Groundwater potential assessment in Gia Lai province (Vietnam) using machine learning, remote sensing and GIS

Abstract

Population growth, urbanization and rapid industrial development increase the demand for water resources. Groundwater is an important resource in sustainable socio-economic development. The identification of regions with the probability of the existence of groundwater is necessary in helping decision makers to propose effective strategies for the management of this resource. The objective of this study is to construct maps of potential groundwater, based on machine learning algorithms, namely deep neural networks (DNNs), XGBoost (XGB), and CatBoost (CB), in the Gia Lai province of Vietnam. In this study, 12 conditioning factors, namely elevation, aspect, curvature, slope, soil type, river density, distance to road, land use/land cover (LULC), Normalized Difference Vegetation Index (NDVI), Normal Difference Built-up Index (NDBI), Normal Difference Water Index (NDWI), and rainfall were used, along with 181 groundwater inventory points, to construct the models. The proposed models were evaluated using the receiver operating characteristic (ROC) curve, the area under the curve (AUC), root-mean-square error (RMSE), mean absolute error (MAE). The results showed that the predictions of groundwater potential were most accurate using the XGB model; CB came second, and DNN was performed the least well. About 4,990 km² of the study area was found to be in the category of very low groundwater potential; 3,045 km² was in the low category; 2,426 km² was classified as moderate, 2,665 km² as high, and 2,007 km² as very high. The methodology used in the study was effective in creating groundwater potential maps. This approach, used in this study, can provide valuable information on the factors influencing groundwater potential and assist decision-makers or developers in managing groundwater resources sustainably. It also supports the sustainable development of the territory, including tourism. This methodology can be used in other geographic regions with a small change of input data.