Robert Masolele
Geoinformation Scientist leveraging AI and satellite technology to map our planet's changing landscape. I transform petabytes of satellite data into actionable insights on agricultural expansion, deforestation, and biodiversity, bridging the gap between advanced algorithms and environmental policy.
My work focuses on developing scalable deep learning models to monitor commodity crops (e.g., oil palm, cocoa, coffee, soy) and assess their environmental impacts, using a combination of radar, optical imagery, and cloud computing.
Sessions
Robert N Masolele1, Katja Berger2, Zoltan Szantoi3, Camilo Zamora2, Johannes Reiche1
1 Wageningen University, Wageningen, The Netherlands; robert.masolele@wur.nl
2 GFZ, German GeoResearch Center Potsdam, Germany
3 Directorate of Earth Observation Programmes, European Space
Agency (ESA), Frascati, RM, Italy
Coffee cultivation underpins agricultural economies worldwide, supporting millions of livelihoods and contributing significantly to global production [1]. At the same time, coffee is among the leading commodities associated with global deforestation risks linked to European Union (EU) consumption. However, accurately mapping coffee farm locations remains challenging due to the heterogeneous landscapes in which coffee is grown, including dense vegetation, diverse land cover types, varying management practices, and phenological stages [2], [3], [4]. Existing mapping efforts are largely limited to major producers such as Brazil, Vietnam, Ethiopia, and Colombia, leaving substantial gaps across other coffee-growing regions [5].
To address this, we first present a global benchmarking framework for commodity crop mapping. We evaluate a combination of Sentinel-1 and Sentinel-2 data, alongside locational variables. Using a comprehensive reference dataset spanning >40 coffee-producing countries, we show that models integrating Sentinel-1 and Sentinel-2 data with location encoding achieve the highest performance (F1-score: 89%), outperforming models without contextual information [4].
Building on this, we apply the best-performing deep learning framework to generate the first high-resolution global map of coffee farm extent, achieving an F1-score of 86%. The integration of Sentinel-1 (radar) and Sentinel-2 (optical) data enables robust feature extraction across diverse conditions, while location encodings enhance geographic contextualization of coffee systems.
This work delivers a consistent, high-resolution global coffee map, supporting sustainable land management, supply chain transparency, and conservation in tropical regions. It directly aligns with the EU Deforestation Regulation (EUDR, Regulation (EU) 2023/1115), which requires monitoring the deforestation footprint of seven key commodities, including coffee relative to the December 31, 2020 cut-off date. The approach is being operationalized within cloud-based platforms (e.g., Copernicus Data Space Ecosystem), facilitating access for policymakers, certification bodies, and stakeholders.
[1] R. Grüter, T. Trachsel, P. Laube, and I. Jaisli, ‘Expected global suitability of coffee, cashew and avocado due to climate change’, PLoS One, vol. 17, no. 1, p. e0261976, Jan. 2022, doi: 10.1371/JOURNAL.PONE.0261976.
[2] D. A. Hunt et al., ‘Review of Remote Sensing Methods to Map Coffee Production Systems’, Remote Sensing 2020, Vol. 12, Page 2041, vol. 12, no. 12, p. 2041, Jun. 2020, doi: 10.3390/RS12122041.
[3] G. Maskell, A. Chemura, H. Nguyen, C. Gornott, and P. Mondal, ‘Integration of Sentinel optical and radar data for mapping smallholder coffee production systems in Vietnam’, Remote Sens. Environ., vol. 266, Dec. 2021, doi: 10.1016/j.rse.2021.112709.
[4] R. N. Masolele et al., ‘Mapping the diversity of land uses following deforestation across Africa’, Sci. Rep., vol. 14, p. 1681, 2024, doi: 10.1038/s41598-024-52138-9.
[5] A. Escobar-López, M. Á. Castillo-Santiago, J. F. Mas, J. L. Hernández-Stefanoni, and J. O. López-Martínez, ‘Identification of coffee agroforestry systems using remote sensing data: a review of methods and sensor data’, Geocarto Int., vol. 39, no. 1, p. 2297555, 2024, doi: 10.1080/10106049.2023.2297555;WGROUP:STRING:PUBLICATION.