Assessment of Methods for Mapping Snow Albedo from MODIS

Water resource managers, weather forecasters, and climate modelers all require accurate snow albedo measurements to drive or validate their models. Several daily albedo products that represent whole pixel albedo or specifically snow albedo are available from the MODIS sensor, including standard NASA products as well as products generated using more mechanistic algorithms. Here, we compare a suite of daily MODIS-based albedo products to terrain-corrected, in situ albedo observations from field sites in California and Colorado, USA. Additionally, we intercompare the products to highlight their strengths and weaknesses.

The products we consider include MOD10A1 and MCD43A3, standard NASA products which provide broadband or shortwave albedo measurements at 463 m spatial resolution. We also examine MCD19A3D, a NASA product that provides grain size at 927 m spatial resolution which can be converted to the albedo of clean snow. Finally, we compare two additional approaches processed at 463 m spatial resolution, STC-MODSCAG/MODDRFS and MODIS SPIRES, that solve for the snow fraction, grain size, and impact of light absorbing particles (LAP) on snow albedo; snow albedo is estimated by combining the grain size with darkening due to LAP.

A review of the approaches reveals differences that impact albedo retrieval accuracy and stability, as seen in the product intercomparison and validation with in situ data. Regarding viewing geometry, the approaches vary from not accounting for different view zenith (MOD10A1), to temporal smoothing (MCD43A3), to temporal and spatial coarsening (MCD19A3D), to explicitly accounting for viewing geometry using a weighting scheme for each pixel based on the view zenith angle (STC-MODSCAG/MODDRFS and MODIS SPIRES). Some products incorporate approaches to account for fractional snow coverage (MCD19A3D, STC-MODSCAG/MODDRFS, MODIS SPIRES) whereas others instead incorporate non-snow land surfaces like vegetation, soil, and rock (MOD10A1 and MCD43A3). Some approaches explicitly account for the impact of LAP on snow albedo (MCD19A3D, STC-MODSCAG/MODDRFS, MODIS SPIRES) while others do not (MOD10A1 and MCD43A3). Finally, some algorithms provide a daily gap-free time series (STC-MODSCAG/MODDRFS and MODIS SPIRES), while others do not (MOD10A1, MCD43A3), while MCD19A3D appears to have large missing square shaped regions from a possible error in processing logic.

Results show that overall, STC-MODSCAG/MODDRFS and MODIS SPIRES snow albedo products present the highest accuracy, especially for pixels not fully snow covered. MOD10A1 and MCD43A3 underestimate snow albedo in pixels not fully snow covered because they incorporate non-snow land surfaces. Spatial smoothing/coarsening in MCD43A3 and MCD19A3D overlook real changes in the snow albedo from new snow, metamorphism, and LAP deposition. MOD10A1 oscillates between high and low snow albedo values perhaps depending on the satellite view zenith. Of the NASA standard products, MCD19A3D is the only one that considers LAP by assuming a static background soot concentration, but cannot account for changes in albedo due to changes in LAP at the snow surface. The STC-MODSCAG/MODDRFS and MODIS SPIRES approaches explicitly account for viewing geometry and allow for tracking of albedo signals from new snow, grain growth, and LAP deposition at a daily time step at 463 m resolution. STC-MODSCAG/MODDRFS and MODIS SPIRES are produced operationally as part of the NSIDC Snow Today website. We solicit research and application users to support future operational production of these products, presumably at the NSIDC DAAC or another supportive facility.