Modelling groundwater level fluctuation in an Indian coastal aquifer
Estimating groundwater level (GWL) fluctuations is a vital requirement in hydrology and hydraulic engineering, and is commonly addressed through artificial intelligence (AI) models. The purpose of this research was to estimate groundwater levels using new modelling methods. The implementation of two separate soft computing techniques, a multilayer perceptron neural network (MLPNN) and an M5 model tree (M5-MT), was examined. The models are used in the estimation of monthly GWLs observed in a shallow unconfined coastal aquifer. Data for the water level were collected from observation wells located near Ganjimatta, India, and used to estimate GWL fluctuation. To do this, two scenarios were provided to achieve optimal input variables for modelling the GWL at the present time. The input parameters applied for developing the proposed models were a monthly time-series of summed rainfall, the mean temperature (within its lag times that have an effect on groundwater), and historical GWL observations throughout the period 1996–2006. The efficiency of each proposed model for Ganjimatt was investigated in stages of trial and error. A performance evaluation showed that the M5-MT outperformed the MLPNN model in estimating the GWL in the aquifer case study. Based on the M5-MT approach, the development of this model gives acceptable results for the Indian coastal aquifers. It is recommended that water managers and decision makers apply these new methods to monitor groundwater conditions and inform future planning.
Keywords: groundwater level estimation multilayer perceptron neural network, M5 model tree Indian coastal aquifers time-series modelling