基于多尺度统计样本的天山山区MOD10A1分类精度评价  

Quality Assessment of MOD10A1 in the Tianshan Mountains Based on Multiple Statistical Sample Scales

作　　者：陈晓娜[1,2] 包安明[1] 刘萍[1,2] 

机构地区：[1]中国科学院新疆生态与地理研究所,乌鲁木齐830011 [2]中国科学院研究生院,北京100049

出　　处：《国土资源遥感》2010年第3期80-85,共6页Remote Sensing for Land & Resources

基　　金：国家科技支撑计划项目(编号:2007BAH12B03);中国科学院知识创新重要方向性项目(编号:KZCX-YW-BR-12)共同资助

摘　　要：以2007年5月15号晴空状态下获取的天山地区MOD10A1积雪遥感影像和Landsat-5 TM数据为基础,通过SNOMAP算法提取基于Landsat-5 TM数据的天山山区积雪分类图(30 m×30 m)。将MOD10A1数据与Landsat-5 TM积雪分类图进行对比分析,并分别在50像元×50像元、10像元×10像元和3像元×3像元统计样本尺度上对其进行定量的质量评价。结果表明:随着统计空间尺度的减小,晴空状态下MOD10A1积雪产品的分类精度有降低趋势,上述不同统计尺度上的分类精度分别为0.94、0.87与0.80,说明受空间分辨率的限制,MOD10A1积雪产品的应用存在有效尺度及最优尺度的问题;同时,在上述不同统计尺度上各代表性样方的精度值间的方差呈增大趋势,分别为0.032、0.074与0.135,说明误差增大的同时,MOD10A1数据质量的稳定性在下降。In this paper,the MOD10A1 and Landsat- 5 TM images were used as the basic data, and the snow information was extracted from Landsat -5 TM with SNOMAP developed by Hell et al.. Furthermore, a comparison between the MOD10A1 data and the classification map from Landsat -5 TM was made, and the quality accuracy of MOD10A1 was calculated at three statistical sample scales （50 pixel × 50 pixel, 10 pixel × 10 pixel and 3 pixel× 3 pixel ）. The results show that, with the decrease of statistical sample scales, the statistical classification accuracy of snow in MOD10A1 images decreases,and the mean quantity accuracies at 50 pixel × 50 pixel,10 pixel × 10 pixel and 3 pixel ~ 3 pixel scale are 0.94,0.87 and 0.80 respectively. These data suggest that, limited by the spatial resolution, there is an efficient or optimum scale when MOD10A1 is applied. Meanwhile, the statistic results show that, with the decrease of the statistical classification accuracy, the stability of the MOD10A1 gradually becomes lower.

关 键 词：天山山区 MODIOA1 数量精度 质量评价 

分 类 号：TP751.1[自动化与计算机技术—检测技术与自动化装置]