MODIS 다중채널 반사도 프로파일을 활용한 DTW 기반 적설탐지 기법

Abstract

Snow cover is one of climate factors that affects the artic environment, and it has an important role about hydrological modeling and water resource management. For this reason, accurate detection of snow cover acts as an essential element for regional water resources management. Snow cover detection using satellite data have some advantages such as obtain wide spatial range data, available time-series observations periodically. In snow cover mapping using satellite, discrimination of snow cover and cloud is very important because it is main error factor. Misclassified cloud and snow pixel can lead directly to error factor for retrieval of surface products from satellite data. Cloud has similar reflectance to snow. But Cloud shows higher reflectance than snow in 1.5~1.7μm because cloud has lower grain size and moisture content than snow. So Cloud and snow show difference reflectance patterns change according to wavelength. Therefore, in this study, we perform algorithm for classifying snow cover and cloud with remote sensing data. For classifying snow cover and cloud, we use Dynamic Time Warping (DTW) Algorithm which is one of commonly used pattern analysis methods such as speech and fingerprint recognitions. Cloud and Snow spectral library as reference data is constructed using MOD21km (MODIS Level1 swath 1km) data that their reflectance is at six channels including 3 (0.466μm), 4 (0.554μm), 1 (0.647μm), 2 (0.857μm), 26 (1.382μm) and 6 (1.629μm). We validate our result using MODIS RGB image and MOD10 L2 swath product which is MODIS snow cover product. And we use PA, UA, CI as validation criteria. We perform quantitative validation analysis with MODIS snow cover product. As result of quantitative validation, PA, UA, and CI present 93.42 %, 59.29 % and 74.07 %, respectively. In the result of qualitative validation, MODIS snow cover did not detect as snow in the several region which is detected as snow in MODIS RGB images. In the same region, result of our study is detect as snow because we consider angle component and status of snow. The result of this study can improve accuracy of surface products from satellite data such as land surface temperature and surface reflectance. Also it can use input data of hydrological and atmospheric modeling.