Implementation and validation of the snow grain size retrieval SGSP from spectral reflectances of the satellite sensor MODIS

Abstract

Snow is part of the cryosphere in the climate system of the Earth. It has a high albedo in the visible, decreasing towards the near-infrared. Snow on ground is a porous medium of ice, air, and possibly impurities like dust or soot. After deposition, it undergoes snow metamorphism changing the grain size, grain shape, and density. In the visible, the reflection characteristics of snow are mainly determined by the amount of impurities, and in the near-infrared by the size of the snow grains. Satellite sensors allow observing the snow in remote areas like the polar regions on a regular basis and on a global scale. A method to compute the snow grain size and impurity amount from optical satellite observations is the Snow Grain Size and Pollution amount (SGSP) retrieval. It uses data of three reflectance channels (here: at 0.47 µm, 0.86 µm, and 1.24 µm), has a reduced dependency on the snow grain shape, and is applicable at solar zenith angles up to 75°. In this work, the SGSP retrieval is implemented in a near-real time processing chain using data from the Moderate Resolution Imaging Spectrometer (MODIS) operating on the satellites Terra and Aqua. A sensitivity analysis reveals that currently only the snow grain size can be determined reliably by the SGSP retrieval, as the uncertainties of the MODIS instrument are too high for the amount of impurities typically occurring in polar regions. Sensitivity studies on the influence of vertically inhomogeneous snow, wet snow, and cirrus clouds show that the SGSP retrieval typically underestimates the grain size by 15% to 25% for those three cases. The SGSP-retrieved snow grain size is validated using six different ground truth data sets from the Arctic, the Antarctic, Greenland, and Japan from the years 2001 to 2009, and various subsurfaces (land, land ice, sea ice, lake ice). In general, the retrieved and ground-measured grain size are in good agreement. 17 cases have small differences (1 14%), 16 cases intermediate differences (18 53%), and four cases large differences (72 178%). The SGSP retrieval tends to underestimate the grain size for wet snow cases (by 18% 31%) and cirrus cloud cases (by 14% 31%), and overestimates it for surface hoar cases (by 30% 53%) and wind crust cases (by 23% 77%). A comparison of the SGSP retrieval with a previous retrieval using ground measurements from the Himalayan basin shows that the SGSP-retrieved grain size tends to be smaller (by 5 48 µm) and that vertically inhomogeneous snow influences the retrieval. A comparison of SGSP-retrieved snow grain size time series on the Ross ice shelf, Antarctica, at three Automatic Weather Stations (AWS) with snow depth change data from those three stations shows that a snow fall event of 6 cm is detected by the sudden decrease of the retrieved grain size from 200 µm to 50 µm. A comparison of the spectral snow albedo for the MODIS Channels 1 to 5 over 16 days on a large-scale area in Greenland between the SGSP-derived albedo and the spectral MODIS albedo product MOD43 shows a correlation of 0.82 for Channel 5,which is most sensitive to the snow grain size.