The Budyko curve, relating a catchment’s water and energy balance, provides a useful tool to analyse how humans may impact long-term runoff. Often a parametric form of the curve, the Fu’s equation, is used to represent the relationship between a catchment’s long-term water partitioning behaviour and climate. Fu’s parameter ω, typically derived from observed climate and runoff data, can further be related to catchments’ physio-climatic characteristics for understanding the main drivers of its water balance. We employ this approach to quantify the impact of human interventions on surface water partitioning across India. We explore the relationship between ω and a curated database of 33 physio-climatic and socio-economic characteristics for 534 regional divisions of India using two related machine learning algorithms: classification and regression trees (CART) and random forest (RF). Both algorithms diagnose the hierarchy of representative vegetation, climate, soil, land use land cover, topography and anthropogenic controls. RF validates CART output while also providing a data-driven model to estimate ω in assumed data-scarce regions, enabling us to assess the value of this dataset for predictions in ungauged basins. The most relevant characteristics controlling ω based on CART and RF analysis were: long-term temperature, percentage of short rooted vegetation, population density, and long-term precipitation. RFs were able to correctly predict the classified ω for 63.9 % of assumed ungauged regions. We found that population density’s influence on ω was comparable to that of climate and vegetation, highlighting the role of humans in controlling long-term surface water partitioning variability across India.