Towards Methodologies for Global Monitoring of Forest Cover Characteristics with Coarse Resolution Data

Principal Investigator: Dr. DeFries

This project addresses the need to develop prototype methodologies for global monitoring of forest cover with coarse resolution data in the context of Global Observations of Forest Cover activities.  It builds on previous research to develop methodologies for characterizing forest cover and changes in forest cover independent of the often varying thresholds of canopy cover considered to be “forest.”  By developing a training and validation data set based on in situ measurements as well as high resolution Landsat data, it is developing a prototype product for the conterminous United States using coarse resolution data (AVHRR and MODIS when available).  The methodology for combining in situ, high resolution, and coarse resolution data will serve as a prototype that can be extended to other parts of the world.  The project also examines the ability of the methodology to identify changes in forest cover by applying it to individual years and assessing the extent to which differences represent actual change.  Until several years of MODIS data are available, this part of the project will use AVHRR 1km data.  Further, it addresses the need within GOFC for methodologies that are automated and repeatable.  A number of techniques such as automated noise reduction for training data, feature selection, and enhancements to decision tree classifiers are assessed for their potential to automate the procedures.