Support Vector Machine (SVM) for Groundwater Quality Modelling – Gaza Coastal Aquifer Salinity as A Case Study

Abstract

Artificial intelligence (AI) techniques such as artificial neural networks (ANNs) and support vector machine (SVM) exhibits a reliable performance in modelling complicated hydrological processes using relatively less cost, effort and data. SVM is a new technique compared with ANNs, and it has been developed based on the statistical learning theory. In this study, SVM based model was developed to simulate the salinity, (described as chloride concentration), of Gaza Coastal Aquifer (GCA). The developed model was trained using 10-years water quality data from 22 municipal wells in Khanyounis governorate in Gaza Strip. The potential model input variables were basically selected based on understanding the physical processes that govern GW salinity in GCA. Different combinations of input variables were evaluated based on the correlation coefficient and error criteria of the model’s output. The best SVM model showed good simulation accuracy where, the mean absolute percentage error (MAPE), and correlation coefficient (r) for test data set were 4.6% and 0.997 respectively. The model results indicated that the most influencing input variables on the chloride concentration in the study area wells were the previous chloride concentration for the past analysis period, abstraction rate, overall recharge in the well area, depth of pump screen, and well’s location with respect to the shoreline and adjacent aquifers. The developed model could effectively be utilized for analyzing the effect of input variables variation on the chloride concentration which is considered as a crucial step for integrated water resources management.