constrains. Therefore, we recommend the integration of these methods for estimating of soil salinity. Keywords: Ordinary Kriging, Regression Models, Supervised Classification, Topsoil Salinity.