Assessment of methods for mapping snow albedo from MODIS

We compare five daily MODIS-derived snow albedo products to terrain-corrected, in situ data from sites in California and Colorado, USA, and to snow albedo derived from airborne hyperspectral imagery over several basins in California and Colorado. The MODIS-derived products we consider are NASA standard products MOD10A1, MCD43A3, and MCD19A3D, along with STC-MODSCAG/MODDRFS and MODIS SPIReS. These products vary in their retrieval algorithms, including whether, for mixed pixels, they represent the albedo of snow within the pixel or the albedo of the whole pixel. When compared to in situ data, STC-MODSCAG/MODDRFS and SPIReS products have the highest accuracy (RMSE ≤0.093) and most spatially and temporally complete data records (∼99 %) because the algorithms each have independently developed gap filling and interpolation methods. The MOD10A1 and MCD43A3 products underestimate snow albedo (RMSE ≤0.248) because they incorporate non-snow land surfaces into their calculations and have less complete data records (∼76 %) due to less accurate snow detection and lack of interpolation. The MCD19A3D product has accuracy similar to STC-MODSCAG/MODDRFS and SPIReS (RMSE = 0.090) but the lowest data completeness of all datasets (56 %). We found similar performance trends when comparing the MODIS products to airborne hyperspectral data. Our analysis shows algorithms that account for fractional snow cover and incorporate all available spectral information result in the best snow albedo products across time and space. Similar algorithms applied to hyperspectral data can better resolve spectral features to retrieve optical properties of snow; hence we can expect improvements in snow albedo retrievals from future hyperspectral satellite missions.