Open-Earth-Monitor Global Workshop 2025

Cocoa cultivation serves as a cornerstone of many agricultural economies across the globe, supporting millions of livelihoods and contributing significantly to global cocoa production. However, accurately mapping cocoa farm locations remains a challenging endeavor due to the complex and heterogeneous nature of the landscapes where cocoa is cultivated. Traditional mapping techniques often fall short in capturing the intricate spatial patterns of cocoa farming amidst dense vegetation, varying land cover types, farming practices and growing stages (Masolele et at., 2024). Moreover, the current mapping efforts mainly focus on two major producing countries, Ivory Coast, and Ghana (Kalischek et al., 2023). Thus, little is known about the location of cocoa farms in other cocoa producing regions, posing a challenge to the sustainability and economic contributions of the cocoa crop.
To address this challenge, we first present a benchmarking approach for mapping commodity crops worldwide. Here we compare different spectral, spatial, temporal and spatial-temporal methods for mapping commodity crops. The benchmarking is based on a variable combination of Sentinel-1 and Sentinel-2, locational and environmental variables (temperature and precipitation). We use a comprehensive list of reference data spanning 36 cocoa-producing countries to do this task. Higher accuracy (F1-score 87%) is obtained when using a model that employs spatial-temporal remote sensing images plus locational and environmental information, compared to other models without locational and environmental information.
Secondly, for demonstration, we employ the developed deep learning methodologies to map the locations of cocoa farms across the Globe with an F1-Score of 88%. By leveraging the rich spatio-temporal information provided by Sentinel-1 and Sentinel-2 satellite data, complemented by location encodings, temperature and precipitation data, we have developed a robust and accurate cocoa mapping framework. The developed deep learning algorithm extracts meaningful features from multi-source satellite imagery and effectively identifies cocoa farming areas. The integration of Sentinel-1 and Sentinel-2 data offers a synergistic approach, combining radar and optical sensing capabilities to overcome the limitations of individual sensor modalities. Furthermore, incorporating location encodings into the modeling process enhances the contextual understanding of cocoa farm distributions within their geographical surroundings.
Through this research effort, we provide the first high-resolution global cocoa map giving, valuable insights into cocoa farm locations, facilitating sustainable cocoa production practices, land management strategies, and conservation efforts across the pan-tropical forests, where cocoa farming occurs. The work aligns with recent European Union (EU) regulations to curb the EU market’s impact on global deforestation and provides valuable information for monitoring land use following deforestation, crucial for environmental initiatives and carbon neutrality goals (European Commission., 2024). Specifically, our product can support monitoring and compliance of the European Union (EU) Regulation on Deforestation-free Products (EUDR, No 2023/1115) by identifying the previous existing and current cocoa farm expansion after the cut-off date of December 31, 2020.
Within the framework of the ESA funded WorldAgroCommodities project, this mapping approach is now being converted into an operational cloud-based service on the Copernicus Data Space Ecosystem, allowing easy access to these crucial tools for the National Competent Authorities in light of enforcing the EUDR regulation. Furthermore, our findings hold significant implications for cocoa farmers, agricultural policymakers, and environmental stakeholders, paving the way for informed decision-making and targeted interventions to support the resilience, sustainability and traceability of cocoa farming systems worldwide.