Date of this Version
Remote Sensing of Environment 118 (2012) 241–248; doi:10.1016/j.rse.2011.11.022
Russian boreal forests are the largest forested zone on Earth and a tremendous pool of organic carbon. Current limited records on forest structure, composition, successional stage and disturbances contribute to large uncertainties in estimates of carbon stocks and fluxes in this zone. Our ability to monitor ongoing changes in forest cover has improved with the influx of remotely sensed data products since 2000 from multiple satellite platforms. Here we present a method aimed at reconstructing disturbance history from a known distribution of land cover. We developed and tested the method over a biologically and topographically diverse region of the Russian Far East. This method explores capabilities introduced through fusion of the long-term but spatially limited Landsat data archive and the spatially continuous but temporally limited 2000-present data record from the Moderate Resolution Spectroradiometer (MODIS). Landsat data from 1972 to 2002 were used to develop a reference disturbance dataset to train and validate a MODIS-based decision tree classification. The results showed a reliable differentiation of disturbed and mature forests with an overall accuracy of 88% (Kappa 0.73). Individual disturbances by type and decade were estimated with an overall accuracy of 70% (Kappa 0.64).