Ensemble based groundwater level prediction using neural network pattern fitting

Abstract

Prediction of groundwater level is implemented using Time-series prediction model and combined prediction model for learning the pattern and trend in groundwater level fluctuation, result show that the combined prediction model using, groundwater level time series and precipitation time series as input predictors is a better predictor. Study also shows that prediction is dependent on the pattern and trends at a particular location as every dataset depends on the dynamics of the location namely the geomorphology of the aquifer, the drainage inside the aquifer and pumping from the aquifer. Ensemble based forecasting is studied to fix the upper and lower limit of the prediction. Ensembles helped in fixing a range for the forecast instead of relying on a single unique value.