全球变化知识资源中心

Analysis of long-term change trend of vegetation covers faces several challenges. First, the change of vegetation covers exhibits remarkable cyclical oscillations and disturbances, which often mask the long-term trend. Second, the remote sensing image based time-series analysis is often complicated by the lack of long-series ground reference data. In addition, current remote sensing based vegetation change analysis focuses on continental, national or regional scales. Local scale analysis of vegetation changes is often neglected. In this paper, we developed a synthesized analysis framework, the domain adaptive and classified smoothing - DACS, to tackle these limitations. We integrated domain adaptive learning with the classified maps of vegetation groups (VGs) to derive the ground reference data supporting the image-based time-series analysis and to support local scale spatial analysis. We applied empirical mode decomposition model to extract long-term trends of VGs NDVI changes. We employed univariate polynomial regression to examine the long-term change trends of VGs and to identify their spatial and temporal patterns. We tested this new algorithm by analyzing the SPOT-VEG NDVI 10-days composites from 1999 to 2007 in Wulate, Inner Mongolia, China. The findings confirmed that the newly developed DACS method effectively captured the long-term trends of NDVI change over various VGs and revealed that the regional trend of NDVI change differed from the local trends of change of different VGs. DACS is suitable to other image sources, such as MODIS and Landsat images, and can be applied in other regions for local scale time-series analysis of vegetation cover changes.