A numerical modelling and neural network approach to estimate the impact of groundwater abstractions on river flows

Evaluation of the impacts of groundwater abstractions on surface water systems is a necessary task in integrated water resources management. A range of hydrological, hydrogeological, and geomorphological factors influence the complex processes of interaction between groundwater and rivers. This paper presents an approach which uses numerical modeling of generic river–aquifer systems to represent the interaction processes, and neural networks to capture the impacts of the different controlling factors. The generic models describe hydrogeological settings representing most river–aquifer systems in England and Wales: high diffusivity (e.g. Chalk) and low diffusivity (e.g. Triassic Sandstone) aquifers with flow to rivers mediated by alluvial gravels; the same aquifers where they are in direct connection with the river; and shallow alluvial aquifers which are disconnected from regional aquifers. Numerical model simulations using the SHETRAN integrated catchment modeling system provided outputs including time-series and spatial variations in river flow depletion, and spatially distributed groundwater levels. Artificial neural network models were trained using input parameters describing the controlling factors and the outputs from the numerical model simulations, providing an efficient tool for representing the impacts of groundwater abstractions across a wide range of conditions. There are very few field data sets of accurately quantified river flow depletion as a result of groundwater abstraction under controlled conditions. One such data set from an experimental study carried out in 1967 on the Winterbourne stream in the Lambourne catchment over a Chalk aquifer was used successfully to test the modeling tool. This modeling approach provides a general methodology for rapid simulations of complex hydrogeological systems which preserves the physical consistency between multiple and diverse model outputs.