Effectiveness of machine learning ensemble models in assessing groundwater potential in Lidder watershed, India

• Published in: Acta geophysica 2024, Vol.72 (4), p.2843-2856
• Main Authors: Ali, Rayees, Sajjad, Haroon, Saha, Tamal Kanti, Roshani, Masroor, Md, Rahaman, Md Hibjur
• Format: Article
• Language: English
• Subjects: Acidic soils; Alluvial plains; Barren lands; Developing countries; Drainage density; Earth and Environmental Science; Earth Sciences; Effectiveness; Elevation; Foothills; Geophysics/Geodesy; Geotechnical Engineering & Applied Earth Sciences; Gravel; Groundwater; Groundwater potential; Groundwater resources; LDCs; Machine learning; Multilayer perceptrons; Rainfall; Regression analysis; Research Article - Hydrology and Hydraulics; Snow cover; Statistical analysis; Structural Geology; Urbanization; Vegetation; Water resources management; Watersheds
• ISSN: 1895-7455; 1895-6572; 1895-7455
• DOI: 10.1007/s11600-023-01237-8

Groundwater is under stress due to the growth of population, urbanization and industrialization, especially in developing countries. Consequently, there is a shortage of water in many parts of Indian cities. Thus, monitoring and assessing groundwater are crucial for effective management of water resources. The main objectives of our study is to explore the effectiveness of machine learning models for assessing groundwater potential in Lidder watershed in India. The site-specific factors influencing groundwater potential were selected for preparing groundwater potential zones in the study area. These factors were integrated for preparing the groundwater potential map using three models, namely logistic regression (LR), ensemble of random forest (LR-RF) and multilayer perceptron layer (LR-MLP). The area under curve of receiver operating characteristic (ROC) and several statistical performance measures were used to validate and compare the performance of the models. The validation revealed that the maximum value of AUC-ROC (0.946) was found for LR-RF for assessing groundwater potential than the LR and LR-MLP models. The results revealed low groundwater potential mostly found in the northern part of the watershed. High elevation, steep slope, highly alkaline and acidic soils, snow cover, barren land and hard basaltic rocks are attributed to low groundwater potential. Rainfall, vegetation, low elevation, younger alluvial plain, low drainage density and high lineament density have made groundwater potential high and very high in the southern part of the watershed. The central part of the watershed experienced moderate groundwater potential due to the piedmont alluvial plain, basal gravel, moderate rainfall, moderate drainage density, and scrub vegetation. We argue that the other geographical areas that are interested in properly managing groundwater resources can make use of the logistic regression and ensemble models of random forest and multilayer perceptron (MLP) for groundwater potential assessment.