Assessing the performance of machine learning and analytical hierarchy process (AHP) models for rainwater harvesting potential zone identification in hilly region, Bangladesh

Water scarcity in hilly regions presents unique challenges, particularly in Bangladesh, where obtaining fresh drinking water has become difficult to access. This study aims to evaluate the potential zones for rainwater harvesting (RWH) using machine learning (ML) algorithms and geospatial analysis. Specifically, four ML algorithms—random forest (RF), boosted regression trees (BRT), k-nearest neighbors (KNN), and naïve bayes (NB)—alongside the analytical hierarchy process (AHP) were employed to delineate potential RWH zones in the Chattogram hilly districts, including Chattogram, Rangamati, Bandarban, Khagrachari, and Cox’s Bazar. Eleven influencing factors were considered: aspect, distance from road, drainage density, elevation, hill shade, lineament density, land use/land cover (LULC), slope, topographic wetness index (TWI), rainfall, and geology. Inventory data from the study area, consisting of 135 suitable and 135 non-suitable points, were randomly split, with 70% used for training the models and the remaining 30% for validation using the area under the curve (AUC) values. The southern regions are highly suitable for harvesting rainwater. Among the five models, BRT and RF demonstrated superior performance with AUC values of 0.93 for both models. In contrast, the AHP method yielded the lowest AUC value at 0.82. Notably, drainage density and elevation emerged as the most influential factors in constructing these models. The application of machine learning algorithms has enhanced the precision of rainwater harvesting zone estimate systems by examining diverse aspects. The findings of this study can provide valuable insights for policymakers in making informed decisions regarding RWH in these regions.