RESUME
A PROPOS
La Revue Africaine d’Environnement et d’Agriculture est placée sous l’autorité scientifique du Comité de Rédaction
et sous l’autorité administrative des ASBL CABD (Centre d’Assistance des Communautés de Base pour le Développement Durable),
GERADIB (Groupe d’Etudes et de Recherches Agropastorales pour le Développement de Bandundu) et SOFT AFRICA.
Utiliser cet identifiant pour créer un lien vers cet article :
https://?pages=article&id=335| Titre : | Canopy density spatial model and binary classification of mangrove presence using the fusion of Landsat composite and Top of Atmosphere Sentinel-2 images in the Parc Marin des Mangroves (Moanda, Democratic Republic of Congo) |
| Auteur(s): | Raymond Lumbuenamo Sinsi, Elie Nsimba Ngembo, Cedric Kompani Daba, Eric Lutete Landu, Jerry Manisa Bomingi, Arielle Mabaya1, Arsène Kayengi Baziota, Hippolyte Ditona Tsumbu, Bonaventure Lele Nyami, Roger Ntoto M’Vubu |
| Mots-clés: | Mangrove mapping, remote sensing, machine learning classifiers, Parc Marin des Mangroves/DRC. |
| Date de publication | 2025-06-28 14:21:03 |
| Resumé : | Description of the subject. The Parc Marin des Mangroves (PMM), a protected coastal area in the Congo Basin, is under increasing anthropogenic pressure and lacks sufficient spatial data on mangrove presence or a canopy density model.
Objective. This study contributes to the knowledge of the mangrove spatial distribution. It aims to (a) map mangrove presence using machine learning classifiers, and (b) generate a spatial model of mangrove canopy density. Methods. High-resolution Planet NICFI imagery was visually interpreted using a systematic sampling approach in Collect Earth Online. Three hundred twenty-four sampling plots, each 50 m × 50 m and spaced 1.5 kilometers apart, were systematically subdivided into dots, with each classified as either mangrove or non-mangrove. The proportion of mangrove dots per plot was interpolated to create a canopy density model. A binary classification of mangrove presence was performed using three classifiers: Random Forest (RF), Support Vector Machine (SVM), and Classification and Regression Trees (CART), and validated through a holdout validation. A Global Mangrove Watch (GMW) mask was applied to standardize the location of mangrove extent. Results. All classifiers achieved an accuracy of over 90 %. CART showed slightly higher performance metrics but did not significantly outperform RF and SVM. Empirical Bayesian Kriging produced the most accurate canopy density model, with the lowest root mean square error. Between 2016 and 2022, canopy density trends revealed signs of degradation. Mean mangrove area estimates (~209 km²) remained constant across classifiers after applying the GMW mask. The mean mangrove area, as determined by the three classifiers before mask application, is 265.16 km² for 2016 and 269.96 km² for 2022. Conclusion.These findings confirm the reliability of machine learning models for mangrove mapping and highlight early signs of mangrove degradation and deforestation in the PMM using canopy density analysis. |
| Editeur : | RAFEA |
| DOI : |
https://dx.doi.org/10.4314/rafea.v8i2.14 |
Document pour cet article:
| Fichier | Description | Taille | Format | |
|---|---|---|---|---|
| ARTICLE-RAFEA | OPEN ACCESS | 1223 ko | Adobe PDF | Lire article |