Mapping specific groundwater vulnerability to nitrate using random forest: case of Sais basin, Morocco

• Published in: Modeling earth systems and environment 2020-09, Vol.6 (3), p.1451-1466
• Main Authors: Lahjouj, Abdelhakim, El Hmaidi, Abdellah, Bouhafa, Karima, Boufala, M’hamed
• Format: Article
• Language: English
• Subjects: Calcium; Calcium carbonate; Calcium carbonates; Carbon-nitrogen ratio; Carbonates; Chemistry and Earth Sciences; Clay; Computer Science; Datasets; Earth and Environmental Science; Earth Sciences; Earth System Sciences; Ecosystems; Environment; Groundwater; Hydraulic gradient; Land use; Lithology; Mapping; Math. Appl. in Environmental Science; Mathematical Applications in the Physical Sciences; Nitrates; Nitrogen; Organic matter; Original Article; Physics; Pollution; Rain; Rainfall; Sand; Slope gradients; Soil; Soil classification; Statistics for Engineering; Training; Vulnerability; Water analysis; Water pollution; Water sampling
• ISSN: 2363-6203; 2363-6211
• DOI: 10.1007/s40808-020-00761-6

The objective of this study was to assess the groundwater vulnerability to nitrate (NO 3 − ) pollution in the Sais basin, based on the drinking threshold (50 mg/L), using the random forest (RF) model. A spatial dataset consists of the nitrate concentrations observed in 154 water samples and 14 explanatory variables was considered in this research. These variables are rainfall, texture (sand, silt, and clay), lithology, organic matter, piezometric level, altitude, land use, calcium carbonate (CaCO 3 ), carbon/nitrogen ratio (C/N), slope, hydraulic gradient, and soil classification. 80% of the dataset was randomly selected for training and validation, and the remaining 20% for testing the RF model. The RF model was validated and tested using out-of-bag (OOB) error and receiver operating characteristic (ROC) curve. The error computed and the area under the curve for success rate were 0.11 and 82.2%, respectively. In addition, the RF result revealed that rainfall, sand content, clay content, piezometric level, organic matter, and lithology are the key factors determining groundwater vulnerability to NO 3 − in the Sais basin. However, using only these most important factors as RF inputs, the prediction accuracy was found to be slightly similar to that obtained using all variables. The groundwater vulnerability maps were created using the groundwater vulnerability indexes predicted. The most reliable groundwater vulnerability maps to NO 3 − showed that about 48 and 63% of the surface area of the basin are under high to very high vulnerability level, using all and most important explanatory variables, respectively. This study serves to determine the most vulnerable areas and to identify the factors affecting NO 3 − pollution in the Sais basin, to properly control and protect groundwater.