Using the Budyko framework to evaluate anthropogenic impacts on long-term surface water partitioning in India

Parsimonious physics based conceptual models, such as parameterized forms of the Budyko curve, are useful in constraining the uncertainty in estimates of long-term freshwater availability in data limited regions. In order to enable predictions in data limited regions, it is necessary to regionalize the Budyko parameter, i.e., relate it to catchment characteristics. This also helps to identify the effects of various physio-climatic and socio-economic drivers on long-term surface water balance. Here, we apply data space splitting based machine learning algorithms to disentangle regionally varying controls on Fu's parameter ω, an often-used variant of the Budyko curve. The classification and regression tree (CART) algorithm and its associated ensemble based random forest (RF) technique are used to analyse the relative importance of climate, vegetation, topography, geologic, and anthropogenic characteristics. These algorithms are applied to 534 regional divisions across India with 33 representative characteristics curated from literature. The CART decision tree revealed the existence of hierarchical scale dependent controls of vegetation, climate and soil composition. Controls exerted by long-term precipitation (MAP) and long-term temperature (MAT) on surface water balance were found contingent upon the extent of short rooted vegetation (ShortRooted). Soil characteristics played a crucial role in relatively drier regions with considerable coverage of short rooted vegetation. Effects of influencing factors was also found dependent on regional conditions. RF correctly predicted ω classes for 63.9 % assumed ungauged regions. Both CART and RF identified ShortRooted, MAT, population density and MAP as the main characteristics governing the variation of ω. Our results highlight the significance of human impacts on long-term surface water partitioning as population density emerges as an important factor comparable in its influence to climate and vegetation factors.