Aboveground Carbon Estimation in a Mangrove Ecosystem Using UAV-Based Remote Sensing and Machine Learning
| dc.contributor.author | Duan, M. | |
| dc.contributor.author | Sanchez-Azofeifa, A. | |
| dc.contributor.author | Abdulmajeed, M. | |
| dc.contributor.author | Turner, D. | |
| dc.contributor.author | Buckingham, K. | |
| dc.contributor.author | Odari, A. | |
| dc.contributor.author | Mtwana, J. | |
| dc.contributor.author | Kipkoech, S. | |
| dc.contributor.author | Kasraee, N. | |
| dc.coverage.spatial | Kenya | |
| dc.date.accessioned | 2026-11-11T12:26:09Z | |
| dc.date.issued | 2025 | |
| dc.description.abstract | Mangroves are an important part of coastal blue carbon ecosystems, efficiently absorbing atmospheric carbon dioxide (CO 2 ). Accurate quantification of mangrove carbon stocks aids climate change mitigation and adaptation strategies. This study uses UAV-based remote sensing datasets to model Aboveground Carbon (AGC) in a juvenile mangrove ecosystem in Kenya, characterized by relatively open canopies. We developed an Ensemble regression model to estimate AGC, achieving a Mean Absolute Error (MAE) of 1.79 kg when validated against ground truth data. Instead of plot-level metrics, which lack detailed spatial information about individual trees or areas smaller than the mapping unit, our model was developed based on tree-level metrics, using data on hundreds of trees from fewer forest inventory plots. This methodology enabled the extraction of detailed spatial information on AGC at the tree level. We also explored the potential of two different UAV-based remote sensing data (LiDAR point cloud data vs point cloud data generated from high overlap images) for estimating mangrove AGC. Furthermore, a pixel-level comparison of difference values (“AGC LiDAR – AGC High overlap ”) was conducted to quantify and evaluate the estimated AGC differences (R 2 = 0.71, RMSE = 0.97 kg/m 2 ). The results suggest that both LiDAR data and superior high overlap images have the potential to accurately predict mangrove biomass/carbon stocks, although LiDAR outperforms high overlap images due to its involvement in unique intensity metrics. The tree-level-based modeling methodology presented in this work offers a different insight for biomass or carbon stock modeling. | |
| dc.identifier.citation | Duan, M., Sanchez-Azofeifa, A., Abdulmajeed, M., Turner, D., Buckingham, K., Odari, A., Mtwana, J., Kipkoech, S., & Kasraee, N. (2025). Aboveground Carbon Estimation in a Mangrove Ecosystem Using UAV-Based Remote Sensing and Machine Learning. Ecological Indicators, 178, 113950. https://doi.org/10.1016/j.ecolind.2025.113950 | |
| dc.identifier.doi | 10.1016/j.ecolind.2025.113950 | |
| dc.identifier.uri | https://repository.kmfri.go.ke/handle/123456789/645 | |
| dc.language.iso | en | |
| dc.publisher | Elsevier BV | |
| dc.relation.ispartof | Ecological Indicators | |
| dc.relation.uri | https://doi.org/10.1016/j.ecolind.2025.113950 | |
| dc.rights | CC BY 4.0 | |
| dc.source | OpenAlex and Crossref metadata, Blue Carbon bulk upload 2026-11-11 | |
| dc.subject | Mangroves | |
| dc.subject | Monitoring | |
| dc.title | Aboveground Carbon Estimation in a Mangrove Ecosystem Using UAV-Based Remote Sensing and Machine Learning | |
| dc.type | Journal article |
