시계열 Landsat 영상 생성을 위한 시공간 다중센서 영상 융합 모델의 평가 - 한라산 사례연구 -

초록

Spatio-temporal fusion of multi-sensor satellite images with different spatial and temporal resolutions can generate time-series images with both high spatial and temporal resolutions to monitor phenological changes of indigenous plants. This paper evaluates the predictive performance of spatio-temporal fusion models to generate Landsat-like images by fusing MODIS into Landsat images. Three spatio-temporal fusion models including STARFM(spatial and temporal adaptive reflectance fusion model), ESTARFM(enhanced STARFM), and FSDAF (flexible spatiotemporal data fusion) are compared via an experiment using MODIS and Landsat red and near infrared images in Mt. Halla where increasing decline and death of Korean fir have been reported. The prediction accuracy of ESTARFM using two pairs of input images was higher than that of the other two models in 2009, but FSDAF using one pair of input images outperformed better than ESTARFM in 2020. These different prediction results were mainly due to the strength of correlation between input images on base and prediction dates. When the correlation was relatively weak, ESTARFM using multiple-pair images yielded the best prediction accuracy. On the contrary, the stronger the correlation, the greater the prediction accuracy of FSDAF. Therefore, the statistical relationships between input images and spatial patterns of input and output images should be considered to selecting the best spatio-temporal fusion model for monitoring phenological changes.

키워드