How Can a Hierarchical Bayesian Approach Bridge the Gap Between Multi-Source Remote Sensing Data and Hydrological Models?

Integrating multi-source remote sensing data with hydrological models presents significant challenges, primarily due to mismatches in spatial resolution between satellite observations and models, and spectral inconsistencies between model outputs and observations. These discrepancies stem from satellite mission sampling, the conversion of measured signals to the variables of interest, and processing steps like filtering to reduce noise. For instance, Terrestrial Water Storage (TWS) data from the Gravity Recovery and Climate Experiment (GRACE) and its follow-on mission (GRACE-FO) represent a vertical summation of all water stored on land, with a footprint of several hundred kilometers. Another example is Surface Soil Moisture (SSM) data from passive and active remote sensing missions, such as the ESA Climate Change Initiative (CCI), which reflects the moisture of the top few centimeters of soil at a spatial resolution of 25 km.

While large-scale hydrological models now target kilometer-level spatial resolution, their ability to represent climate-driven and anthropogenic changes remains limited. In this study, we propose a hierarchical Bayesian approach to merge GRACE/GRACE-FO TWS changes and ESA CCI’s SSM with the water storage outputs of a high-resolution hydrological model, while accounting for uncertainties in both observational data and model simulations. Our methodology aims to downscale GRACE/GRACE-FO observations and achieve vertical separation of GRACE/GRACE-FO TWS components. By refining the spatial and spectral alignment between observations and model results, this approach enhances the representation of individual water storage components, such as soil water and groundwater storage changes.

The proposed method involves several key steps to ensure data consistency within the multi-sensor fusion. First, all input datasets, including hydrological model outputs and remote sensing observations, are filtered to align their spectral signal contents. Then, a hierarchical Markov Chain Monte Carlo (MCMC) algorithm is applied to constrain all modeled and filtered TWS with GRACE/GRACE-FO and the SSM datasets. This is achieved by computing a temporal scaling factor that aligns the individual water storage compartments of the hydrological model with both observations. Finally, the residuals between filtered and unfiltered model outputs are incorporated to refine TWS estimates and enhance the downscaling process. The implementation and validation of the proposed approach are demonstrated using the W3RA hydrological model at a 10 km resolution over Europe. Model performance is evaluated by comparing updated groundwater and topsoil water estimates with other model outputs such as WGHM and independent observations. Results highlight the effectiveness of the hierarchical Bayesian method in resolving spectral and spatial mismatches. This study underscores the potential of advanced Bayesian techniques to enhance the utility of remote sensing data in hydrological applications.