The Stepwise Discriminant Algorithm for Snow Cover Mapping based on FY-3/MERSI Data

Medium Resolution Spectral Imager (MERSI) on board China's new generation polar orbit meteorological satellite FY-3A provides a new data source for snow monitoring in large area. As a case study, the typical snow cover of Qilian Mountains in northwest China was selected in this paper to develop the algorithm to map snow cover using FY-3A/MERSI. By analyzing the spectral response characteristics of snow and other surface elements, as well as each channel image quality on FY-3A/MERSI, the widely used Normalized Difference Snow Index (NDSI) was defined to be computed from channel 2 and channel 7 for this satellite data. Basing on NDSI, a tree-structure prototype version of snow identification model was proposed, including five newly-built multi-spectral indexes to remove those pixels such as forest, cloud shadow, water, lake ice, sand (salty land), or cloud that are usually confused with snow step by step, especially, a snow/cloud discrimination index was proposed to eliminate cloud, apart from use of cloud mask product in advance. Furthermore, land cover land use (LULC) image has been adopted as auxiliary dataset to adjust the corresponding LULC NDSI threshold constraints for snow final determination and optimization. This model is composed as the core of FY-3A/MERSI snow cover mapping flowchart, to produce daily snow map at 250m spatial resolution, and statistics can be generated on the extent and persistence of snow cover in each pixel for time series maps. Preliminary validation activities of our snow identification model have been undertaken. Comparisons of the 104 FY-3A/MERSI snow cover maps in 2010-2011 snow season with snow depth records from 16 meteorological stations in Qilian Mountains region, the sunny snow cover had an absolute accuracy of 92.8%. Results of the comparison with the snow cover identified from 6 Terra/MODIS scenes showed that they had consistent pixels about 85%. When the two satellite resultant snow cover maps compared with the 6 supervise-classified and expert-verified snow cover maps derived from integrated MERSI and MODIS images, we found FY-3A/MERSI has higher accuracy and stability not only for nearly cloud-free scenes but also the cloud scenes, namely, FY-3A/MERSI data can objectively reflect finer spatial distribution of snow and its dynamic development process, and the snow identification model perform better in snow/cloud discrimination. However, the ability of the FY-3A/MERSI model to discriminate thin snow and thin cloud need to be refined. And the limitation, error sources of FY-3A/MERSI snow products would be assessed based on the accumulation of large amounts of data in the future.