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 Cote d'Ivoire

2026-07-09 · Source: Computer Vision and Pattern Recognition · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics, Geospatial AI for Agriculture · Depth: Expert, quick

Summary

A study in Cote d'Ivoire evaluated cocoa mapping performance across varying landscape conditions, comparing very high resolution (VHR) imagery, decametric Earth observation inputs, and operational products. Researchers developed models using 0.5 m Pleiades VHR imagery, a 10 m Sentinel-2 annual composite, and embeddings from TESSERA and AlphaEarth Foundations (AEF). They also assessed four publicly available cocoa mapping products. Performance was evaluated using 2,821 independently interpreted reference points, stratified by tree cover density and landscape fragmentation. The VHR model achieved the highest performance with an F1-score of 0.92, maintaining F1-scores above 0.90 across all strata. Among decametric inputs, TESSERA performed best (F1 = 0.86), followed by AEF (F1 = 0.82) and Sentinel-2 (F1 = 0.76). The Kalischek product was the best existing product (F1 = 0.83). Performance differences between VHR and decametric approaches were more pronounced in fragmented landscapes and under extreme tree cover densities.

Key takeaway

For Computer Vision Engineers developing large-scale agricultural monitoring systems, you should consider a hybrid approach for cocoa mapping. Prioritize targeted 0.5 m VHR imagery acquisition for highly fragmented or extreme tree cover density regions to achieve F1-scores above 0.90. For broader, scalable coverage, integrate foundation-model embeddings like TESSERA with 10 m decametric data, which offers F1-scores up to 0.86, providing a robust balance between accuracy and operational efficiency.

Key insights

Sub-metre VHR imagery significantly outperforms decametric inputs for accurate cocoa mapping, especially in complex landscapes.

Principles

• VHR maintains high accuracy across diverse landscape conditions.
• Foundation model embeddings enhance decametric mapping performance.
• Fragmentation increases VHR's advantage over decametric data.

Method

Models were developed using 0.5 m Pleiades VHR, 10 m Sentinel-2, and TESSERA/AEF embeddings, then evaluated with 2,821 stratified reference points.

In practice

• Use targeted VHR acquisition for highly fragmented cocoa areas.
• Employ foundation model embeddings for scalable large-area mapping.

Topics

• Cocoa Mapping
• Very High Resolution Imagery
• Sentinel-2
• Foundation Models
• Earth Observation
• Cote d'Ivoire

Best for: AI Scientist, Research Scientist, Computer Vision Engineer

Related on AIssential

• Context-Aware Slum Mapping in Sub-Saharan Africa Using Sentinel-1 Texture and Local Climate Zones — Integrating Sentinel-1 SAR textures with Sentinel-2 optical data significantly enhances informal settlement mapping accuracy and seasonal stability in Sub-Saharan Africa.
• CHMv2 is already supporting public sector efforts in the United States, Europe, and beyond. By making these advances open source, we aim to accelerate research and inform carbon offsetting, reforestation, and land management decisions globally. 🔗 Read t - x.com — CHMv2, an open-source global canopy height mapping model, enhances forest carbon quantification and land management.
• Self-Supervised Tree-level Biomass Estimation in Urban Environments From Airborne LiDAR and Optical Observations — Self-supervised deep learning combined with LiDAR and orthophotography enables accurate, scalable urban tree biomass estimation.
• MapSR: Prompt-Driven Land Cover Map Super-Resolution via Vision Foundation Models — MapSR enables high-resolution land-cover mapping with minimal annotation and compute by decoupling supervision from training.
• Multi-Scale ViT Inference with Habitat-Fit Priors and kNN Retrieval for Multi-Species Plant Identification — Multi-scale ViT inference combined with habitat-fit priors and kNN retrieval significantly improves multi-species plant identification.
• Integrating national forest inventory, airborne lidar, and satellite imagery for wall-to-wall mapping of forest structure with computer vision — Integrating lidar and satellite data with computer vision improves large-scale forest attribute mapping accuracy and consistency.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Computer Vision and Pattern Recognition.