Publications

Zhang, HB; Zhang, F; Zhang, GQ; Yan, W; Li, SE (2021). Enhanced scaling effects significantly lower the ability of MODIS normalized difference snow index to estimate fractional and binary snow cover on the Tibetan Plateau. JOURNAL OF HYDROLOGY, 592, 125795.

Abstract
MODIS fractional (FSC) and binary (BSC) snow-cover data are important for obtaining accurate spatiotemporal snow-cover information for the Tibetan Plateau (TP) where rapid warming is closely related to snow-cover changes. However, FSC and BSC data are no longer provided in the newly released version (v6) of the MODIS snow cover product, having been replaced with normalized difference snow index (NDSI) data. Recent studies have observed clearly lower accuracy of MODIS snow cover data on the TP than in other areas, possibly implying that there are strong scaling effects due to the complex terrain and land cover which are not well understood. A total of 353 Landsat-8 scenes covering most parts of the TP are used to establish a new empirical relationship between FSC and MODIS NDSI for FSC estimation and a new NDSI threshold for BSC estimation. The results indicate that the new regression model (mean root-mean-squared-deviation (RMSD): 0.22) has a better FSC estimation accuracy than the previously used global reference equation (mean RMSD: 0.24) and that the new NDSI threshold of 0.29 (mean Cohen's Kappa (CK): 0.49) outperforms the global reference NDSI threshold of 0.4 (mean CK: 0.40) in BSC estimation. The relatively low accuracy could be due to an enhanced scaling effect. The 30-m Landsat-8 NDSI data are upscaled to 500-m (MODIS) to analyze the scaling effects on FSC and BSC estimates made using MODIS NDSI. We find that the methods using MODIS NDSI have much lower estimation accuracy, for both FSC and BSC, compared with those using upscaled Landsat-8 NDSI. An analysis of variance (ANOVA) test which considers 512 combinations of aspect, slope and normalized difference vegetation index (NDVI) further demonstrates that the enhanced scaling effects are mainly caused by terrain factors (i.e. aspect and slope). The optimal NDSI threshold for estimating BSC generally increases with slope and decreases as the aspect varies from the southeast to northwest. This study has important implications for the optimal use of MODIS NDSI snow cover data on the TP and highlights the importance of developing more advanced methods which take more factors into account.

DOI:
10.1016/j.jhydrol.2020.125795

ISSN:
0022-1694