Environmental and Sustainability Indicators (Dec 2020)
Predictive mapping of soil electrical conductivity as a Proxy of soil salinity in south-east of Algeria
Abstract
In semi-arid and arid areas soil salinity has adverse effects both on the environment and agricultural production. The region of Biskra (South-East of Algeria) underwent a strong agricultural transformation from traditional oasis agriculture to an almost exclusive production of dates involving market gardening throughout the year. The main goal was to predict the spatial variation of EC using geostatistics and a Geographic Information System (GIS), comparing also the performance of two classical geostatistical interpolators - Ordinary Kriging (OK), using only point data, and Cokriging (CK), introducing also auxiliary variables to improve prediction accuracy (SI gypsum and SO42−, obtained from the analysis of the chemical and geochemical processes of soil salinization). For this study, a total of 42 soil samples were randomly collected from topsoil (0–15 cm) in the irrigated perimeter of El Ghrous, a representative rural community located in the west of Biskra. Aiming to better understand the processes that most influence the evolution of soil salinity in this area, some chemical parameters were determined, among which the electrical conductivity (EC). Moreover, some terrain parameters were derived from a digital elevation model as auxiliary information, and Normalized Difference Vegetation Index (NDVI) was calculated from satellite imagery. The prediction efficiency of the methods was evaluated by calculating the mean error (ME) and the root mean square error (RMSE). The resulting maps showed that soils in the study area are affected by salinization. Cross-validation results showed a better performance in estimating EC of CK, after the introduction of the covariates, than OK, with an RMSE value of 0.92 vs. 1.53. This suggests a greater efficiency of CK in EC prediction in this area, confirming that the introduction of some auxiliary data correlated to the target variable significantly improves the interpolation. A third kriging technique, Indicator Kriging (IK) was applied to generate a map of the probability of exceeding a given threshold.
