Factors of Forest Loss using An Object-based Mapping Method -A Case Study in North America

Changes in forest cover affect the delivery of important ecosystem services, including biodiversity richness, climate regulation, carbon storage, and water supplies. In 2013, Hansen et al. published a scientific paper in Science. They mapped global tree cover extent, loss and gain for the period from 2000 to 2012 at a spatial resolution of 30m, with loss allocated annually. They found a total of 2.3 million km2 of forest loss and 0.8 million km 2 of new forest established.

Concerning loss, all stand-replacement disturbances were mapped, including mechanical removal (logging), fire and other factors such as storms and disease. The implications on carbon cycle dynamics are very different for these various causes of forest loss. For example, emissions from the various change dynamics vary greatly, and labeling forest loss by change factor will improve carbon cycle and carbon accounting efforts.

In this study, we will try to distinguish forest loss caused by fire and logging from other causes. We take Canada and the continental US as the study area, which were estimated to have lost 52.8 MHa forest from 2000 to 2012. Our preliminary results show that fire and logging are the dominant factors (>90%) of North American forest loss. We employ annual minimum NDVI band, annual ETM+ composites for band 3, band 4, band 5, band 7 from 1999 to 2012, and three derived bands from ETM+ composites (dataset A), together with the annual 30m global forest loss map (2001- 2012) (dataset B). The method can be divided into four parts: 1. Generate patches from dataset B. 2. Extract shape, spectral, texture and contextual features from dataset A, with a total of 445 features for each patch; 3. Classify the patches with those features using a bagged regression tree model; 4. Validate and evaluate the results. Mapping results and analysis will be presented at the meeting.