Prediction of groundwater levels using a long short-term memory (LSTM) technique

The precise prediction of groundwater levels is a challenging task due to the complex relationships between hydrological parameters and the lack of in situ climate data. The present research proposed an integrated machine learning model for groundwater level prediction based on long short-term memory (LSTM) along with principal component analysis (PCA) and discrete wavelength transform (DWT), i.e. PCA–DWT–LSTM model. The proposed model was developed using 23 years (2000–2022) of seasonal groundwater level data and climatic variables for nine wells in the district of Kangra in Himachal Pradesh, India. The proposed model attains higher ranges of R2 (0.8253–0.8828) and lower ranges of root mean square error (RMSE) (0.1011–2.0025) than the alternative model (PCA–LSTM), having R2 and RMSE values in the range of 0.7019–0.8005 and 0.2662–2.9565, respectively. Moreover, when compared to the hybrid models, the accuracy of the DWT-based models is much higher. The developed model (PCA–DWT–LSTM) improves the accuracy and interpretability of groundwater level prediction and has the potential to estimate the accurate groundwater level, particularly in the regions where obtaining the hydrogeological data is difficult.