Groundwater level forecasting model using hybrid support vector regression -particle swarm optimization for aquifer in Udupi region

Groundwater level in Udupi region decline invariably because of variation in rainfall and the ruggedness of topography and porous nature of lateritic rock. The groundwater level forecasting model needs to be developed for investigating the varying water level to save this precious water resource. This research work is focused in the Udupi region located in Karnataka state, India for two different types of major geological formations, lateritic terrain and Banded Gneissic Complex (BGC). In this paper hybrid Particle Swarm Optimization guided support Vector Regression (SVR) approach is employed to forecast the future trend. The particle swarm optimization algorithm (PSO) is used to select optimal SVR based parameters. Hybrid SVR –PSO model is tested with historical groundwater level data and rainfall data collected from Udupi Region. Forecasting performance of the Artificial Neural Network (ANN) and SVR models were analyzed using statistical metrics like MAE, NMSE. The ANN model shows high performance for larger dataset and SVR models shows high performance for limited dataset. The result indicates that the SVR shows less relative error than ANN and SVR could be a better alternative for forecasting GWL with limited data. Thus SVM shows promising result and can be used as a practical tool for cases where less measured data is available for future prediction.