MAPPING THE BIOMASS OF GOLA FOREST RESERVE IN SIERRA LEONE WITH REMOTE SENSING DATA AND NEURAL NETWORKS

摘要: This work describes the estimation of the above ground live biomass (AGLB) of a tropical forest area, the Gola Forest Reserve in Sierra Leone, from field plot measurements and Landsat TM/ETM+ images, using machine learning techniques for retrieval, and correlating the results with lidar derived metrics from the Geoscience Laser Altimeter System (GLAS) instrument onboard the NASA ICESat satellite. The study will also contribute to a research modeling effort, supported by the Cambridge Conservation Initiative, aimed at comparing alternative predictive models for carbon emission reductions calculation in the framework of the UN-REDD (United Nations Reduction in Emission from Deforestation and Forest Degradation in Developing Countries) program. Mapping and monitoring carbon stocks and forest changes in tropical regions are essential steps in the implementation of a future carbon market. While accounting methods are still under discussion, efforts are needed to test different carbon monitoring approaches. Selected methods should be efficient, costeffective and robust enough to enable countries to measure their stock at different scales, from national to local; the former for reporting and accounting purposes and the latter to allow internal forest management, planning and evaluation, conservation prioritization, and to identify the contribution from different areas to the national carbon budget.The study site (Fig. 1), the Gola Forest Reserve, is the newest National Park in the country with an area of about 710 square km, and the largest closed canopy lowland rain forest remaining in Sierra Leone. The forest holds several endangered …