Automated Characterization of Global Land Cover Change Dynamics

Principal Investigator:

Kuan Song

Global Land Cover Change Dynamics is an important field that impacts many other fields, such as carbon cycle dynamics and global forest inventory dynamics. Global Land Cover Change Dynamics is especially relevant in light of recent economic development worldwide. During the past twenty years, significant deforestation has occurred in developing countries, while some primary forests have been protected in reserves. The past twenty years also saw significant climate change and atmospheric carbon dioxide level increase. In debate is whether or not there is a greening trend in the mid and high latitudes. Evidence shows that the NDVI generated from AVHRR data of the past twenty years increased, but atmospheric-inversion gets inconsistent results. While in the US, gradual conversion has been observed between grassland and shrubland. Thus the global land cover dynamics during the past twenty years is a possible mix of anthropogenic and climate-driven changes. This research will try to separate these two factors by automatic methods applied on multi-date Landsat collection GeoCover offered by EarthSat and hosted by GLCF, and the GIMMS dataset for the past twenty years. The idea is to view the GIMMS dataset as a 'video' of earth's vegetation cover and apply a video-detection method, Optical Flow Field, to find the gradual, climate-driven changes; while Landsat imagery for years 1973, 1990, and 2000 are used to detect human-induced changes. To overcome the difficulty of handling all the Landsat data, we choose to perform a 1-degree sampling over the earth's surface, with each sampling a 10km by 10km size.