Data-driven models, especially artificial neural networks (ANN), have proven their suitability for modelling and predicting groundwater levels many times. In particular, ANNs are characterised by the lower effort and are less dependent on field data availability compared to conventional numerical-physical modelling approaches, especially when it comes to (supra-)regional problems. New developments in the field of deep learning promise a significant improvement of already existing prediction approaches.
A reliable forecast of groundwater levels is, for example, the basis for deriving water availability for drinking water supply and agricultural irrigation, the demarcation of potential soil subsidence zones due to extremely low groundwater levels in connection with droughts and/or water abstraction, the demarcation of areas of potential groundwater floodings to protect transport infrastructure, buildings and agricultural land, as well as the development of suitable avoidance and adaptation strategies.
Our goal is to develop a method that enables nationwide short-, medium- and long-term forecasts of groundwater levels and spring discharges, especially with regard to extreme events. To this end we improve existing approaches based on artificial neural networks, which already provide robust weekly, monthly and seasonal forecasts at individual groundwater wells, and investigate promising new deep learning approaches. Data basis mainly are groundwater data of the federal state measuring networks as well as weather and climate forecasts and projections of the German Meteorological Service (DWD) and other research institutions (e.g. MPI-M).
Example Groundwater Level Prediction