Consistent forest change maps 1981–2000 from the AVHRR time series: Case studies for South America and Indonesia

This study predicts global forest cover change for the 1980s and 1990s from AVHRR time series metrics in order to show how the series of consistent land cover maps for climate modeling produced by the ESA climate change initiative land cover project can be extended back in time. A Random Forest model was trained on global Landsat derived samples. While the deforestation was underestimated by the model, major global patterns were effectively reproduced. Compared to reference data for the Amazon satisfying accuracies (>0.8) were achieved, but results are less promising for Indonesia.