GIS-based multivariate adaptive regression spline and random forest models for groundwater potential mapping in Iran

Abstract

This study evaluated and compared groundwater spring potential maps produced with two different models—namely multivariate adaptive regression spline (MARS) and random forest (RF)—using geographic information system (GIS). In total, 234 spring locations were identified in the Boujnord, North Khorasan, Iran and a GIS spring inventory map was prepared. Of these, 176 (70 %) locations were employed to produce spring potential maps (training), while the remaining 58 (30 %) cases were used to validate the model. The explanatory variables used to predict spring location were altitude, slope aspect, slope degree, slope length, topographic wetness index (TWI), plan curvature, profile curvature, land use, lithology, distance to rivers, drainage density, distance to faults, and fault density. Furthermore, the spatial relationships between spring occurrence and explanatory variables were performed using a Certainty Factor (CF) model. For validation, area under a receiver operating characteristics (ROC) curves (AUC) was used. The validation results showed that the AUC for calibration is almost identical (0.79) in both models, while for prediction, the MARS model (73.26 %) performed better than RF (70.98 %) model. These results indicate that the MARS and RF models are good estimators of groundwater spring potential in the study area. These groundwater spring potential maps can be applied to groundwater management and groundwater resource exploration.