基于混合像元分解的MODIS积雪面积信息提取及其精度评价——以天山中段为例

A Study on Methods and Accuracy Assessment for Extracting Snow Covered Areas from MODIS Images Based on Pixel Unmixing:A Case on the Middle of the Tianshan Mountain

MODIS(Moderate Resolution Imaging Spectroradiometer)是现阶段积雪遥感监测及积雪水文学研究中积雪面积信息获取的重要平台,但其空间分辨率相对较低,影像中混合像元现象普遍存在。本文以MOD02HKM数据为基础,通过线性光谱混合模型(LSMM,Linear Spectral Mixing Model)对研究区MODIS影像进行像元分解,从中提取积雪面积信息,并进行精度评价。将线性光谱混合模型得到的积雪面积信息与美国国家冰雪数据中心提供的MOD10A1日积雪覆盖数据影像进行对比分析。结果表明:利用线性光谱混合模型可以较好的分解出像元中积雪面积信息,其分类精度达0.88;相同位置上MOD10A1的积雪分类精度为0.80。说明,对MODIS影像上积雪信息提取来说,线性光谱混合模型的分类精度较高,具有较强的适用性。

Moderate Resolution Imaging Spectroradiometer（MODIS）is a critical remote sensing data source in snow monitoring and snow hydrology study.However,mixed pixel is a common problem encountered in using satellite data with moderate or low spatial resolution.Relatively low spatial resolutions（i.e.,250 m,500 m,and 1000 m）have limited widespread applications of MODIS data in research,such as snow area extraction,snow water evaluation and snowmelt runoff simulation.In this paper,the author extracted the snow area from MOD02 HKM image,one of the three MODIS L1B products（MOD02 QKM,MOD02 HKM,MOD02 1KM）at 500 m spatial resolution acquired from U.S.National Aeronautics and Space Administration（NASA）,aiming to extract snow area at sub pixel scale on the basis of the line spectral mixture model（LSMM）.In addition,the authors compared the classification accuracy of extracted snow area with the snow cover map derived from MOD10A1 grey level snow and ice products of the same image acquisition time and spatial resolution provided by the U.S.National SnowIce Data Center（NSIDC）,and subsequently estimated their classification accuracies with the binary snow-covered area derived from Landsat 5 TM data at 30 m spatial resolution based on the SNOMAP algorithm using the quality accuracy assessment method.For better running LSMM and eliminating cloud effects on snow area extraction,cloud-free days in May 15,2007,were selected for this study.Results indicated that the use of the line spectral mixture model in snow area extraction can provide better snow classification accuracy,showing the quantity accuracy assessment result of 0.88 and a standard deviation of 0.087 at a 3×3 pixel scale,while the classification accuracy of MOD10A1 grey level snow and ice product was found to be 0.80 and 0.135,respectively,at the same locations and statistical scales.This suggests that the line spectral mixture model could be effective in snow area extraction from MODIS data of relatively low spatial resolutions.