Groundwater level prediction using genetic programming: the importance of precipitation data and weather station location on model accuracy

Groundwater (GW) level prediction is important for effective GW resource management. It is hypothesized that using precipitation data in GW level modelling will increase the overall accuracy of the results and that the distance of the observation well to the weather station (where precipitation data are obtained) will affect the model outcome. Here, genetic programming (GP) was used to predict GW level fluctuation in multiple observation wells under three scenarios to test these hypotheses. In Scenario 1, GW level and precipitation data were used as input data. Scenarios 2 only had GW level data as inputs to the model, and in Scenarios 3, only precipitation data were used as inputs. Long-term GW level time series data covering a period of 8 years were used to train and test the GP model. Further, to examine the effect of data from previous time periods on the accuracy of GW level prediction, 12 models with input data up to 12 months prior to the current period were investigated. Model performance was evaluated using two criteria, coefficient of determination (R2) and root mean square error (RMSE). Results show that when predicting GW levels through GP, using GW level and precipitation data together (Scenario 1) produces results with higher accuracy compared to only using GW level (Scenario 2) or precipitation data (Scenario 3). Additionally, it was found that model accuracy was highest for the well located closest to the weather station (where precipitation data were collected), demonstrating the importance of weather station location in GW level prediction. It was also found that using data from up to six previous time periods (months) can be the most efficient combination of input data for accurate predictions. The findings from this study are useful for increasing the prediction accuracy of GW level variations in unconfined aquifers for sustainable GW resource management