The research, led by Manika Mallick of Tripura University together with Y. V. Krishnaiah, Vajana Mondal and Kausik Panja, focuses on Jalpaiguri, a district straddling the foothills of the eastern Himalayas. The region receives some of the heaviest rainfall in India, yet its groundwater story is far more complicated than the drenching monsoons suggest. Water availability below ground depends on an intricate interplay of geology, topography, soil properties, land cover and the rhythm of recharge and extraction. When those factors shift, as they are doing under a warming climate and expanding agriculture, the consequences can appear suddenly, in the form of falling water tables, failing wells and stressed ecosystems.

To map where groundwater is most likely to be found and most easily extracted, the team assembled an unusually comprehensive set of fifteen environmental and hydrological factors. These included the region’s geology and geomorphology, the curvature of the land profile, the density of geological lineaments where fractured rock can channel water, elevation and slope, the depth to the water table, drainage density, the stream power index, the topographic wetness index, rainfall, soil texture, soil moisture, the normalised difference vegetation index known as NDVI, and land-use and land-cover classifications derived from satellite imagery. Each of these layers tells part of the story. Steep slopes shed water rather than absorbing it; coarse soils let rainfall percolate downward; dense drainage networks can signal either abundant surface water or rapid runoff that never recharges the aquifer.

The researchers then applied two fundamentally different analytical philosophies to these layers. The first was a multi-criteria decision-making approach built on the multi-influencing factor model, a technique in which experts assign relative weights to each factor based on its known influence on groundwater occurrence. This method, long a staple of groundwater prospecting in data-scarce regions, has the advantage of transparency: every weight can be inspected and debated. The second approach was a random forest algorithm, a machine learning method that trains an ensemble of decision trees on real-world evidence, in this case the observed presence or absence of groundwater at known locations, and learns the relationships between the fifteen factors and groundwater occurrence without any human-imposed weighting scheme.

The results of the two approaches were strikingly similar in their broad outlines. According to the multi-influencing factor model, 46.87 percent of the study area falls within high to very high groundwater potential zones, while the random forest algorithm placed 44.6 percent of the district in those same categories. In both maps, the most promising zones cluster in the alluvial plains where permeable sediments, gentle slopes and abundant recharge combine to create productive aquifers. But similarity at the regional scale does not settle the question of which method a water manager should actually trust, and for that the team turned to formal validation.

Model performance was evaluated using the receiver operating characteristic curve, a standard statistical tool that measures how well a model separates true positives from false positives across all possible thresholds. The models were also validated against independent field evidence in the form of dug well depth data collected with the cooperation of local communities. Both models performed credibly, but the machine learning approach clearly outpaced its expert-weighted rival. The random forest algorithm achieved an accuracy of 88 percent, compared with 79 percent for the multi-criteria method. That nine-point gap may sound modest, but in a district where millions of litres of water are allocated on the basis of such maps, the difference between an 88 percent and a 79 percent reliable prediction translates into wells that either strike water or run into dry sediment.

The superiority of the random forest model is consistent with a broader trend in the geospatial sciences. Machine learning algorithms excel precisely where expert judgment struggles: in high-dimensional problems where many factors interact in nonlinear ways, and where the relative importance of a variable such as soil moisture or profile curvature shifts from one landscape to the next. An expert weighting scheme must average across the entire district, while a trained random forest can capture the fact that slope matters enormously in the Himalayan foothills but very little on the flat plains. The result is a map that reflects the actual behaviour of the hydrological system rather than a generalised template of it.

Yet the study does not stop at identifying where water is abundant. Its most consequential contribution is a vulnerability assessment that layers a second dimension onto the potential maps: how much the water table fluctuates after the monsoon season. Using the innovative trend analysis technique, a statistical method capable of detecting trends in hydrological time series without requiring strict assumptions about data distribution, the researchers quantified post-monsoon groundwater level fluctuations across the district and then combined this information with the groundwater potential zonation. The logic is intuitive but powerful. An area with high groundwater potential that also experiences dramatic seasonal swings in water level is a very different management proposition from an area with moderate potential and a stable water table.

The combined analysis produced the study’s most sobering figure: 59.79 percent of Jalpaiguri district is vulnerable to groundwater stress, owing to the pairing of moderate to low groundwater potential with high post-monsoon water level fluctuations. In practical terms, nearly six out of every ten square kilometres of the district host aquifers that are either naturally limited or seasonally unstable, or both. For the farmers, tea estates and rural households that depend on these aquifers, that classification signals a heightened risk of wells failing in the dry months and of extraction outpacing recharge in the years ahead. The finding challenges the complacency that abundant monsoon rainfall can breed, demonstrating that a water-rich climate does not guarantee a water-secure future.

The implications reach well beyond one district in West Bengal. The sub-Himalayan belt stretches across northern India and into neighbouring countries, and its aquifers face parallel pressures everywhere: intensifying agriculture, growing populations, and shifting monsoon patterns under climate change. The framework developed in this study, which fuses geospatial data, competing modelling paradigms and trend analysis into a single integrated assessment, offers a replicable template for the region. Because all of the underlying datasets are drawn from openly accessible national and international repositories, including satellite imagery from the USGS Earth Explorer, geological layers from the Geological Survey of India’s Bhukosh portal, groundwater depth records from the India-WRIS system, soil maps from the National Bureau of Soil Survey and Land Use Planning, and rainfall grids from the India Meteorological Department, the approach can be reproduced by any regional authority with basic computational resources.

For policymakers, the study delivers a concrete planning instrument. The high-potential, low-vulnerability zones identified on the maps are candidates for sustainable intensification of groundwater use, while the vulnerable areas flagged by the trend analysis demand demand-side management, artificial recharge structures and community-based water governance. The authors emphasise that sustainable management requires both regional-scale strategic planning and governance rooted in local communities, and their maps provide the spatial scaffolding on which such governance can be built. In an era when groundwater depletion is quietly emerging as one of the defining resource challenges of the century, this fusion of satellite data, machine learning and hydrological trend analysis shows how the tools of the digital age can be turned toward one of humanity’s oldest problems: finding water, and learning to live within its means.

Subject of Research: Mapping groundwater potential zones and vulnerability in the sub-Himalayan belt of West Bengal, India, using geospatial analysis, machine learning and trend analysis

Article Title: Integrated assessment of groundwater potential and vulnerability for sustainable groundwater management in the sub-Himalayan belt of West Bengal, India: A novel geospatial, machine learning, and trend analysis framework

Article References: Mallick, M., Krishnaiah, Y. V., Mondal, V., & Panja, K. (2026). Integrated assessment of groundwater potential and vulnerability for sustainable groundwater management in the sub-Himalayan belt of West Bengal, India: A novel geospatial, machine learning, and trend analysis framework. Hydrogeology Journal. https://doi.org/10.1007/s10040-026-03160-w

Image Credits: AI Generated

DOI: 10.1007/s10040-026-03160-w

Keywords: groundwater potential zones, machine learning, random forest, multi-criteria decision-making, groundwater vulnerability, Jalpaiguri, West Bengal, sub-Himalayan belt, innovative trend analysis, remote sensing, GIS, sustainable groundwater management