Influences of watershed characteristics on long-term annual and intra-annual water balances over India

Evaluation of the engrossment of watershed surface characteristics on partitioning of precipitation to runoff and evapotranspiration is key to inspect the availability of water at watershed scale. It is more evident in the cases of ungauged watersheds. The present study develops models using multiple linear regression method and machine learning techniques (ANN: Artificial Neural Network and RVM: Relevance Vector Machine) over 793 (25 major river basins and 768 watersheds across India) to estimate the watershed parameter ‘ω’ (in Fu’s Budyko based equation) that represents intrinsic watershed attributes. In addition, seasonality factor is incorporated in the model due to intra-annual variability in vegetation across India. The models attempt to explain the intricate relationship between vegetation alterations and regional water balance. It is seen that the ANN and RVM models have performed better in estimating ω, than the MLR (Multiple Linear Regression) models. In addition, NDVI has shown more engagement in explaining the partitioning process of water in intra-annual low NDVI period compared to high NDVI period. We have also found the present models to be more accurate than the previously developed Budyko based methods in predicting ω. The newly improved models have closely imitated the intrinsic basin attributes and enhanced the functionality of Budyko framework in estimation of water availability, which would play a crucial role in assessment of hydrology of ungauged watersheds of India.