Predicting Groundwater Level Fluctuations Using Hybrid SVM-SSA Algorithm in Cuttack, Odisha: A Case Study

Accurate groundwater level (GWL) estimation is significant to attain sustainable developmental goals and management of integrated water resources. But, its all-time accessibility is of serious concern. Hence, it is crucial to understand the groundwater potential for utilisation of water resources. There has been a significant decline in urban groundwater resources in past few decades because of over-exploitation, climatic change, urbanisation, and population growth. For understanding the effect of climate change variables on the fluctuation of GWL, a machine learning-based model is developed integrating sparrow search algorithm (SSA) and support vector machine (SVM). The developed model is utilised for predicting GWLs in Cuttack, a heavily inhabited city with decreasing groundwater resources. Precipitation, minimum relative humidity, and GWL of current, one-month lag, and two-month are taken as input parameters to predict the GWL on monthly basis. Three statistical indices, namely root mean square error (RMSE), coefficient of determination (R2), and Nash–Sutcliffe efficiency (NS), are applied for performance evaluation of SVM-SSA method. The results reveal that SVM-SSA with RMSE of 0.5876 (m), R2 of 0.9698, and NS of 0.968 significantly outperforms SVM with RMSE of 9.9891 (m), R2 of 0.9345, and NS of 0.9326. The study’s findings indicate that SVM-SSA model is an acceptable data-driven method for predicting skewed and nonstationary monthly GWL time series, demonstrating a suitable tool for monthly GWL prediction.