Detecting oil palm plantations and mangroves overlooked by IGBP land-cover products with “ground-truthing from the sky” across Southeast Asia

Tropical deforestation in Southeast (SE) Asia is largely driven by expansion of oil palm plantations, which are difficult to detect with global- or regional-scale remote-sensing products. This paper demonstrates a straight-forward method for “ground-truthing from the sky” with very high resolution (VHR) photographic images to detect oil-palm cultivation and other drivers of deforestation across SE Asia. The MCD12Q1 land-cover is a widely used land-cover product that generates annual 500-m-pixel maps of 17 IGBP (International Geosphere-Biosphere Programme) classes. While this resolution is practical for regional assessments across SE Asia (∼4.5 million km 2 ), it is too coarse to identify fine-scale, agriculture-driven conversions such as oil-palm expansion. Our approach was to integrate MCD12Q1 data with VHR imagery from Google Earth by random subsampling within each major land cover type to which forests had transitioned during two periods (2001-2010 and 2010-2018). Each of the 260 verification points in each post-transition land type was visually interpreted, from which we generated correction factors to calibrate land-use change probabilities. Whereas uncalibrated results showed that 12% and 9% of forested areas were converted to woody savanna and savanna, VHR-images showed that 32% to 46% of them were oil palm plantations. After application of these correction factors, we estimated that 40 % and 48% of forest losses were due to oil palm expansion, whereas the remaining forest loss was attributed to degradation to savannas and grasslands. Additionally, permanent wetland classified by MCD12Q1 was found to consist of mangroves (63%), aqua farms (20%), and oil palms (8%) rather than lakes, rivers and marshes in the region. While MODIS remains a valuable source for analyzing land use changes across large areas, detection of deforestation driven by agricultural activities benefits from calibration with VHR imagery. Our approach is straightforward and requires minimal expertise, making it easily adoptable by local governments, NGOs, land managers and others.