Advanced neural network approaches for optimal check dam site selection in sub-tropical climates

This study explores the suitability of check dam sites in the Kangsabati River Basin using three advanced neural network models: Convolutional Neural Networks (CNN), Long Short-Term Memory (LSTM), and Recurrent Neural Networks (RNN). The objective is to identify optimal locations for check dam construction to improve water conservation, sediment trapping, soil erosion control, and groundwater recharge. A thorough multi-collinearity assessment ensured the datasets’ robustness. Among the models, CNN demonstrated the best performance, achieving a sensitivity of 0.93, specificity of 0.85, and AUC scores of 0.91 during training and 0.82 during validation. LSTM and RNN models showed AUC scores of 0.799 and 0.787, respectively. Spatial analysis indicated that 14.33% of the study area was classified as highly suitable and 17.21% as very highly suitable in the CNN model. These findings provide a reliable framework for identifying optimal check dam sites, which could contribute significantly to sustainable water resource management in the region. Future studies should focus on land use, climate change impacts, and design modifications to improve long-term check dam performance.