Groundwater potential model using statistics and hybrid machine learning approaches

Abstract

The present study was conducted to delineate the groundwater potential in Kanchanaburi Province, Thailand based on groundwater yield, groundwater contamination risk, and groundwater quality. In this study, an ensemble model was created by combining Analytical Hierarchy Process, Frequency Ratio, and Random Forest to evaluate the spatial distribution map of the groundwater resources. Additionally, a new hybrid approach was developed based on maximum entropy and analytical hierarchy process to delineate the Ni contamination risk in groundwater. Finally, four machine-learning models, including Random Forest - Cross validation, Random Forest – Bootstrap, Artificial Neural Network - Cross validation, and Artificial Neural Network - Bootstrap were used to decipher groundwater quality. The results indicated that the ensemble model was better than individual models in delineating groundwater yield potential. Poor and moderate potential with groundwater yield > 10 m3/h was distributed mainly in the western Kanchanaburi, while the eastern regions showed high groundwater yield potential (bao nhiêu m3/h). In terms of contamination risk, the hybrid model between maximum entropy and analytical hierarchy process gave a high performance with an Area Under Curve of 0.86 and Accuracy of 0.85. The map of Ni contamination risk in groundwater showed that approximately 24.79% of the eastern Kanchanaburi (1691.82 km2) was a very low contamination risk of Ni, whereas the zone with high Ni contamination risk accounted for around 6.56% (447.65 km2). Moderate contamination risk zone of Ni occupied 68.65% of the eastern region. In the groundwater quality assessment, Random Forest - Cross validation was the best in deciphering the groundwater quality map, compared to the Random Forest – Bootstrap, the Artificial Neural Network - Cross validation, and the Artificial Neural Network – Bootstrap models. According to the best model (Random Forest - Cross validation), around 64.78% and 29.39% of the eastern Kanchanaburi were good and very good groundwater quality while only 0.58% and 0.08% were poor and very poor groundwater quality, respectively. Meantime, 5.17% was designated to be moderate groundwater quality. In conclusion, groundwater agencies can release policies on groundwater management. It can be done by publicizing the list of restricted areas from the exploitation of groundwater, orienting in reasonable groundwater exploitation, and usage with different purposes.