Cluster and regression analysis for predicting salinity in groundwater

Abstract

Groundwater salinity is a major problem particularly in the northeastern region of Thailand. Saline groundwater can cause widespread saline soil problem resulting in reducing agricultural productivity as in the Lower Nam Kam River Basin. In order to better manage the salinity problem, it is important to be able to predict the groundwater salinity. The objective of this research was to create a cluster-regression model for predicting the groundwater salinity. The indicator of groundwater salinity in this study was electrical conductivity because it was simple to measure in field. Ninety-eight parameters were measured including precipitation, surface water levels, groundwater levels and electrical conductivity. In this study, the highest groundwater salinity at 3 wells was predicted using the combined cluster and multiple linear regression analysis. Cross correlation and cluster analysis were applied in order to reduce the number of parameters to effectively predict the quality. After the parameter selection, multiple linear regression was applied and the modeling results obtained were R2 of 0.888, 0.918, and 0.692, respectively. This linear regression model technique can be applied elsewhere in the similar situation.