In global tree cover products, how much forest change is actually agricultural expansion?

Global tree cover products are widely used for analyses of deforestation, fragmentation, and conservation decision making, especially in the tropics. Critical assessments of global forest products have pointed out that misclassification of agricultural land uses as forest is widespread, but the impact of this misclassification on estimates of agricultural expansion is not well understood. To what extent are net expansion estimates inflated by "false deforestation" of agricultural fields, or lowered by "false reforestation" of agricultural fields? In this study, we examine agricultural expansion in a tropical country with a diverse set of agricultural systems, Costa Rica. We ask a simple question: for two global tree cover maps, what proportion of mapped tree cover loss, or gain, was attributable to agricultural expansion between 2000 and 2015? We first assess map accuracy in georeferenced agricultural parcels, finding that that both the Hansen GFC and Sexton GCS global tree cover predictions over-predict tree cover in agricultural land in the years 2000 and 2015, across both row and tree crops. Next, we assess agricultural expansion by generating two highly accurate (>90%) agricultural land-use maps for 2000 and 2015, derived from Landsat and ALOS Palsar data. We show that both Hansen and Sexton markedly overestimate recent rates of agricultural expansion into forest (forest loss to agriculture) in Costa Rica. We further show that cropland expansion outside forest makes up the majority of estimated forest gains. Errors were spatially concentrated near tree crops such as banana and oil palm, but small errors were common even in open row crop systems. Accuracy was highest in forest-dominated or forest-pasture landscapes. Studies that uncritically use global tree cover products as a baseline for monitoring agricultural expansion should restrict their inferences to pastoral systems not dominated by row or tree crops.