17:15 〜 19:15
[ACG44-P05] Investigation of surface changes in Tsambagarav National Park using summer Landsat normalized difference spectral indices
キーワード：地表面変化、正規化差分スペクトル指数、線形混合モデル、ランドサット、ツァンバガラウ国立公園
The Tsambagarav National Park in the Mongolian Altai is a glacial mountain whose summit is at an altitude of over 4,000 meters, and its land surface continues to be changed under global warming. Satellite remote sensing makes it possible to monitor the park with the aim of long-term conservation. For example, historical data records from Landsat series contain rich amounts of environmental information on ecotones between glaciers of the park. To investigate such a long-term trend, our previous work introduced a pixel-based temporal averaging procedure for the Landsat Collection 2 Level-2 Science Product (L2SP). We prepared average images from multitemporal data of thermal, visible, and near-infrared (NIR) bands, and the normalized difference vegetation index (NDVI) in summer, and found trends in warming, darkening, and greening of the park over a few decades. Therefore, an expansion of the procedure to various types of normalized difference spectral indices (NDSIs) designed with other variables of interest, such as water or snow, is expected to be attractive for the extraction of geospatial information over the park. One of the difficulties raised with the combined use of average NDSIs for such a practical purpose is relating averaged NDSIs with physical variables. This work attempts to bridge averaged NDSIs and fractional covers of interest at mixels from a reflectance-based approach. The objective of this study is to introduce an analytical expression of NDSIs with a linear mixture model (LMM) under a simple land cover, and then to investigate surface changes in the Tsambagarav National Park using summer Landsat NDSIs. First, the NDSI is defined as a ratio of the subtraction and sum of a set of reflectances for two certain bands at a mixel. Subsequently, a vector with two elements is set for the LMM: one element is a pure reflectance spectrum from a land surface with a highly sensitive index, and the other is one with low sensitivity. Finally, the analytical expression of NDSIs is yielded using a multitemporal LMM-based average reflectance. A set of numerical experiments were conducted using surface reflectance images from the L2SP during the summer months of June, July, and August in 1990–2012 over the Tsambagarav National Park. In the experiment, the averaging operation is applied to time series surface reflectance data over an 11-year range, and then three types of NDSIs for water, snow, and vegetation are computed with combinations of green and NIR, green and shortwave-infrared, and NIR and red bands, respectively. The Wilcoxon signed-rank test (p<0.05), performed on the average NDSIs for 1995 and 1999, considered as the central years in the averaging period, indicated statistical significance for the snow index in a glacier of the park. NDSI differences between 1995 and the years 1999, 2003, and 2007, on the other hand, showed negative trends for the water and snow indices and a positive trend for the vegetation index around ecotone zones, which implies the possibility of varying land surface information related to each index. These findings suggest that an approach with a multitemporal LMM-based NDSI average is useful for measuring decadal changes in land surfaces over the Tsambagarav National Park with complicated topographies.