Characterizing forest disturbance dynamics in the humid tropics using optical and LIDAR remotely sensed data sets

Thumbnail Image


Publication or External Link





Human-induced tropical deforestation and forest degradation are widely recognized as major environmental threats, negatively affecting tropical forest ecosystem services, such as biodiversity and climate regulation. To mitigate the effects of forest disturbance, particularly carbon emissions, national forest monitoring systems are being established throughout the tropics. Multiple good practice guidelines aimed at developing accurate, compatible and cost-effective monitoring systems have been issued by IPCC, UNFCCC, GFOI and other organizations. However, there is a lack of consensus in characterization of the baseline state of the forests and carbon stocks. This dissertation is focused on the improvement of the current methods of remotely-sensed forest area and carbon loss estimation. A sample-based estimation method employing Landsat-based forest type and change maps and GLAS Lidar-modeled carbon data was first prototyped for the Democratic Republic of the Congo (DRC), and then applied for the entire pan-tropical region. The DRC study found that Landsat-scale (30m) map-based forest loss assessments unadjusted for errors may lead to significant underestimation of forest aboveground carbon (AGC) loss in the environments with small-scale land cover change dynamics. This conclusion was supported by the pan-tropical study, which revealed that Landsat-based mapping omitted almost half (44%) of forest loss in Africa compared to the sample-based estimate (sample-based estimate exceeded map-based by 78%). Landsat performed well in Latin America and Southeast Asia (sample-based estimate exceeded map-based by 15% and 6% respectively), where forest dynamics are dominated by large-scale industrial forest clearings. The pan-tropical validation sample also allowed disaggregating forest cover and AGC loss by occurrence in natural- (primary and mature secondary forests, and natural woodlands) or human-managed (tree plantations, agroforestry systems, areas of subsistence agriculture with rapid tree cover rotation) forests. Pan-tropically, 58% of AGC loss came from natural forests, with proportion of natural AGC loss being the highest in Brazil (72%) and the lowest in the humid tropical Africa outside of the DRC (22%). The pan-tropical study employed a novel forest stratification for carbon estimation based on forest structural characteristics (canopy cover and height) and intactness, which aided in reducing standard errors of the sample-based estimate (SE of 4% for the pan-tropical gross forest loss area estimate). Such a stratification also allowed for the quantification of forest degradation by delineating intact and non-intact forest areas with different carbon content. This indirect approach to quantify forest degradation was advanced in the last research chapter by automating the process of intact (hinterland) forest mapping. Hinterland forests are defined as forest patches absent of and removed from disturbance in near-term history. Their utility in using spatial context to map structurally different (degraded and non-degraded) forests points a way forward for improved stratification of forest carbon stocks. Conclusions from the dissertation summarize strengths and challenges of sample-based area estimation in monitoring forest carbon stocks and the possible use of such estimates in the revision of spatially explicit maps by adjusting them to match the unbiased sample-based estimates. Hinterland forest maps, in addition to providing a valuable stratum for sample-based carbon monitoring, may serve as a baseline for the near real-time monitoring of remaining ecologically intact tropical forests.