Using statistical models for the prediction of an indicator parameter of the groundwater quality: Case study of aquifers fissured groundwater of Ivory Coast

MICHEL AMANI KOUASSI

Abstract


The objective of this study aims the evaluation and the comparison of the predictive power of two statistical models using hydrochemical variables to simulate the electric conductivity of groundwaters in the hard rock aquifers. The quality assessment models in calibration and internal validation was based on the principle of "split-sample test" which consists in the calibration of two-thirds (2/3) of the sample and the internal data on the remaining third validation. In validation by spatial transposition, several test samples were used. By using several evaluation criteria (correlation coefficient and Nash-Sutcliffe) and graphic representations, an analysis of the performance and robustness of the two models was performed. For both models, high values ​​of performance criteria were generally obtained in calibration, validation and internal validation by spatial transposition reflecting a greater capacity for models to simulate the electrical conductivity of groundwater. Both models are efficient and robust. But, the model 2 is more efficient and robust than model 1 and is presented as the most suitable to study the reliability of the hydrochemical data with the Nash-Sutcliffe as the most appropriate criterion model. Based on the Nash-Sutcliffe, could retain the 60% ​​threshold for which a performance that is superior lead a balanced balance.



Keywords


Statistical Models, Hydrochemical variables, Performance, Robustness, Ivory Coast.

Full Text:

PDF (Français)




.