关键词: 特征空间, MODIS, NA模型, 积雪丰度

Abstract: As one of the most important surface water resources in Xinjiang，accumulated snow （snowcover） plays an irreplaceable role in sustainable economic and social development. And it is also an important factor affecting the ground water resource management，climate change，disaster prevention and snowmelt simulation forecast. While the snowcover area is the main parameters of accumulated snow，as well as one of the main input parameters for water management，climate change and snowmelt flood simulation and prediction research. At present，the access to snowcover area is via the image data’s [NDSI] obtained from Landsat TM and MODIS，mixed pixel decomposition or direct application of MODIS snowcover products MOD10A2 data. However，the applicability of the various ways in different locations varies. Snowcover abundance represents the pixel within the snowcover content，so the improvement of snowcover abundance extraction accuracy can promote snowcover classification accuracy increases，thereby increasing the extraction accuracy of the snowcover area. For the problem of the fluctuation of snowcover the normalized difference snowcover index （NDSI） values （usually 0.4） for different locations，this paper introduces Albedo which is sensitive to accumulated snow，and normalized difference snowcover index （NDSI） to construct feature space. According to the scatter in the feature space and the actual situation of the study area to establish the inversion NA model which is applicable to Xinjiang snowcover abundance. In this study，two indicators of [NDSI] and Albedo are calculated by the 500 m-resolution MODIS images data. The NA model is built based on the upper left arc distribution of the scattered points，and further use the higher resolution Landsat TM data relative to MODIS data to test the accuracy of model inversion results，and then compare with the results obtained from Support Vector Machine （SVM） method. In this paper，53 validation points were randomly selected based on MODIS images，and determine the pixel corresponding to the NDSI image calculated by Landsat TM data，then take this pixel as the center to acquire 5 pixels ×5 pixels square，furthermore，get the mean of the 25 pixels as the center pixel value.  Finally，two indicators of the root mean square error （RMSE） and correlation coefficient （R²） are used for the test of two extraction methods respectively. The results show that the NA model established by two-dimensional feature space can better invert snowcover abundance of the study area. The correlation coefficient （R²） and the root mean square error （RMSE） of the first method is 0.857 and 0.093；while the correlation coefficient （R²） and root mean square error （RMSE） of the second method is 0.833 and 0.199，the correlation coefficient （R²） of NA model is 2.4 percentage points higher than the model of SVM，while the root mean square error （RMSE） improved 0.106，which shows that the NA model retrieval accuracy is better than the SVM method. This paper introduces the feature space method to the inversion of snow information，which is widely used in many areas，such as desertification monitoring，drought monitoring and soil salinization monitoring. Through the test of high-resolution image and comparison with the SVM method，it proves that the method of this paper is feasible，and its accuracy is better than the SVM method. So it provides a valuable reference to the inversion of accumulated snow information in the arid areas.

Key words: feature space； , MODIS； , NA model； , snowcover abundance ratio