This project aims to address two broad and general challenges related to assimilation of remotely sensed data into terrestrial hydrology models: 1. Incorrect or incomplete specification of model and…

This is a grant award from the National Aeronautics and Space Administration (NASA) to the University of Alabama Office For Sponsored Programs Division, located in Tuscaloosa, Alabama. The $184,520.87 grant is for research aimed at improving the integration of remote sensing data into terrestrial hydrology models. Key aspects include developing a machine learning-based data assimilation approach to address issues with model and retrieval biases, as well as providing a pathway from data assimilation to improved model understanding and development. The research will leverage the Structure for Unifying Multiple Modeling Alternatives (SUMMA) land surface model integrated into NASA's Land Information System. No set-aside designation is mentioned. This grant aligns with NASA's Program Element 2.3 on Multivariate Hydrological Simulation and addresses fundamental challenges in terrestrial hydrologic data assimilation.

This project aims to address two broad and general challenges related to assimilation of remotely sensed data into terrestrial hydrology models: 1. Incorrect or incomplete specification of model and retrieval biases and statistical error distributions causes information loss during DA. 2. There currently exists no systematic path from DA to improved process understanding through either process-level model diagnosis or model improvement. To improve the use of remote sensing information in land models, we propose to apply a machine-learning-based DA approach to address these two issues. A kernel-based non-parametric correction factor is added to both the model state transition function and the retrieval operator in place of the statistical error distributions typically required by DA. These correction factors will be estimated during DA using expectation-maximization. This approach avoids the use of an a priori bias correction like CDF matching, which removes valuable information that retrievals might otherwise contain about systematic model biases. This approach also removes the need for a priori and/or parametric specification of model and retrieval error distributions, which can be difficult to estimate. Instead of a prescribed error distribution, our DA approach learns a correction factor to the dynamic behavior of the terrestrial hydrology model. This results not only in adaptions to model structure that result in persistent improvements to model-simulated variables outside of the period where assimilation observations are available (examples of this are given in the proposal), but also provides a tool for assessing process-level error structures in the model. To capitalize on this latter aspect of our DA approach, we also propose a quantitative strategy for process-level model diagnostics, to measure which individual modeled state couplings change the most due to application of structure-updating DA. This provides a direct pathway for information gained from data assimilation to guide future model developments. This proposal is directly relevant to program element 2.3 multivariate hydrological simulation, and as noted above, addresses fundamental problems in terrestrial hydrologic DA. Our expectation-maximization DA strategy will be demonstrated on the structure for unifying multiple modelling alternatives (summa), which is a process-flexible land surface model that has recently been integrated into the NASA land information system. This will allow US to apply our assimilation approach to many different land model process configurations. We will use a suite of in situ (fluxnet) and remote sensing (smap, modis, gome-2, alexi) experiments to specifically investigate model realism, and to identify systematic structural errors in certain biogeophysical process representations contained in modern terrestrial hydrology models (e.g., stomatal resistance, root water uptake, canopy infiltration, soil characteristic functions, etc.). This project responds directly to a recent