Daily Groundwater Level Prediction and Uncertainty Using LSTM Coupled with PMI and Bootstrap Incorporating Teleconnection Patterns Information

Daily groundwater level is an indicator of resources. Accurate and reliable (GWL) prediction crucial for resources management land subsidence risk assessment. In this study, a representative deep learning model, long short-term memory (LSTM), adopted to predict with the selected predictors by partial mutual information (PMI), bootstrap employed generate different samples combination training many LSTM models, predicted values models are used uncertainty assessment prediction. Two wells climate zones in USA were as case study. Different significant GWL two identified PMI from candidate incorporating teleconnection patterns information. The results show that significantly affected antecedent GWL, AO, Niño 3.4, 1 + 2, precipitation humid areas, 3, PNA arid areas. Predictor selection can assist improving performance model. relationship between modeled it achieved higher accuracy while areas was poorer due limited improved increasing correlation coefficient (R2) 10% 25% 2 compared generalized regression neural network (GRNN). Three evaluation metrics indicate reduced GRNN coupling be promising approach accurate zones.