Mapping groundwater potential zones in Kanchanaburi Province, Thailand by integrating of analytic hierarchy process, frequency ratio, and random forest

Abstract

At the basic level, groundwater potential zone (GWPZ) mapping plays an important role in sustainable water resource management. There are different approaches to delineating GWPZ, and each has unique advantages and disadvantages. Incorporating these approaches into an ensemble could provide a more efficient tool for GWPZ evaluation and mapping. In this study, the frequency ratio (FR), random forest (RF), and analytic hierarchy process (AHP) models, and their ensemble were compared in delineating GPWZs in Kanchanaburi Province, Thailand. These models predicted the potential of groundwater yield at > 10 m3/h and were trained based on the measured groundwater yield of 1,601 wells in the study region, coupled with the spatial data of eight influencing factors, including altitude, distance to faults, distance to waterbodies, geology, land use, rainfall, soil type, and slope. The Areas under the curve (AUC) metric was used to assess the model’s performance. The results demonstrated that all models achieved similarly good performance with an AUC of 0.80, 0.76, 0.74, and 0.72 for the ensemble, RF, FR, and AHP models, respectively. Areas with high groundwater yield potential were primarily reported in the eastern part of Kanchanaburi, where the terrain is flat. The ensemble approach slightly improved the predictive power, but at the cost of model complexity.