CONCEPTION DE MODÈLES STATISTIQUES A VARIABLES HYDROCHIMIQUES POUR LA PRÉDICTION DE LA CONDUCTIVITÉ ÉLECTRIQUE DES EAUX SOUTERRAINES

Abstract

The study takes place on the regions of N'zi-Iffou-Moronou (ex-N'zi-Comoé), located in East-Central of Ivory Coast with an area of 19,560 km². The objective of this study is to develop statistical models using multiple linear regressions able to simulate the electrical conductivity of groundwater from physical and chemical parameters of water. Sample from 193 boreholes were used for this study. The physical parameters are temperature, pH, turbidity and electrical conductivity. Chemical parameters selected are the major cations and major anions. The methodology consisted on the one hand, a normalized principal component analysis (NPCA) to identify the relevant variables, on the second hand, the model calibration to determine the coefficients of linear regressions and standard errors associated. The results of the normalized principal component analysis show that the most relevant and expressive modeling of the electrical conductivity of groundwater variables are the sum of the major cations and the sum of the major anions. These two parameters have been used as variables for designing a first model. Also, to show the impacts of anthropogenics activities on groundwater quality, a second model was proposed by integrating the temperature and pH in the first model. The values of the coefficient of variation on the coefficients of regression of the explanatory variables are relatively low (lower than 10%) with regard to the chemical parameters whatever the model. They are followed by the pH (23%) then the temperature (511%).