Machine learning applied to ground water-surface water connectivity in the wet pampa plain: ground water level forecast and assistance for baseflow separation methods

Setting up a physically based model to determine the connectivity between a river and its underlying aquifer can be complex, making machine learning (ML) techniques attractive for processing and exploiting time series associated with hydrometeorological variables acquired by remote sensors and in situ measurements in and around a local section of a channel. Data-driven models do not require knowledge of the basin’s characteristic physical parameters or its spatiotemporal distribution. If the inputs are of sufficient quality, the dynamics of ground water-surface water (GW-SW) exchange can be predicted at different time scales. The experience of the authors working with Ground Water Level (GWL) forecasting and baseflow filtering using the Machine Learning for Flood Forecasting (ML4FF) framework at the topographically “depressed” or wet Argentinian Pampa plain is described.