Improving hydrological prediction across different spatial scales with observation-informed modeling

Abstract:

Physical modeling approaches are important tools for developing spatially and temporally continuous estimates of the terrestrial water cycle components. In addition to the limitations imposed by uncertainties in boundary conditions and model parameters, representing the heterogeneity and impacts of human management across different spatial scales are significant challenges in terrestrial hydrological modeling. The availability of remote sensing measurements offers the opportunity to mitigate some of these limitations, which are typically incorporated within physical models through data assimilation. Though assimilation studies have demonstrated the beneficial impact of remote sensing measurements both for improving the representation of catchment scale processes, there are significant challenges related to assimilation strategies, limitations in model formulations, and observational data processes that limit the potential utility of the remote sensing measurements. In this presentation, results from recent studies that highlight these challenges will be discussed. The information utilization within data assimilation approaches will be contrasted with data driven techniques based on machine learning. The use of nonparametric metrics founded in information-entropy methods that are more efficient for characterizing the information utilization efficiency will be discussed. The presentation argues for the need for hybrid approaches that integrate both physical and data-driven modeling to improve the representation of hydrologic processes across spatial scales.