Highlighting the Limitations of Forest Change Products for REDD+ Evaluation: a Case Study in Sierra Leone

By Hannah Graham

Abstract
Remote sensing based forest change products provide valuable data on global deforestation trends, making them essential resources for evaluating Reducing Emissions from Deforestation and Forest Degradation (REDD+) policies. However, using one dataset over the other can lead to different conclusions on REDD+ impacts, posing a critical challenge to the credibility of carbon financing. This study assesses Global Forest Change (GFC) and Tropical Moist Forest (TMF) products to explore their utility for REDD+ assessments. Using data from 2013-2023 and the focus region of Sierra Leone’s voluntary Gola REDD+ project, findings showcase alarming differences between GFC and TMF datasets concerning the trends, spatial extent, and accuracy of estimated deforestation events. Low spatial agreement (<30% throughout 2013-2023) between the datasets introduces unwanted ambiguity in map-based deforestation estimates, exacerbated by 0.2-1ha differences in the average size of detected deforestation patches per year. GFC and TMF datasets exhibit large discrepancies in overall accuracy, with values of 78.1% and 63.2%, respectively. Increasing trends in omission errors in the TMF dataset highlight the risk of underestimating future deforestation, neglecting up to 2,300ha of deforestation in 2023. Markedly, stratified sample-based area estimates offer a statistically rigorous alternative to map-based deforestation estimates and report up to 24,000ha in recurrent forest disturbances otherwise unaccounted for in GFC and TMF datasets. In the context of REDD+ evaluations, this suggests that GFC and TMF datasets alone have a limited capacity to deliver credible deforestation estimates, underscoring the necessity to integrate forest regrowth and recurrent disturbances in future assessments.

Keywords: REDD+; Deforestation; Global Forest Change; Tropical Moist Forest; Accuracy; Area Estimation; Credibility