How accurately may we project tropical forest-cover change? A validation of a forward-looking baseline for REDD

The Reduced Emissions from Deforestation and forest Degradation (REDD+) mechanism of a future post-2012 global climate-change treaty would aim to give incentive to tropical countries to reduce deforestation and thus forest-carbon emissions. It would do so by crediting tropical countries for reducing deforestation relative to a baseline scenario describing carbon emissions and removals from forest-cover change expected in the absence of REDD+. Defining a credible and accurate baseline is both critical and challenging. One approach considered promising is spatial modelling to project forest-cover change on the basis of historical trends; yet few such projections have been validated at a national scale. We develop and validate a novel GEOMOD projection of forest-cover change in Panama over 2000–2008, based on trends over 1990–2000 and 25 drivers of forest-cover change. Compared with the actual landscape of 2008, our projection is 85.2% accurate at a 100-m pixel resolution. More error is attributable to the location of projected forest (8.6%) than to its area (6.2%). Accuracy was least where forest regeneration predominated (80%), and greatest where deforestation predominated (90%). Despite the sophistication of our projection, it is slightly less accurate than if we had assumed no forest-cover change over 2000–2008. We identify factors limiting projection accuracy, including the complexity of forest-cover change, the spatial variability of forest-carbon density, and the relatively small area of change at the national scale. We conclude that, with the exception of contexts where forest-cover change is significant and straightforward and where forest-carbon density relatively uniform (e.g., agricultural frontiers), spatially projected baselines are of limited value for REDD+ – their accuracy is too limited given their relative lack of transparency. Simpler, relatively coarse scale, retrospective baselines are recommended instead.