Mapping Slight and Moderate Saline Soils in Irrigated Agricultural Land Using Advanced Land Imager Sensor (EO-1) Data and Semi-Empirical Models

Abstract

ABSTRACTAround the world, especially in semi-arid regions, millions of hectares of irrigated agricultural land are abandoned each year because of the adverse effects of irrigation, mainly secondary salinity and sodicity. Accurate information about the extent, magnitude, and spatial distribution of salinity and sodicity will help create the sustainable development of agricultural resources. In Morocco, south of the Mediterranean region, the growth of the vegetation and potential yield are limited by the joint influence of high temperatures and water deficit. Consequently, the overuse of surface and ground water, coupled with agricultural intensification, generates secondary soil salinity. Knowing when, where, and how salinity may occur is very important to the sustainable development of any irrigated production system. Remedial actions require reliable information to help set priorities and to choose the type of action that is most appropriate in each situation. Ground-based electromagnetic measurements of soil electrical conductivity (EC) are generally accepted as the most effective method for quantification of soil salinity. Unfortunately, these methods are expensive, time consuming, and need considerable human resources for land surveying. Moreover, the dynamic nature of soil salinity in space and time makes it more difficult to use conventional methods for comparisons over large areas. A major challenge of remote sensing, as a potential alternative technique, is to detect different levels of soil salinity. The main aim of this research is to assess the potential of the Advanced Land Imager (ALI) sensor on board the Earth Observing-1 (EO-1) satellite, with its rich infrared bands, for the discrimination and mapping of slight and moderate soil salinity in the Tadla’s irrigated agricultural perimeter in Morocco. To achieve this goal, semi-empirical predictive models developed in a previous study using second order regression analysis between the EC of salt-affected soils and different spectral salinity indices were applied to the ALI image. This was atmospherically corrected and the radiometric sensor drift was calibrated. Visual comparisons and statistical validation of these models using ground truth were undertaken in order to identify the best semi-empirical model for slight and moderate salinity mapping. The obtained results show that the model based on the Normalized Difference Salinity Index (NDSI) does not give any results. The model based on the Salinity Index-1 (SI-1) and the SI-Advanced Space-borne Thermal Emission and Reflection Radiometer (SI-ASTER) confuses vegetation with high soil salinity, although the model does bring out areas of lower salinity. Both R2 of 0.67 for the SI-1 and 0.65 for the SI-ASTER further reinforce that these models cause too much confusion to be used with accuracy for salt-affected soil detection. The semi-empirical model based on Soil Salinity and Sodicity Index-1 (SSSI-1) performs better than the two last models. However, there is a relative confusion between the classes in the slight and moderate salinity and in areas that are shown by the validation map; the higher class of salinity does not appear to contain higher levels of salinity. The statistical validation of this model reinforces what is seen on the derived map with only an R2 = 0.68. The model based on the SSSI-2 clearly provides the best results in comparison to the ground truth. Its derived map gives the closest overall visual approximation of the EC map, with a whole range of values. With a statistical validation of R2 = 0.97 to the ground truth, it is by far the best performance of any of the other models, and the different classes are statistically well separated, which further reinforces the accuracy of the visual analysis.