MODIS-derived snow-free ground reflectance statistics of selected Eurasian non-forested land cover types for the application of estimating fractional snow cover

Abstract

This work aims at the improvement of the reflectance model-based methodology for mapping of fractional snow cover (FSC) in seasonally snow covered areas in the Northern Hemisphere. The investigations are focused to the determination and analyses of snow-free ground reflectance (ρground). The objective is to derive information on typical average snow-free ground reflectance for predominant non-forested land cover classes in Eurasia and to investigate the implications of their variations to the FSC estimation accuracy. The snow-free ground reflectances are derived from MODIS time series, using single band reflectances and their related indices. Our approach for the determination of ρground is based on the assumption that the behavior of ground reflectance in conditions occurring directly after the snow melt adequately represents the snow-free ground reflectance during snow ablation i.e. melting snow cover including snow-free patches. Additionally, the effects of the ρground variability to snow mapping accuracy for different land cover types are analyzed focusing to the SCAmod algorithm. The current operational implementations of SCAmod use a fixed value for snow-free ground reflectance at the visible wavelengths around 555nm (corresponding to Terra/MODIS band 4). The deviation between the true snow-free ground value and the fixed value causes error in FSC estimation and, thus, it is necessary to investigate the magnitude and variations of the true ρground for different land covers. Even so, our investigation for the target area of Europe, shows that the currently used fixed value of ρground (10.0%-units) in SCAmod seems to work adequately for land cover classes investigated here. For example, the obtained ρground mean and standard deviation 10.0±1.3%-units for the agricultural areas and steppe at 555nm seem to widely coincide with the fixed value. According to SCAmod, the error in FSC estimation caused by the deviation between the fixed and the land cover class-specific value of ρground for agricultural areas and steppe yields a systematic error close to zero and a random error ranging from 2.5 to 1.5%-units with the corresponding FSC range of 0–50%. However, in the case of wetlands (7.4±0.8%-units), the systematic error caused to FSC estimation using the fixed value compared to using the estimated class-specific value is as large as from 5 to 2.5%-units with the FSC range of 0–50%. For FSC retrievals larger than 50% the error caused by the variability in ρground is considerably smaller for all of the studied land cover classes.Our results show that the performance of SCAmod improves if land cover-specific ρground values are applied at least for wetlands. We also suggest that the global application of SCAmod would benefit from the generation of a ρground map for all land cover types by using the methods presented in this paper. This post-winter reflectance climatology map can be constructed by calculating and using the class-specific ρground statistics together with suitable land cover information.