Random Forest-Based Groundwater Potential Zone Mapping in Latur District, Maharashtra, India

Abstract

Groundwater is a major source of water for agriculture, domestic use, and rural development in semi-arid regions of India. In Latur District, Maharashtra, groundwater occurrence is spatially variable because the Deccan basalt terrain has low primary porosity and depends mainly on weathered material, fractures, joints, vesicular zones, lineaments, and favourable surface conditions. This research applies Remote Sensing, Geographic Information System (GIS), and Random Forest (RF) machine learning to delineate groundwater potential zones (GWPZs) in Latur District. Seven spatial conditioning parameters—lithology, geomorphology, soil, slope, drainage density, lineament density, and land use/land cover (LULC)—were integrated with groundwater-well observations. A total of 109 wells were classified using a 12.8 m mean-depth threshold. Seventy percent of the observations were used for RF training and 30% were retained for independent validation. All input maps were standardized to WGS 84/UTM Zone 43N and a 30 m raster grid. The RF probability output was classified into very-low, low, moderate, high, and very-high groundwater-potential zones. The model produced an ROC-AUC value of 0.8192, showing good discriminative capability between groundwater occurrence and non-occurrence conditions. The final GWPZ map provides a district-scale planning tool for groundwater exploration, recharge planning, conservation measures, and future field investigation.

Introduction

The text presents a Random Forest (RF) and GIS-based approach for mapping groundwater potential zones (GWPZ) in Latur District, Maharashtra. Groundwater is especially important in Latur because the region frequently experiences drought and depends heavily on groundwater for drinking, agriculture, livestock, industry, and rural development.

Background

Groundwater availability in the Deccan Trap basalt terrain is highly variable because basalt generally has low primary porosity. Groundwater is mainly stored and transmitted through weathered basalt, fractures, joints, vesicular zones, red-bole layers, and contacts between lava flows. Therefore, identifying favorable groundwater areas requires considering several environmental and geological factors together.

A groundwater potential zone map does not guarantee groundwater at a particular location. Instead, it shows the relative likelihood of groundwater occurrence and can help with borehole planning, detailed groundwater surveys, artificial recharge, conservation, and monitoring.

Study Area

Latur District is located in the Marathwada region of Maharashtra, covering about 7,157 km². It is part of the Deccan Plateau and is dominated by basaltic formations. Because groundwater occurrence is strongly controlled by secondary porosity and geological structures, Latur is suitable for GIS and machine-learning-based groundwater assessment.

Data and Methodology

The study combines 109 groundwater-well observations with seven spatial parameters:

• Lithology – influences permeability, fractures, weathering, and groundwater storage.
• Geomorphology – represents landforms and their effects on runoff and recharge.
• Soil – controls infiltration and percolation.
• Slope – affects runoff speed and infiltration time.
• Drainage density – indicates drainage and runoff characteristics.
• Lineament density – represents fractures, faults, joints, and other structural features.
• Land Use/Land Cover (LULC) – represents vegetation, agriculture, built-up areas, and surface sealing.

Rainfall was excluded as an RF predictor because suitable spatially varying rainfall data matching the 30 m modelling grid and well-observation period were unavailable. The study recognizes rainfall as an important recharge factor and suggests including reliable rainfall data in future research.

Main Methodological Steps

The research follows a systematic workflow:

Groundwater-well data + spatial layers → Preprocessing → Raster alignment → Well classification → Feature extraction → Random Forest modelling → Probability mapping → ROC-AUC validation → GWPZ classification

The 109 wells were divided into:

• 76 training wells
• 33 validation wells

All spatial layers were aligned to a 30 m grid using WGS 84/UTM Zone 43N. Wells were classified using a 12.8 m mean-depth threshold, and the seven predictor values were extracted at each well location.

The Random Forest model was then used to produce both a binary groundwater-potential prediction and a continuous probability surface. The probability values were subsequently divided into five groundwater-potential classes, with ROC-AUC used to evaluate model performance.

Important Factors

The study particularly discusses two parameter pairs:

1. Lithology and Geomorphology

Lithology determines the availability of fractures, weathering, permeability, and storage. In basaltic areas, massive basalt generally has poor groundwater potential unless fractured or weathered, while vesicular and weathered basalt can provide better groundwater conditions. Geomorphology also affects runoff, regolith thickness, and recharge. Valleys and pediment-pediplain areas may offer favorable conditions when combined with gentle slopes, permeable soils, and fractured rocks.

2. Soil and Slope

Soil texture, thickness, porosity, and permeability control how much water infiltrates into the ground. Gentle slopes generally favor infiltration because water remains on the surface longer, while steep slopes encourage rapid runoff and reduce infiltration. However, soil and slope must be considered together with the underlying geology rather than independently.

Conclusion

This research developed a Random Forest-based groundwater-potential zone map for Latur District using GIS, Remote Sensing, seven spatial conditioning parameters, and groundwater-well observations. The selected parameters were lithology, geomorphology, soil, slope, drainage density, lineament density, and LULC. All layers were standardized to a common 30 m grid before RF modelling. A total of 109 wells were used in the study. The well observations were classified using a 12.8 m mean-depth threshold, and the dataset was divided into 76 training wells and 33 independent validation wells. The reported ROC-AUC value of 0.8192 shows good model performance for distinguishing groundwater occurrence and non-occurrence conditions. The final map classified the district into very-low, low, moderate, high, and very-high groundwater-potential zones. Drainage density and slope were the strongest fitted parameters in the RF model, while soil, LULC, lineament density, lithology, and geomorphology provided additional environmental information. The final GWPZ map should be used as a planning and screening tool. It can support groundwater exploration, recharge planning, conservation, and monitoring. However, site-specific decisions should be confirmed through borehole data, pumping tests, groundwater-level measurements, geophysical surveys, and water-quality assessment. Future research should incorporate spatial rainfall data, recharge indicators, aquifer transmissivity, weathering thickness, geophysical resistivity, and spatial cross-validation.

References

[1] Arulbalaji, P., Padmalal, D., & Sreelash, K. (2019). GIS and AHP techniques based delineation of groundwater potential zones: A case study from Southern Western Ghats, India. Scientific Reports, 9, 2082. [2] Jerin Joe, R. J., Pitchaimani, V. S., Kaliraj, S., Abishek, S. R., Promilton, A. A. A., & Karuppannan, S. (2025). Spatial modelling of groundwater potential zones in the Neyyar Basin using machine learning and morphometric analysis. Scientific Reports, 15, 36171. [3] Kalantar, B., Al-Najjar, H. A. H., Pradhan, B., Saeidi, V., Halin, A. A., Ueda, N., & Naghibi, S. A. (2019). Optimized conditioning factors using machine learning techniques for groundwater potential mapping. Water, 11(9), 1909. [4] Lee, S., Hyun, Y., Lee, S., & Lee, M. J. (2020). Groundwater potential mapping using remote sensing and GIS-based machine learning techniques. Remote Sensing, 12(7), 1200. [5] Madani, A., & Niyazi, B. (2023). Groundwater potential mapping using remote sensing and Random Forest machine learning model: A case study from Lower Part of Wadi Yalamlam, Western Saudi Arabia. Sustainability, 15(3), 2772. [6] Nguyen, N. N. T., Thunyawatcharakul, P., Ngu, N. H., & Chotpantarat, S. (2022). Global review of groundwater potential models in the last decade: Parameters, model techniques, and validation. Journal of Hydrology, 614, 128501. [7] Rahmati, O., Pourghasemi, H. R., & Melesse, A. M. (2016). Application of GIS-based data-driven Random Forest and maximum entropy models for groundwater potential mapping: A case study at Mehran Region, Iran. CATENA, 137, 360–372. [8] Razavi-Termeh, S. V., Sadeghi-Niaraki, A., & Choi, S. M. (2019). Groundwater potential mapping using an integrated ensemble of three bivariate statistical models with Random Forest and logistic model tree models. Water, 11(8), 1596.

Copyright

Copyright © 2026 Suchivrutta R. Kamble, Dr. Ganesh K. Patil. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.