Quantifying Snow–Ground Backscatter Uncertainty: A Bayesian Approach Using Multifrequency SAR and In-Situ Observations

Highlights: This study validates the stepwise approach to estimate the snowpack bottom boundary condition by splitting snow–ground backscatter from volume backscatter in active snow microwave retrievals. The parsimonious three-parameter QHN frequency-independent soil reflectivity model is best suited for active snow microwave. There is a substantive dependence of total backscatter on ground properties at the X-band (increasing ground surface roughness reduced the simulated backscatter by ~1.5 dB across the tested range, and increasing the specular-to-total reflectivity ratio (STRR) produced an additional ~1.0 dB decrease), while there was negligible to very weak sensitivity to ground parameters at the Ku-band. The retrieval sensitivity to STRR is minimized in the 0.6–0.7 range and it can be fixed at 0.65 without having discernible impact. Backscatter uncertainty at the snow–ground interface preferentially impacts the estimation of soil dielectric properties. Accurate estimation of snowpack microwave backscatter is critical for retrieving key physical properties of snow, such as snow depth (SD) and snow water equivalent (SWE), typically modeled using radiative transfer models (RTM). Among the various sources of uncertainty in RTM simulations, snow–ground reflectivity—used as a boundary condition—plays a critical role in influencing the accuracy of simulated backscatter. This study leverages high-resolution X- and Ku-band synthetic aperture radar (SAR) backscatter aircraft measurements using SWESARR and SnowSAR from NASA’s SnowEx campaigns, co-located with in situ snow pit observations in Grand Mesa, Colorado, and uses a Bayesian MCMC parameter optimization model with RTM framework to estimate the key ground parameters such as surface roughness, moisture content, and specular-to-total reflectivity ratio (STRR) governing the estimation of the snow–ground reflectivity and quantify the uncertainties associated with them. At the X-band, increasing ground surface roughness reduced the simulated backscatter by ~1.5 dB across the tested range, increasing the STRR produced an additional ~1.0 dB decrease while the dielectric properties of the ground are highly sensitive to the moisture content of frozen soil, and increasing the moisture content even by 2% increased the backscatter by 2–3 dB. The retrieval sensitivity to the STRR is minimized in the 0.6–0.7 range and it can be fixed at 0.65 without having discernible impact. The Bayesian inversion reveals that the extreme parameter values act as diagnostic indicators of unmodeled complexity rather than retrieval failures, with representativeness error often dominating over instrument noise. The study provides a robust methodology for the estimation of the snow–ground backscatter boundary condition for forward modeling, ultimately aiding SWE and SD retrieval from active microwave observations. While this study relied on Grand Mesa, the framework developed here is general and, along with the model uncertainty, is directly transferable and broadly applicable to other snow-dominated mountain regions where active microwave observations can be used for snowpack monitoring.

Duke Scholars

Author Ana P. Barros Civil and Environmental Engineering