Tropentag, September 16 - 18, 2026, Göttingen "Towards multi-functional agro-ecosystems promoting climate-resilient futures"

Spatial mapping of associated trees in cocoa agroforestry systems using remote sensing: Evidence from Côte d’Ivoire

University of Würzburg, Remote Sensing, Germany

Abstract

Côte d’Ivoire cultivates cocoa either as monoculture or in association with forest trees within cocoa agroforestry. Cocoa agroforestry offers important advantages in terms of long-term sustainability, biodiversity conservation and climate resilience. Furthermore, the adoption of the EU Deforestation Regulation (EUDR), which restricts the import of products linked to deforestation into the EU market, may enable farmers to negotiate premium prices for sustainability produced cocoa. The aim of the study was to propose a Earth Observation (EO)-based approach for mapping associated trees in cocoa plantations. This work contributes to improving the economic situation of local farmers while supporting sustainable cocoa production in Côte d’Ivoire. The study area was conducted in the Gueyo region in southern Côte d’Ivoire, characterised by a Guineo-Congolian climate. Sentinel^-1 and Sentinel^-2 data (10 m spatial resolution) together with high resolution NICFI Planet imagery (4.7 m spatial resolution). Reference data consisted of field polygons from farmers in the study area. The classification workflow was implemented in two stages. First, a binary cocoa/non-cocoa map was generated using Sentinel data. Vegetation indices and texture metrics were derived and used to train a Random Forest classifier (RF). The resulting cocoa map was then used as a mask for the Planet imagery. In the second stage, sampling points representing tree/non-tree areas were collected with the cocoa boundaries, and a tree/non-tree classification was performed using RF. Tree cover percentage was estimated and adjusted for each farmer's plot. The proposed approach achieved an Overall Accuracy (OA) of 0.87 for cocoa mapping, while the associated tree classification reached an OA of 0.91. Tree cover percentages were reported successfully for each farmer plot, and a spatially explicit tree cover density map of cocoa agroforestry systems was produced for the study area. The study demonstrates the potential of combining Sentinel and high resolution Planet imagery to generate 10 m resolution products capable of identifying cocoa agroforestry systems. Such information can support farmers, research institutions, and local authorities by improving the monitoring and spatial characterisation of sustainable cocoa production landscapes.

Keywords: Agroforestry, cocoa mapping, Côte d’Ivoire, planet, Sentinel, tree cover density

Contact Address: Dan Emmanuel Kanmegne Tamga, University of Würzburg, Remote Sensing, Würzburg, Germany, e-mail: kanmegnedangmail.com