Quantitative Precipitation Estimates (QPE) from merged rain gauge and radar measurements have become widely available in the last two decades. The errors associated with these products are yet to be fully understood, especially in complex terrain where ground clutter and overshooting artifacts are significant and vary in space and time depending on the storm and underlying synoptic conditions. The location and timing of precipitation in addition to rainfall intensity and duration are critical to the simulation of flood response in headwater basins. This work proposes a generalizable Physics-guided Artificial Intelligence (PAI) framework for QPE error modeling. First, QPE error climatology derived from the hydrologic Inverse Rainfall Correction (Liao & Barros, 2022) to historical floods in selected headwater basins is analyzed to identify dominant precipitation regimes. Second, for each precipitation regime, a Multilayer Perceptron (MLP) error prediction model is trained using event-specific precipitation metrics at hourly scale as input, and subsequently used to predict estimation errors for various QPE products. The corrected QPE can then be used for hydrologic simulations and flood nowcasting. The PAI framework is demonstrated in the Southern Appalachian Mountains using the 57 largest floods over 2008-2017. The Probability Distribution Function of predicted precipitation errors follows a Gaussian-like distribution but varies significantly between cold and warm season events, while the spatial distribution is inextricably connected to basin geomorphology. On average, large improvements on hourly KGE from -0.5 to 0.4 are achieved, and the peak flood error is reduced by 70%, with distinctively better results for cold season events.

Yueqian Cao1, and Ana P. Barros21. Department of Civil and Environmental Engineering, Duke University, Durham, NC, USA2. Department of Civil and Environmental Engineering, University of Illinois at Urbana-Champaign, Urbana, IL, USA* Correspondence: [email protected]: A radar observing system simulator consisting of a coupled snow hydrology (MSHM) and radiative transfer model (MEMLS) was modified to include vegetation contributions to the total backscatter from the ground-snow-vegetation system. Vegetation parameters were estimated from airborne SnowSAR (X- and Ku-band) and Sentinel-1 (C-band) measurements in Grand Mesa (flat topography) and Senator Beck Basin (steep topography) by solving the inverse problem via simulated annealing. Physics-based constraints were imposed to address indeterminacy with good results, which highlights that the forward-inversion system accounting for complex multiple scattering within the ground-snow-vegetation system reliably regulated compensation effects of vegetation and snow-ground interface, including simulating observed background backscatter under snow-free conditions. The proximal goal of this study is to quantify the integrated effect of complex multiple scattering within the ground-snow-vegetation system toward isolating volume scattering from subcanopy snowpack, and subsequent retrieval of snowpack properties such as snow water equivalent (SWE). The stretch goal is to develop a vegetation correction to expand the operational utility of radar remote sensing of snow in the boreal forests at northern latitudes. The proposed approach has high operational utility for retrieving large-scale SWE from satellite-based SAR measurements.Keywords: forward-inversion system; MEMLS; simulated annealing; SAR; vegetation heterogeneity

An uncalibrated distributed multiphysics snow model driven by downscaled weather forecasts (30-m, 15-min) was implemented as a Radar Observing System Simulator (ROSS) in Senator Beck Basin (SBB), Colorado to elucidate topographic controls on C-, X-and Ku-bands active microwave sensing of mountain snowpacks. Phase-space maps of time-evolving grid-scale ROSS volume backscatter show the accumulation branch of the backscatter-snow water equivalent (σ-SWE) hysteresis seasonal loop that is the physical basis for radar retrieval (direct inference) of SWE and snowpack physical properties. ROSS results with snow-ground scattering correction inferred from snow-free conditions capture well the seasonal march of Sentinel-1 C-band backscatter, including spatial patterns tied to elevation, slope, and aspect. Root Mean Square Deviations (RMSDs) do not exceed ±3.2 dB for ripening snowpacks in early spring and ±2.4 dB for dry snowpacks in the accumulation season when the mean absolute bias is < 1 dB for all land-cover types with topographic slopes 30°. Grid-point RMSDs are attributed to the underestimation of snowfall on upwind slopes compounded with forecast errors for the weather near the ground. Like Sentinel-1, ROSS backscatter fields exhibit frequency-independent single-scaling behavior within the 60-150 m scale range for dry snowpacks in the accumulation season, while frequency-dependent scaling behavior emerges in the ablation season. This study demonstrates skillful physical modeling capabilities to emulate Sentinel-1 observations in complex terrain. Conversely, it suggests high readiness to retrieve snow mass and snowpack properties in mountainous regions from radar measurements at high-spatial resolutions enabled by SAR technology.