Automated Strategies for Distinguishing Deforestation from Mosaiced Landsat Data

Principal Investigator:

Christopher S. R. Neigh

Large-scale deforestation disturbance estimates are currently generated from remote sensing instruments such as MODIS and AVHRR, but they do not have the fine spatial resolution to distinguish many anthropogenic impacts that are occurring in our terrestrial world. MODIS has enhanced our vision to 250m, but it is limited temporally to the last four years. Recent data set creation and increasing processing speed has enabled free access to globally orthorectified mosaiced Landsat data from base years of 1990 and 2000. Establishing better estimates of change in vegetative cover will enable greater knowledge of how humans are altering above ground biomass (stocks of carbon), species habitat, biogeochemistry, and biogeophysical terrestrial properties. Information on where and to what extent humans are altering the land is a pressing issue because of our pervasiveness in global change. Developing an automated technique to extract land use and/or land cover change for extensive regions at 28.5m will quickly provide a base metric for terrestrial world change at a fine spatial scale. Previous localized studies performed at this magnitude have proven to be laborious tasks. Preprocessed datasets now freely exist to make this work possible. Research challenges will have to be addressed in processing schema to overcome cloud cover/cloud shadow, terrain shadow, atmospheric haze, and seasonality between dates. This research will focus on multiple pathways for extracting the desired product, while establishing the most valid method.