Toward global baselines and monitoring of forest cover for REDD: the Global Forest Cover Change project

Monitoring, Reporting, and Verification (MRV) procedures in support of Reducing Emissions from Deforestation and forest Degradation (REDD) require the establishment of historical baselines of forest cover and changes, as well as consistent monitoring of subsequent forest gains and losses over time. Under the NASA MEaSUREs program, the Global Forest Cover Change project is using the USGS Global Land Survey (GLS) dataset of Landsat images to generate Earth Science Data Records (ESDRs) for monitoring forest cover over multiple decades at sub-hectare spatial resolution. These data products include layers representing forest cover, change, and fragmentation in 1975, 1990, 2000, and 2005, as well as atmospherically corrected surface reflectance images for these same GLS “epochs”. Monitoring at this scale requires high levels of automation and radiometric precision. Atmospheric correction is accomplished with the 6S radiative transfer code, and classification is performed with Support Vector Machines fit with training data gathered by automated procedures. Surface reflectance images for the 2000 and 2005 epochs were recently released for public use. For the 2000 epoch, 94% of images had Root-Mean Squared Difference (RMSD) less than 5% reflectance compared to coincident MODIS daily surface reflectance (MOD09) across all bands. For 2005, 92% of images based on Landsat-7 and Landsat-5 met this specification relative to MODIS daily surface reflectance and 16-day NBAR composites (MCD43A4), respectively. Forest cover and change maps are being validated against visually interpreted reference data; pilot studies conducted in several countries showed accuracies above 90%. Classification errors are predominantly due to poor discrimination of deciduous forests from crops and other herbaceous cover types, and so procedures have been devised for flagging and/or replacement of phenologically unsuitable GLS images. Web-based tools have been developed for rapid collection of multi-temporal reference data, and we are exploring ways to incorporate community efforts for validation through this interface. Accuracy assessments will be updated as algorithms are refined. This talk will describe project goals, algorithms, and product status, with additional emphasis on accuracy assessment of surface reflectance and forest cover products. These and forthcoming data records will provide crucial information to resource managers and scientists monitoring changes in forest cover over recent decades.