Assessment of groundwater scenario in drought-prone semi-arid regions of the Indian subcontinent using time series decomposition and the machine learning models

Groundwater is most vital resources to survive people around the globe in terms of drinking, domestic, and irrigation purposes. About two billion people are heavily reliant on this resource for drinking. In recent days, both the quality and quantity of groundwater have deteriorated in the semi-arid region, particularly in the plateau fringe of the Bundelkhand in India. In this connection, it is very much essential to evaluate and assess the variations of groundwater storage and groundwater status over the Bundelkhand plateau in the heart of India. Therefore, the principal objectives of this study are i) to estimate the terrestrial water storage (TWS), Soil moisture (SM) and Groundwater storage (GWS) using Gravity Recovery and Climate Experiment (GRACE) and TerraClimate monthly datasets and ii) to assess the groundwater fluctuations and regional groundwater depletion trends using analytical methods using local insitu well data for water resource management. However, this study also emphasized predicting future conditions using both the time series decomposition model (Prophet) and the machine learning model (Random Forest) over the Bundelkhand plateau of India. The result indicates that the estimated water thickness value ranges from 23.23 cm to -22.07 cm (2002-2017). Average soil moisture changes vary from 1.58 cm (2016) to 17.41 cm (2003). In addition, maximum groundwater storage was 10.52 (2003), and minimum negative water storage occurred in -23.99 cm (2012). However, the average groundwater storage was -7.59 cm in the study period (2002 to 2017). Predicted results demonstrate that in the future (up to 2032) this region will face in shortage of groundwater ( -28.65 cm). In terms of R 2 value, RF performs slightly better (0.99) than Prophet (0.98). Each forecasting model indicates that a continuous decline trend up to 2032 in the Bundelkhand Plateau region. In this connection, the application of Google Earth Engine (GEE) and Python Notebook has a decent result on hydrological research using observed and satellite datasets, particularly in terms of a huge region with inadequate borehole datasets. Still, the weakness of this study is highly associated with the data shortage. In addition, this study incorporated simplified subtraction approaches without explicitly accounting for surface water and other hydrological components. However, this study will be helpful for future researchers and administrators to formulate short-term measures and long-term strategies in a semi-arid dry environment.