Introduction and Goal In the coming century, the intensification of droughts and increasing human pressure on groundwater resources will make assessing aquifer system resilience a key indicator of water resource sustainability, increasingly important. This study aims to apply a comprehensive method to predict groundwater resilience in the Marvdasht Plain watershed. The quantitative Calamity Resilience System (CRS) index was used as a criterion to measure an aquifer's ability to recover to a desirable state following hydrological shocks. By integrating the Weight of Evidence (WoE) model, advanced feature selection techniques, and multiple validation methods, this study provides a reliable and practical predictive map to support management decisions.A spatial distribution map of groundwater resilience was generated, and its performance was evaluated using three methods: bootstrap, calibration curve, and random validation. The methodology employed in this research proves to be a reliable and effective means of predicting groundwater resilience in the Marvdasht Plain watershed. Materials and Methods The study area is located in the Marvdasht-Kharameh plain, part of the Maharloo and Bakhtegan basins in Fars Province, Iran. It lies between longitudes 52°15' to 53°27' E and latitudes 29°19' to 30°25' N. As one of the largest plains in Iran, the Marvdasht plain has a maximum elevation of 3,099 m at Mount Dashtak and an average elevation of 1,590 m, covering 3.8% of the province's area (Fosfizadeh et al., 2014; Khoshakhlaq et al., 2010). This plain is a major agricultural hub for Fars province, with 148,000 hectares of irrigated land and 22,000 hectares of dry farmland. However, this agricultural and economic prosperity has led to the over-exploitation of groundwater resources, resulting in significant challenges. The declining groundwater level has triggered widespread land subsidence and fissures (Rahnama and Mirasi, 2013). Alarmingly, the risk of subsidence has now extended to the historic sites of Persepolis and Naqsh-e Rostam, underscoring the critical need for this study. Methodologically, 21 initial environmental, hydrogeological, and climatic variables identified as determinants of groundwater status were prepared as raster layers. A multi-stage feature selection framework based on variance, correlation, collinearity, and mutual information was then applied to identify the most effective predictors. This process selected 10 key factors (including kernel density of exploitation wells, watercourse density, and distance from agricultural wells) for the final modeling. The relationships between the CRS index and these factors were quantified by integrating evidence and calculating WoE weights. The final groundwater resilience map was generated by integrating these WoE weights and classified into five distinct classes using a quantile-based method. Model validation was conducted through a comprehensive approach, including random cross-validation, the bootstrap method (with 1,000 iterations), and calibration analysis (using the Brier Score). Results and Discussion The comprehensive validation of the advanced Weight-of-Evidence (WoE) model demonstrates that the predictive model for the Groundwater Resilience Index (CRS) possesses excellent discrimination performance, with a very high AUC value of 0.920, and remarkable statistical stability.In the Marvdasht-Kharameh plain watershed, groundwater resilience was successfully mapped with appropriate accuracy. The final map, classified into five distinct classes using the quantile method, is as follows: Very Low (0.100 - 0.223), Low (0.223 - 0.367), Medium (0.367 - 0.503), High (0.503 - 0.669), and Very High (0.669 - 1.000). This classification reflects the inherent variability of the data, as each class contains approximately 20% of the study area, based on the empirical distribution of the normalized index values derived from the integrated WoE weights of the effective factors.The low resilience index values prevalent across most of the area (mean = 0.48) indicate a severe weakness in the aquifer system's capacity to recover from hydrological shocks. Consequently, zones identified with the lowest resilience should be prioritized for critical management interventions, such as reducing groundwater withdrawals, implementing artificial recharge projects, and establishing intensive monitoring networks. Conclusion and Suggestions The results of this study provide a critical tool for identifying areas with the lowest groundwater resilience. It is therefore strongly recommended that zones classified as having "very low" resilience be given the highest priority in management interventions. Such interventions should include curbing unauthorized withdrawals, implementing artificial recharge projects, establishing protection zones, and enforcing strict technical and legal oversight. Furthermore, the modeling framework developed here serves as a robust and generalizable scientific basis for assessing groundwater resilience in other critically endangered aquifers across the country. |