Uncertainty Propagation of Snowfall and Atmospheric Forcings in Modeled Seasonal Snowpack Evolution during SnowEx 2017 — Implications for Microwave Remote Sensing of Snow and Data Assimilation

Direct assimilation of remotely sensed radiances in lieu of retrieved geophysical variables has been successful in operational numerical weather prediction, which is attributed to the lower uncertainty and less ambiguity of direct radiance measurements compared to indirect estimates of nonlinear processes. This study's objective is to characterize the uncertainty propagation of snowfall and atmospheric forcings in simulated seasonal snowpack evolution of a coupled snow hydrology — radiative transfer model to determine minimum requirements of potential remote sensing measurements from space. To this end, an ensemble including 52 distinct snowfall and atmospheric forcing members was established from the High-Resolution Rapid Refresh (HRRR) model output (3km, hourly) and European Centre for Medium-Range Weather Forecasts (ECMWF) reanalysis dataset (5 km, hourly) at high elevations in Grand Mesa, Colorado. First, the snowpack predictions (SWE, snow depth, snow density, temperature and grain size profiles) from the Multilayer Snow Hydrology Model (MSHM) were evaluated against the SnowEx campaign observations in Colorado throughout February 2017 to assess model physics and sensitivity to snowfall and atmospheric forcing uncertainty. The results show that the MSHM captured well the vertical structure, snow accumulation and melt processes over Grand Mesa with spatial variability in field measurements generally consistent with the MSHM ensemble spread. Second, the MSHM was coupled to the Microwave Emission Model of Layered Snowpacks (MEMLS) to examine the sensitivity of C- and L-band brightness temperatures (Tb) and dual-polarization backscatter coefficients (σ) to the snowpack ensemble. The results indicate that the forward radiative transfer simulations of σ and Tb exhibit highly nonlinear sensitivity to snowpack properties relied on atmospheric forcings, suggesting that data assimilation could disambiguate significantly model results under uncertainty. The behavior of the ECMWF-driven members of the coupled MSHM-MEMLS ensemble is governed by the severe snowfall underestimation errors; while the HRRR-driven ones capture uncertainty tied to spatial variability across Grand Mesa and inform the lower requirements for microwave remote-sensing measurements of snow.