Mapping smallholder plantation as a key to sustainable oil palm: A deep learning approach to high-resolution satellite imagery

Abstract

The adverse impacts of oil palm plantations on socio-ecological conditions in Indonesia are a prominent issue. While farming improvement in large-scale plantations has shown good progress, accelerated expansion coupled with low productivity of smallholder plantations has raised concerns. Smallholder plantations are rarely monitored, as they are small, scattered, fragmented, often located in remote areas, with irregular shapes, susceptible to rapid change, and commonly mixed with other commodities. Accordingly, the present study aimed to (1) develop a new model for mapping smallholder oil palms, (2) describe the spatial distribution of smallholder oil palms, and (3) identify their potential impact on the environment. An object detection model based on a deep-learning approach was applied to high-resolution satellite imagery from Google Earth, Pleiades, and GeoEye to detect individual oil palm trees. The results showed that the derived model is highly accurate and can successfully map smallholder plantations. Village-level maps showed that the spatial distribution of smallholder oil palms was strongly related to local socio-ecological dynamics, whereas regency-level maps showed that smallholder oil palms have largely encroached on conservation zones and forested areas. These findings highlight the urgency of smallholder oil palm spatial data for reaching sustainable global palm oil industries.