基于不同土地覆盖类型NDSI阈值优化下的青藏高原积雪判别

美国国家雪冰数据中心（NSIDC）发布的MODIS第6版本逐日积雪范围产品（V6）仅提供了归一化积雪指数（NDSI），而用户往往关心的是积雪范围或积雪覆盖率。NSIDC推荐全球积雪范围最佳的NDSI阈值为0.4，但是青藏高原地形复杂，积雪斑块化特征明显，单一的NDSI阈值并不能精确地判识不同下垫面上的积雪。不同的土地覆盖类型可能影响积雪判别的NDSI阈值。以青藏高原为研究对象，基于高分辨率卫星Landsat-5 TM数据，获取了青藏高原不同土地覆盖类型下判识积雪的最优NDSI阈值。结果表明，在草地和稀疏植被地表类型下，最优NDSI阈值分别为0.33和0.40；在其他下垫面类型下，最优NDSI阈值为0.47。以Landsat 8 OLI数据为"真值"对该NDSI阈值确定的积雪范围进行了精度检验。结果表明，采用新的NDSI阈值获取的MOD10A1 V6积雪范围产品的总体精度OA、错分误差OE和漏分误差UE分别为87.88%、5.20%和6.87%。而采用传统的0.4阈值时，其OA、OE和UE分别为87.36%、3.98%和8.60%。这表明考虑不同土地覆盖类型下的NDSI阈值优化可以有效地提高青藏高原积雪判别精度，特别是对占比面积较大的草地区域，通过NDSI阈值优化可以更加准确地识别积雪范围。

Snow cover mapping algorithm in the Tibetan Plateau based on NDSI threshold optimization of different land cover types

The sixth version of MODIS (V6), released by the National Snow and Ice Data Center (NSIDC), offers only normalized difference snow index (NDSI). Users are often concerned about snow cover or fraction of snow cover. NSIDC recommends the world's best snow cover NDSI threshold of 0.4, but the terrain of the Tibetan Plateau is complex, the characteristics of snow patching are obvious, and a single NDSI threshold does not accurately identify snow on different land cover types. Different types of land cover may affect the NDSI threshold. Based on the high-resolution Landsat-5 TM data, in this paper, the optimal threshold of NDSI is obtained for identifying snow cover under different land cover types in the Tibetan Plateau. The results show that the optimal thresholds of NDSI are 0.33 and 0.4, respectively, under the type of grassland and sparse vegetation. The optimal threshold of NDSI is 0.47 under other land cover types. The snow cover determined by the NDSI threshold is accurately validated using Landsat 8 OLI data as "true value". The results show that the overall accuracy (OA), overestimated error (OE) and underestimated error (UE) of MODIS snow cover product based on new NDSI threshold are 87.88%, 5.20% and 6.87%, respectively. It is found that the OA, OE and UE based on traditional 0.4 threshold are 87.36%, 3.98% and 8.60%, respectively. It shows that NDSI threshold optimization under different land cover types can effectively improve the accuracy of snow cover in the Tibetan Plateau, especially in grassland areas with a large proportion of area, and the snow cover can be more accurately identified by NDSI threshold optimization.