DAPE: Data Assimilation and Model Parameterization for Earth Science

nbsp;nbsp;We will develop methods and a software framework for data assimilation and parameterization (DAPE) using Unsupervised and Physics-Informed Machine-Learning (UML and PIML). We will apply DAPE to parameterize processes represented by the Variable Infiltration Capacity (VIC) Hydrological Model. VIC is critical for simulating surface/subsurface water flow in several Earth System Models (ESMs).DAPE will be capable of assimilating and merging discrete in-situ (e.g., weather/gage stations, monitoring wells) and continuous (e.g., satellite/airborne/geophysics images) datasets. DAPE will apply ML to generate high-resolution spatiotemporal mappings of the meteorologic, surface water, and groundwater parameters (e.g., precipitation, evapotranspiration, infiltration, surface runoff, soil moisture, temperature, and heat flux). The mappings will be applied to parameterize the surface water and groundwater flow in the VIC model. The parameterization will combine analyses of field measurements, remote data, and modeling outputs.In the next project phases, we will apply DAPE to develop a PIML model representing all energy/mass balance processes simulated in VIC. We will also address VIC limitations related to lateral surface-water/groundwater flow. We will also explore alternative models (PRMS, Noah-LSM, Noah-MP). We will evaluate the impacts of input and conceptual model uncertainties on the predictive uncertainties. We will also analyze NOAA/NASA satellite images representing land surface water and energy dynamics. We will also explore how the ML analyses of these datasets will impact the characterization, parameterization, and prediction of Earth system processes, with an emphasis on floods and droughts. These analyses will account for anthropogenic activities and climatic changes. We will apply our work to update the Livneh dataset.nbsp;