Triple-feature fusion from UAV multispectral imagery enhances species-level mangrove carbon assessment

Abstract Accurate estimation of mangrove ecosystem carbon stocks is essential for effective blue carbon management. Significant interspecific variations in carbon storage capacity and estimation methods arise due to species-specific biophysical characteristics, highlighting the need for precise mangrove species identification and species-level carbon stock assessment. However, limited studies assessed mangrove carbon stocks at species-level. This study, conducted in the Gaoqiao Mangrove Nature Reserve in Zhanjiang, Guangdong Province, applied UAV multispectral technology to simultaneously acquire spectral, structural and textural vegetation feature variables for mangrove species identification and established species-specific carbon stock models, thereby achieving species-level carbon stock estimation. Results showed that (1) by integrating spectral and structural features, the study achieved 89.87% overall accuracy in species identification. (2) Species-level carbon stock estimation models, incorporating spectral, structural and textural feature variables alongside field-measured carbon data, demonstrated strong predictive performance (R2 = 0.48-0.95). (3) The most effective vegetation feature variables for carbon estimation varied significantly across species, emphasizing the necessity of accounting for species heterogeneity in mangrove carbon stock estimations. (4) Carbon stocks exhibited significant interspecific variation, with Rhizophora stylosa demonstrating the highest aboveground (97.06 t hm⁻2) and belowground (37.22 t hm⁻2) stocks, compared to Aegiceras corniculatum’s minimum values of 49.14 and 19.88 t hm⁻2, respectively. This study established a UAV-based multispectral framework for mangrove species-level carbon stock estimation and provided new insights for mangrove carbon assessment and management by demonstrating the importance of considering species-specific influences on carbon stocks and their estimation.