remote sensing and machine learning for groundwater spring potential assessment in eastern and northeastern iran

Groundwater (gw) resources are being over-exploited in many parts of the world due to the increasing demand for water driven by population growth and industrialization. this study addresses the critical need for assessing gw potential for sustainability, focusing on eastern and northeastern iran. this research leverages a comprehensive analysis of environmental variables using advanced machine learning algorithms to model spring potential in this specific area. sixty-six environmental variables were analyzed, including physiographic, climatic, soil, geological, vegetation cover, and hydrological factors. various machine learning models, such as glm, gbm, cta, ann, sre, fda, mars, rf, maxent, and esms were employed. model accuracy was evaluated using kappa, tss, and roc indices, with 70% of the data used for training and 30% for evaluation through five repetitions. the findings indicated that random forest (rf) model achieved the highest accuracy based on the evaluation criteria. relative importance analysis revealed that topographic factors (altitude, twi, slope), climatic factors (bio7, bio19, bio12), and soil factors (sand 60-100 cm, silt 60-100 cm, clay 0-5 cm, land surface temperature) were the most influential in predicting spring potential. the rf and ensemble (esms) models identified 13.04% to 15.07% of the study area as having high to very high groundwater potential. the high performance of rf model and the identified key influencing factors provide valuable insights for sustainable water resource management in this data-scarce region. the findings underscore the utility of remote sensing-derived variables and machine learning for groundwater assessment and offer a practical gwpm for governmental and private sector use.