Input data, NARX model and groundwater level forecast
The input data precipitation, relative humidity, temperature and a sinusoidal curve adapted to the temperature are shown above, which provide the model with noise-free information on seasonality. These data are incorporated into a NARX model (with one input and output layer, three hidden layers and a feedback connection). The groundwater level forecast shown below shows the results of 10 pseudo-randomised model initialisations (median with a strong red line).
Source: KIT & BGR