Divergence of apparent and intrinsic snow albedo over a season at a sub-alpine site with implications for remote sensing

Abstract

Abstract. Intrinsic albedo is the bihemispherical reflectance independent of effects of topography or surface roughness. Conversely, the apparent albedo is the reflected radiation divided by the incident and may be affected by topography or roughness. For snow, the surface is often rough, and these two optical quantities have different uses: intrinsic albedo is used in scattering equations whereas apparent albedo should be used in energy balance models. Complementing numerous studies devoted to surface roughness and its effect on snow reflectance, this work analyzes a time series of intrinsic and apparent snow albedos over a season at a sub-alpine site using an automated terrestrial laser scanner to map the snow surface topography. An updated albedo model accounts for shade, and in situ albedo measurements from a field spectrometer are compared to those from a spaceborne multispectral sensor. A spectral unmixing approach using a shade endmember (to address the common problem of unknown surface topography) produces grain size and impurity solutions; the modeled shade fraction is compared to the intrinsic and apparent albedo difference. As expected and consistent with other studies, the results show that intrinsic albedo is consistently greater than apparent albedo. Both albedos decrease rapidly as ablation hollows form during melt, combining effects of impurities on the surface and increasing roughness. Intrinsic broadband albedos average 0.056 greater than apparent albedos, with the difference being 0.052 in the near infrared or 0.022 if the average (planar) topography is known and corrected. Field measurements of spectral surface reflectance confirm that multispectral sensors see the apparent albedo but lack the spectral resolution to distinguish between darkening from ablation hollows versus low concentrations of impurities. In contrast, measurements from the field spectrometer have sufficient resolution to discern darkening from the two sources. Based on these results, conclusions are as follows: (1) impurity estimates from multispectral sensors are only reliable for relatively dirty snow with high snow fraction; (2) a shade endmember must be used in spectral mixture models, even for in situ spectroscopic measurements; and (3) snow albedo models should produce apparent albedos by accounting for the shade fraction. The conclusion re-iterates that albedo is the most practical snow reflectance quantity for remote sensing.

Does higher spatial resolution improve snow estimates?

“…For example, from water years 2017-2019, the Root Mean Squared Error (RMSE) for MODIS-SPIReS, calculated using the best value for a 3×3 neighborhood around the validation site, is 2.3% with no bias (Table 2). These albedo errors are similar to the accuracy of the HDRF surface reflectance products, evaluated over dark targets (Vermote et al, 2016;Bair et al, 2022). These improvements in remotely sensed snow albedo over previous assessments, showing RMSE values of 4.6 to 4.8% with 0.7-1.3% bias for MODIS (Bair et al, 2019;Bair et al, 2021), come from improved cloud snow discrimination filters and adjustments to thresholds such as the minimum grain size for dirty snow (Section III-J of Bair et al, 2021).…”

“…Mountain (e.g., Bair et al, 2022), only 23 km from Mount Lyell, the highest point in the Tuolumne River Basin. For example, from water years 2017-2019, the Root Mean Squared Error (RMSE) for MODIS-SPIReS, calculated using the best value for a 3×3 neighborhood around the validation site, is 2.3% with no bias (Table 2).…”

Landsat, MODIS, and VIIRS snow cover mapping algorithm performance as validated by airborne lidar datasets

“…Spectral unmixing approaches have already been shown to be robust when transitioning 620 between sensors like MODIS and VIIRS (Rittger et al, 2021a). Based on this work, we expect SPIReS to have similar performances on VIIRS and expect these spectral unmixing algorithms to be insensitive to bandpass differences among other sensors such as Sentinel 2a/2b (Bair et al, 2022) and the upcoming Thermal infraRed Imaging Satellite for Highresolution Natural resource Assessment (TRISHNA) mission. Global standard MODSCAG and VIIRSCAG products are currently being undertaken by the NSIDC DAAC, and SPIReS MODIS will be produced operationally for North 625 America, Greenland, and High-Mountain Asia as part of Snow Today at NSIDC (Rittger and Raleigh, 2022).…”

“…Second, it is difficult to find a spectral unmixing solution for fSCA=1 that is substantially better than the solutions available for slightly lower snow cover fractions 735 when dealing with many pixels. This is especially true when shading is present in the pixels, as shading has been shown to significantly lower the apparent albedo and snow cover reflectance (Bair et al, 2022).…”

Evaluation of snow processes over the Western United States in E3SM land model

“…There are some inconsistencies between STC-MODSCAG and SPIReS (Figure 3-4), due to the different algorithms and data processing (e.g., interpolation and filtering). Although the physically-based STC-MODSCAG and SPIReS provide higher quality unbiased fsno estimates than the MOD10A1 snow product based on empirical algorithms against field measurements across different forest cover, snow cover, snow climate and viewing angles (Bair et al, 2021c;Rittger et al, 2013;Stillinger et al, 2022), the issues of reflectance errors, one to many problems intrinsic to spectral unmixing, cloud contamination, topographic shadows, sun-sensor geometric effects, and the impacts of forest cover can still affect their reliabilities (Bair et al, 2021b;Raleigh et al, 2013;Stillinger et al, 2022). These issues can also affect the accuracy of extracted snow phenology (Section 2.4).…”