Physics-Informed Machine Learning for Agricultural Drought Prediction in Vidarbha, Maharashtra: A Multimodal Geospatial Fusion Framework for ESG

Anirudh Khajuria

Abstract: Vidarbha is one of the most drought-stricken areas of South Asia. This semi-arid agricultural belt in Maharashtra, India is characterised by extreme deficits in annual rainfall (up to 55% in some districts) and ongoing depletion of the groundwater supply, which poses a significant threat to the livelihoods of millions of small farmers. Drought monitoring methods currently used in traditional systems are based on large statistical indices that do not reflect the multi-scale localised variations in soil moisture, crop stress and aquifer depletion at the scale of the farm. In this work we present a Physics-Informed Neural Network (PINN) algorithm that integrates different forms of geospatial data (Sentinel-2 satellite imagery (at 10m resolution), India-WRIS piezometer time-series data, and IMD gridded rainfall fields) through a novel water-balance-constrained loss function, in an effort to produce accurate estimates of soil moisture and drought indices, at a recent time. Using this approach, we validate the model across five vulnerable districts in Vidarbha (Yavatmal, Amravati, Akola, Buldhana and Washim) over the time frame of 2018-2026 at the tehsil level. By providing a framework for transferability of this method, we anticipate that this approach can be adopted globally using large datasets acquired from the ERA5-Land and Copernicus Open Access Hub.

Keywords: Physics-Informed Neural Networks, Sentinel-2, Drought Stress, Vidarbha, Groundwater Depletion, Water Balance, Remote Sensing, NDVI, NDWI, India-WRIS

How to Cite?: Anirudh Khajuria, "Physics-Informed Machine Learning for Agricultural Drought Prediction in Vidarbha, Maharashtra: A Multimodal Geospatial Fusion Framework for ESG", Volume 15 Issue 6, June 2026, International Journal of Science and Research (IJSR), Pages: 1581-1593, https://www.ijsr.net/getabstract.php?paperid=SR26629131639, DOI: https://dx.doi.org/10.21275/SR26629131639

Download Citation: APA | MLA | BibTeX | EndNote | RefMan