Synergistic Autoregressive Fuzzy-SVM Model for Monthly Groundwater Level Prediction: Case Study of the Birjand Aquifer

چکیده مقاله

While advanced models for groundwater level (GWL) forecasting have proliferated, their reliance on extensive auxiliary data and computationally intensive hyperparameter tuning limits real-world deployment, particularly in data-scarce arid regions. To bridge this, the present study introduces the synergistic autoregressive fuzzy-support vector machine (SAR-FSVM) framework for monthly GWL forecasting in the Birjand aquifer, Iran, and comparing its performance against conventional approaches, including multiple linear regression (MLR), support vector machine (SVM), and autoregressive SVM (AR-SVM). The proposed framework uniquely integrates three complementary components: the temporal dependency modeling of autoregressive (AR) models, the uncertainty quantification of fuzzy logic, and the nonlinear pattern recognition of SVMs within a single, end-to-end architecture that operates exclusively on historical GWL data. Using monthly data from 11 observation wells (1998–2017), the model was developed and validated. Results showed that SAR-FSVM outperforms conventional models, achieving a Nash–Sutcliffe efficiency coefficient (NSE) of 0.937 (training) and 0.907 (testing), a root-mean-square error (RMSE) of 0.228 m (training) and 0.267 m (testing), and a mean absolute error (MAE) of 0.166 m (training) and 0.190 m (testing). The fuzzy logic component effectively managed input variability, while SVM captured nonlinear dynamics. Analysis revealed that 1–3-month lags (