BOGOR, Indonesia (18 October, 2011)_A metastudy by Porter-Bolland et al., shows how most peer-reviewed case studies have found community-managed forests to disappear less rapidly than strictly protected forests. By its very nature, a metasample is heterogeneous in time and space, and areas are not selected randomly by their case study authors. Alternatively, could one compare the fate of each protected forest with that of a similar matched forest without protection status?
This is the approach taken in a new study by Andrew Nelson from the International Rice Research Institute and Ken Chomitz from the World Bank’s Independent Evaluation Group. Lacking globally comparable deforestation data, the authors instead used forest fire incidence as a pantropical proxy of forest threats: In most regions, deforestation involves fire use.
They looked at high-resolution SPOT and MODIS satellite data, defining pixels with at least 25% tree cover as forest, and examined fire occurrence in the entire tropical biome of developing countries; 27% of the target area of almost 20 million km2 had some protection status.
In all three tropical continents, fire incidence from 2000 to 2008 was lower in protected areas than in unprotected ones. For instance, in Latin America and the Caribbean (LAC) fire had occurred on 7.4% of unprotected land. In strictly protected areas (IUCN categories I-IV), it was only 1.6%, four and a half times less.
In multiple-use protected areas (IUCN categories V-VI), fire rates were 3%, and in indigenous protected areas 1.5%. Similar proportions were observed for Africa; in Asia fire rates in strictly protected areas were somewhat higher (4.5%).
However, protected areas are typically located in more remote sites with less conversion pressures. Comparing their record directly with high-threat areas would thus ‘make them look too good’, compared to what counterfactually would have happened to these forests without protection status.
Matching can potentially correct this bias. Each protected pixel was compared to some non-protected ones with similar distance to roads and major cities, elevation, slope, and rainfall. The assumption was that deforestation pressures would be similar when these background variables were controlled for, thus isolating the true impact of the protection status.
The matched results show that protection still increases conservation effectiveness. Strictly protected areas still do between 2.0 (Asia) and 4.3 (LAC) percentage points better than their unprotected pairs. That’s less than in the crude, unmatched comparisons, though.
Multi-use areas at least preserve their unmatched protection premium (e.g. 6.4 points in LAC). However, indigenous areas (only present in LAC) apparently include many high-threat zones, and thus increase their advantage over matching unprotected lands to a smashing 13 percentage points.
What do the results mean for researchers? Applying matching methods is technically challenging. Selecting control variables can be controversial, especially in social sciences where theories are more complex. As we all know, finding the perfect match is not easy. Yet the notable differences between matched and unmatched results – and vis-à-vis literature reviews – underline just how important it can be to control for many potential sampling biases, in a world that progressively differentiates.
What does this mean for policymakers? While protected areas are facing some headwinds in the current debates, they are consistently doing a better job than unprotected ones in avoiding fires, and thus carbon and biodiversity losses – independent of continent, protection category and evaluation method.
Notably, multi-use and especially indigenous lands do even better than strictly protected lands, especially after matching (reinforcing a conclusion by Porter-Bolland et al.). So far, both categories are much more prominent in the Neotropics. This might also indicate a scope for diversifying protection strategies in Africa and Asia toward more ‘parks with people’, at least where local people’s low land-use expansion favours this approach.