Improved quantification of forest cover change and implications for the carbon cycle

Abstract

Changes in forest cover significantly affect the global carbon cycle, the hydrological cycle and biodiversity richness. This dissertation explores the potential of satellite-derived land cover datasets in quantifying changes in global forest cover and carbon stock. The research involved the following three components: 1) improving forest cover characterization, 2) developing advanced methods for detecting forest cover change (FCC) and 3) estimating the amount and trend of forest carbon change.

The first component sought to improve global forest cover characterization through data fusion. Multiple global land cover maps have been generated, which collectively represent our current best knowledge of global land cover, but substantial discrepancies were found in their depiction of forest. I demonstrated that the extent and density of forest cover could be much better characterized by integrating existing datasets. However, these independent map products cannot be directly compared to quantify FCC, because post-classification change detection requires significant consistency in land cover definition, satellite data source and classification procedure. The yearly vegetation continuous field (VCF) product derived from the Moderate Resolution Imaging Spectroradiometer (MODIS) provides a prototype that fulfills such requirement. The second component was intended to explore the features of this time series dataset in change analysis. A new algorithm called VCF-based Change Analysis was developed that can explicitly characterize the timing and intensity of FCC. The efficiency and robustness of this algorithm stem from two realistic assumptions—the spatial rarity and the temporal continuity of land cover change/modification. The developed method was applied to continental scales for mapping forest disturbance hotspots.

The third component of the research combined MODIS-based deforestation indicators, a Landsat sample and a biomass dataset to estimate annual carbon emissions from deforestation with a regional focus on the Amazon basin. I found that deforestation emissions varied considerably not only across regions but also from year to year. Moreover, deforestation has been progressively encroaching into higher biomass lands in the Amazon interior. These observed deforestation and emission dynamics are expected to provide scientific support to policies on reducing emissions from deforestation and forest degradation (REDD+). The generated panel data are also of great value for evaluating forest protection policies.