Comparison of methods for estimation of Kyoto Protocol products of forests from multitemporal Landsat

The Kyoto Protocol requires nations to report on their reforestation, afforestation, and deforestation (RAD). Using 1990 as a baseline, nations are also required to monitor changes in carbon stocks leading up to the reporting period 2008 to 2012. A study was conducted using three dates of boreal summer Landsat 5 imagery to estimate above-ground carbon for a forested test site near Hinton, Alberta. The carbon estimates were compared with those derived from Canada's national forest inventory. The remote sensing estimates for areas that had not changed were consistent year to year within 3%. The experiment was repeated with the addition of leaf-on and leaf-off image pairs and Landsat-7 imagery. A comparison was made of the classification accuracies achieved for forest classes with single date and paired leaf-on and leaf-off image sets. Spatial properties were incorporated into the image analysis by first creating a multitemporal segmentation. This paper reports on the classification methods used, compares the classification accuracies achieved, and gives recommendations for the creation of Kyoto Protocol products for temperate forests derived from remotely sensed imagery.