Global tree cover products face challenges in accurately predicting tree cover across biophysical gradients, such as precipitation or agricultural cover. To generate a natural forest cover map for Costa Rica, biases in tree cover estimation in the most widely used tree cover product (the Global Forest Change product (GFC) were quantified and corrected, and the impact of map biases on estimates of forest cover and fragmentation was examined. First, a forest reference dataset was developed to examine how the difference between reference and GFC-predicted tree cover estimates varied along gradients of precipitation and elevation, and nonlinear statistical models were fit to predict the bias. Next, an agricultural land cover map was generated by classifying Landsat and ALOS PalSAR imagery (overall accuracy of 97%) to allow removing six common agricultural crops from estimates of tree cover. Finally, the GFC product was corrected through an integrated process using the nonlinear predictions of precipitation and elevation biases and the agricultural crop map as inputs. The accuracy of tree cover prediction increased by ≈29% over the original global forest change product (the R2
rose from 0.416 to 0.538). Using an optimized 89% tree cover threshold to create a forest/nonforest map, we found that fragmentation declined and core forest area and connectivity increased in the corrected forest cover map, especially in dry tropical forests, protected areas, and designated habitat corridors. By contrast, the core forest area decreased locally where agricultural fields were removed from estimates of natural tree cover. This research demonstrates a simple, transferable methodology to correct for observed biases in the Global Forest Change product. The use of uncorrected tree cover products may markedly over- or underestimate forest cover and fragmentation, especially in tropical regions with low precipitation, significant topography, and/or perennial agricultural production.
This is an open access article distributed under the Creative Commons Attribution License
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited