Enhancing groundwater salinity estimation through integrated GMDH and geostatistical techniques to minimize Kriging interpolation error

This study investigates the effectiveness of combining the Group Method of Data Handling (GMDH) models and Kriging to reduce errors in groundwater salinity estimation. The study uses two datasets of electrical conductivity (EC) values collected from 109 observation wells in an agricultural region of Iran. The methodology comprises a two-step process. Initially, GMDH models were trained and validated to estimate groundwater EC values at any desired location based on the observed EC values from surrounding wells. Subsequently, these models were employed to expand the EC datasets, followed by using the Kriging interpolation approach to investigate the spatial variability of groundwater EC across the study area. During the cross-validation phase, the GMDH models achieved robust results. For the first dataset, the model maintained an average R-squared value of 0.81, with a corresponding root mean square error (RMSE) of 0.61 dS/m. Similarly, for the second dataset, the model performed well with an R-squared value of 0.84 and RMSE of 0.72 dS/m. The findings also indicated that employing GMDH to extend the original datasets improves the results provided by Kriging, particularly when the original datasets are increased by 40%. On average, for the first dataset, employing the integrated approach decreased the RMSE of Kriging by 18.34%, while for the second dataset, the decrease in RMSE was 19.21%. Overall, the findings suggest that expanding datasets can enhance the accuracy of estimating water quality variables, which has important implications for environmental monitoring and management.

Automated summary

This study investigates the effectiveness of combining GMDH models and Kriging to reduce errors in groundwater salinity estimation. The findings suggest that expanding datasets can enhance the accuracy of estimating water quality variables.