Beyond "forest" change: assessing and improving global forest cover products

Global forest cover change products are the foundation for broad-scale analyses of forest loss and gain, but their capacity to accurately monitor forest regrowth has rarely been assessed. Systematic biases in forest regrowth maps could have important consequences for global estimates of carbon sequestration and habitat. In this study, we use two distinct lines of evidence to evaluate the accuracy and biases of the most widespread current global forest change product, the high-resolution Hansen global forest change map (HGFC; Hansen et al. 2013; 3000+ citations). First, we tested the ability of the HGFC to detect regrowth of tree cover using visual assessment of Landsat time series and high-resolution imagery at 3000 points across the humid tropics. As has been found previously, the Hansen gain product mapped tree cover, not natural forest cover, with relatively high errors of omission. Importantly, we found no bias towards detecting the regrowth of agricultural tree cover, in comparison to natural regrowth. Second, we assessed the degree to which agricultural tree plantations dominate the regrowth detected by the HGFC product. By integrating Landsat and SAR image metrics with a large training dataset of tree plantations and secondary forests, we reclassified the HGFC gain product using machine learning into two new classes: plantation or natural regrowth (>80% mean class accuracy). Most large regrowth patches mapped by the HGFC in the humid tropics consisted of agricultural tree plantations, and agricultural tree plantations dominated the mapped regrowth area in certain regions (e.g., southeast Asia and southeastern Brazil). We present global and per-country statistics on the occurrence of regrowth in agricultural tree cover (i.e., tree crop expansion) and the regrowth of natural forests. We conclude that while the Hansen global forest cover product is not biased towards the detection of agricultural regrowth in the humid tropics, its use as a map of natural forest regrowth should be limited in regions dominated by the production of tree crops.