Identifying areas of deforestation risk for REDD+ using a species modeling tool

Background To implement the REDD+ mechanism (Reducing Emissions for Deforestation and Forest Degradation, countries need to prioritize areas to combat future deforestation CO 2 emissions, identify the drivers of deforestation around which to develop mitigation actions, and quantify and value carbon for financial mechanisms. Each comes with its own methodological challenges, and existing approaches and tools to do so can be costly to implement or require considerable technical knowledge and skill. Here, we present an approach utilizing a machine learning technique known as Maximum Entropy Modeling (Maxent) to identify areas at high deforestation risk in the study area in Madre de Dios, Peru under a business-as-usual scenario in which historic deforestation rates continue. We link deforestation risk area to carbon density values to estimate future carbon emissions. We quantified area deforested and carbon emissions between 2000 and 2009 as the basis of the scenario. Results We observed over 80,000 ha of forest cover lost from 2000-2009 (0.21% annual loss), representing over 39 million Mg CO 2 . The rate increased rapidly following the enhancement of the Inter Oceanic Highway in 2005. Accessibility and distance to previous deforestation were strong predictors of deforestation risk, while land use designation was less important. The model performed consistently well (AUC > 0.9), significantly better than random when we compared predicted deforestation risk to observed. If past deforestation rates continue, we estimate that 132,865 ha of forest could be lost by the year 2020, representing over 55 million Mg CO 2 . Conclusions Maxent provided a reliable method for identifying areas at high risk of deforestation and the major explanatory variables that could draw attention for mitigation action planning under REDD+. The tool is accessible, replicable and easy to use; all necessary for producing good risk estimates and adapt models after potential landscape change. We propose this approach for developing countries planning to meet requirements under REDD+.