Forecasting soil moisture dynamics across diverse Indian river basins using a hybrid ConvLSTM model

In this study, a Convolutional long short-term memory network (ConvLSTM) is employed to forecast soil moisture for three Indian river basins with varying climatic conditions namely Godavari, Narmada and Cauveri. Utilizing a complete dataset from AgERA5 ranging over a decade, a number of meteorological forcings inducing soil moisture dynamics are incorporated to inform our forecast model. Starting with a global scale, the data undergo rigorous preprocessing, being refined to cater to the Indian basin scale, and subsequently tailored for our deep learning paradigm. By configuring the methodology around ConvLSTM network, the intrinsic patterns within the dataset were captured. This unification of Convolution neural network (CNN) and Long-short term memory network (LSTM) safeguarded complete data processing in both spatial and temporal perspectives, thereby bestowing an unparalleled basis for dismembering complex spatial-temporal sequences, making it ideal for tasks like soil moisture forecasting using extensive meteorological data. An all-inclusive evaluation of the proposed network is presented in form of a comparative analysis with four baseline models across all the river basins mentioned. Results in terms of evaluation metrices, underscore the ConvLSTM-based model’s ability in untying the nuanced spatial and temporal variability of soil moisture ahead of the baseline models. The robustness of the proposed network is further scrutinized by correlating ConvLSTM-derived soil moisture forecasts with those derived from another satellite-based product, namely the Soil Moisture and Ocean Salinity (SMOS), juxtaposed against the AgERA5 reanalysis data for 3-day and 5-day forecast horizons across the same river basins, showing good correlation. Such proficiency, overlay the means for possible progressions in agricultural approaches, improved drought prediction, and advanced management of water resources across various Indian river basins.