Using the regression estimator with Landsat data to estimate proportion forest cover and net proportion deforestation in Gabon

摘要: Abstract Forest cover maps were produced for the Gabonese Agency Space Studies and Observations (AGEOS) 1990, 2000 2010 an area of approximately 102,000 km 2 corresponding to 38% total Gabon representative range human pressure on forest resources. The constructed using a combination semi-automated classification procedure manual enhancements ensure greatest possible accuracy. A two-stage frame sampling approach was adopted collect reference data assessing accuracy estimate proportion net deforestation. 251 2 × 2 km segments or primary sample units (PSUs) visually interpreted by team photo-interpreters independently from map production produce dataset representing about 1% study area. Paired observations extracted random selection 50 secondary (SSUs) in form pixels within each PSU. Overall accuracies greater than 95%. PSU SSU outputs used deforestation both direct expansion model-assisted regression (MAR) estimators. All estimates similar, but variances MAR smaller factors as great 50. In addition, SSU-level had standard errors slightly those PSU-level estimates, differences small, particularly when auxiliary variables obtained maps. Therefore, justified collecting reliable estimating Finally, despite large overall accuracies, alone can be misleading indicated finding that which included adjustment bias twice non-adjusted periods 1990–2000 1990–2010. results confirmed expected generally small level Gabon. However, loss appears have almost stopped last 10 years. One explanation could creation national parks implementation concession management plans onward, this should further explored.