Students and satellites are helping close Indonesia’s land cover data gap

Oil palm is the biggest driver of deforestation in Indonesia. And yet the maps we use to track it are often unreliable, out of date or locked behind paywalls.

In 2025, RECOFTC completed a project to address this gap. University students were trained to help build a free, open land cover dataset that anyone can access and use.

From classrooms to field sites

In January 2025, RECOFTC and the Spatial Informatics Group ran three-day GIS training workshops at two Indonesian universities – Hasanuddin University in South Sulawesi and the University of Riau. Sixty students learned how to read satellite images, identify land cover types, and digitize what they saw using mapping software.

Following the training, 30 of the top-performing students were engaged as paid data interpreters for five months. Over that time, they annotated satellite images across Riau and West Sulawesi, labelling every land cover feature they could identify – oil palm, forest, rice fields, rubber, coconut groves and more.

The students also visited field sites to check if what the satellites showed matched what was actually on the ground. In West Sulawesi, those visits included conversations with local communities about how they use and manage their land.

Building evidence for better decisions

The project originally aimed to produce at least 5,000 labeled land cover patches. The final count is 52,225 polygons covering nearly 260,000 hectares across 15 land cover types. The oil palm class alone – the one that matters most for deforestation monitoring – achieved 96 per cent completeness and 88 per cent precision in independent accuracy testing.

The dataset is freely available through Zenodo (DOI: 10.5281/zenodo.15618532) in standard GIS formats, with interactive visualization tools on Google Earth Engine. A scientific paper describing the methodology has also been submitted for peer review.

Open data for communities and policymakers

Most commercial oil palm maps are proprietary. They are not transparent about their methods, governments often cannot officially recognize them, and the communities most affected by land cover change rarely have access to them at all. This dataset was built in Indonesia, by Indonesian students, in landscapes where social forestry communities depend on knowing what is happening to their forests. It is free to use, fully documented, and designed to be built on.