Assessing the power of non-parametric data-driven approaches to analyse the impact of drought measures

Commonly used hydrological models often require much implementation and computational efforts, while their accuracy is limited, especially in areas with strong anthropogenic controls. In this study, two alternative, nonparametric data-driven approaches are tested to supplement existing hydrological models for the assessment of water scarcity along rivers and the potential impact of mitigation strategies. These approaches can assess the water availability at a regional scale in a spatially detailed way, taking into account both the flow regulation effects and other anthropogenic influences as reflected in river flow observations. After application to a network of rivers in Flanders (Belgium), a leave-one-out cross-validation shows that the data-driven approaches are promising, with good NSE values on daily river flows of 0.58-0.65 and high coefficient of determination of 0.72-0.76. The overall performance in representing the relative changes of flows in space and time is similar to that of two state-of-the-art hydrological models.