Groundwater level forecasting using data-driven approaches

Abstract

<!--!introduction!--><b></b><p>Groundwater is a key freshwater resource used for agricultural, industrial, and municipal water supply. Climate change and human activities have resulted in an increased dependence on groundwater in many aquifers around the world. Climate change extremes such as floods and droughts, population growth along with increasing dependence on using groundwater for irrigation and industrial activities pose challenge to the future availability of groundwater. Accurate and reliable methods of groundwater level (GWL) forecasting are a key tool which can inform the groundwater managers about the future quantitative availability of groundwater. Although a variety of numerical and statistical approaches have been applied for GWL forecasting of single wells, a global forecasting method utilizing the GWL time series from several well sites in an aquifer is applied in this study. The global forecasting approach leverages algorithms such as Deep Learning (DL), Long Short-Term Memory (LSTM), Nonlinear Autoregressive Neural Network (NARX) which can model complex non-linear time series relationships between input and output variables. We primarily use precipitation and temperature meteorological data along with GRACE derived Terrestrial Water Storage (TWS) as the input for models and GWL as the output. &nbsp;We implement the algorithm in several stressed aquifers around the world, such as Central Valley in California, High Plains in Kansas, Indo-Gangetic basin in northern India, and North China Plain in China.&nbsp; The data-driven approaches provide robust forecasting and can thus form the basis for effective management decisions and strategies.</p>The 28th IUGG General Assembly (IUGG2023) (Berlin 2023)