Is Sub-Metre Resolution Necessary for Cocoa Mapping? A Landscape-Stratified Evaluation of Very High Resolution Imagery, Decametric Earth Observation Inputs, and Operational Products in Côte d'Ivoire

Jul 16, 2026

Upload by Riley

Accurate cocoa mapping supports deforestation monitoring, supply-chain transparency, and regulatory compliance, yet medium-resolution Earth observation can underperform in heterogeneous smallholder areas due to spatial aggregation effects. In Côte d’Ivoire, landscape conditions were used to test whether very high resolution imagery improves cocoa detection and whether foundation-model embeddings enhance decametric mapping. Models used 0.5 m Pléiades VHR, a 10 m Sentinel-2 annual composite, TESSERA/AEF embeddings, and four external products. A landscape-stratified assessment across 2,821 reference points showed VHR achieved the strongest F1 (0.92), with decametric results and foundation-model performance varying by fragmentation and tree cover density.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages40

Reading Time1 h 41 min

Introduction

Cocoa mapping need and regulatory relevance

Challenges in heterogeneous tropical landscapes

Methods and data

EO inputs and foundation-model embeddings

External cocoa products evaluated

Landscape-stratified accuracy assessment

Results

Performance across image resolutions and models

Effects of fragmentation and tree cover density

Discussion and implications

Title Is sub-metre resolution necessary for cocoa mapping? A landscape-stratified evaluation of very high resolution imagery, decametric Earth Observation inputs , and operational products in Côte d'Ivoire Authors Kasimir Orlowski 1,2 , Filip Sabo3 ,* , Michele Meroni4 , Astrid Verhegghen5,6 , Mariana Belgiu 1 , Felix Rembold3 1 Faculty of Geo-Information Science and Earth Observation (ITC), University of Twente, Hallenweg 8, 7522 NH Enschede, The Netherlands. 2 FINCONS S. P.A. , Via Torri Bianche 10, Palazzo Betulla, 20871 Vimercate, Italy. 3 European Commission, Joint Research Centre (JRC), Via Enrico Fermi, 2749, 21027 Ispra, Italy. 4Seidor Consulting, Carrer de Veneçuela 74, 08019 Barcelona , Spain. 5ARHS Developments Italia S. R. L. , Via Fratelli Gabba 1/A, 20121 Milan , Italy. 6 Flemish Institute for Technological Research (VITO), Boeretang 200, 2400 Mol , Belgium. *Corresponding author: filip.szabo@ec.europa.eu Abstract Accurate cocoa mapping is increasingly important for deforestation monitoring, supply-chain transparency, and regulatory applications. Spatial aggregation in conventional medium-resolution Earth observation (EO) imagery may limit cocoa detection in heterogeneous smallholder landscapes. In Côte d’Ivoire, we therefore evaluated how mapping performance varies across landscape conditions, whether very high resolution (VHR) imagery provides a meaningful advantage, and whether foundation-model embeddings improve decametric cocoa mapping. We developed models using 0.5 m Pléiades VHR imagery, a 10 m Sentinel-2 annual composite, and embeddings from TESSERA and AlphaEarth Foundations (AEF), and additionally assessed four publicly available cocoa mapping products. Performance was evaluated through a landscape-stratified accuracy assessment using 2,821 independently interpreted reference points distributed across gradients of tree cover density and landscape fragmentation. The VHR model achieved the highest performance (F1 = 0.92) and maintained F1-scores above 0.90 across all strata. Among the decametric inputs, TESSERA performed best (F1 = 0.86), followed by AEF (F1 = 0.82) and Sentinel-2 (F1 = 0.76) . Of the existing cocoa products, the Kalischek product performed best (F1 = 0.83), comparable to the internally trained AEF model. Performance differences between VHR and decametric approaches increased with fragmentation and under low and high tree cover density conditions. Targeted VHR acquisition may therefore be particularly beneficial in complex cocoa landscapes, while foundation-model embeddings offera scalable alternative for large-area mapping. Keywords Cocoa mapping; Very high resolution imagery; Geospatial foundation models; Tree cover density; Landscape fragmentation; Stratified accuracy assessment; Deep learning 1 . Introduction Cocoa cultivation supports the livelihoods of millions of smallholder farmers across tropical regions while simultaneously representing one of the most important drivers of tropical deforestation in West Africa (Squicciarini and Swinnen, 2016; Kalischek et al. , 2023; Masolele et al. , 2024; Renier et al. , 2025; Singh and Persson, 2026) . Historically, cocoa expansion has followed forest frontier dynamics, where newly cleared land initially provides high yields that decline over time as soils degrade and plantations age (Clough et al. , 2009; Ruf and Siswoputranto, 1995) , triggering further expansion and deforestation. Côte d'Ivoire, the world's largest cocoa producer, exemplifies this pattern: the country contained an estimated 44,500 km² of cocoa cultivation in 2021 and has experienced extensive cocoa-associated forest loss over recent decades (Kalischek et al. , 2023; Masolele et al. , 2024; Renier et al. , 2025) . The European Union Regulation on Deforestation-Free Products (EUDR, Regulation (EU) 2023/1115) requires that cattle, cocoa, coffee, palm oil, soya, wood, and rubber placed on the EU market are not associated with deforestati

FAQs

What question does the study address about cocoa mapping?

The study tests whether sub-metre (very high resolution) imagery is necessary for cocoa mapping and whether decametric approaches can be improved using foundation-model embeddings and compared against existing operational products.

Which data sources and modelling approaches are evaluated?

Evaluated inputs include 0.5 m Pléiades VHR imagery, a 10 m Sentinel-2 annual composite, and embeddings from TESSERA and AlphaEarth Foundations, alongside four publicly available cocoa mapping products.

How was model performance measured and stratified?

Performance was assessed using a landscape-stratified accuracy approach based on 2,821 independently interpreted reference points spanning gradients of tree cover density and landscape fragmentation.

Jul 16, 2026

Upload by Riley