Integrating Remote Sensing, Field-Measured Tree Heights, and Machine Learning to Enhance Mangrove Above-Ground Carbon Estimation in Baluran National Park, Indonesia

Abstract

Abstract Mangroves play a critical role in coastal ecosystem services, particularly through their capacity to sequester large amounts of atmospheric carbon, contributing to climate change mitigation. Developing accurate mangrove carbon models is therefore essential for monitoring ecosystem condition and carbon stocks at relevant scales. This study aimed to estimate mangrove Above-Ground Carbon (AGC) in Baluran National Park by integrating field measurements and remote sensing data within a Machine Learning (ML) framework. The study utilised an extensive field data collection programme of 60 sampling plots of girth at breast height, canopy cover, tree height, and tree density. Mangrove AGC was estimated using allometric equations. AGC was also modelled by processing satellite images, conducting statistical analyses, developing models with five ML algorithms (Random Forest (RF), Support Vector Machine (SVM), Decision Tree (DT), k-Nearest Neighbour (k-NN), and Gradient Boost (GB)), and checking accuracy using 5-fold cross-validation (CV) of Root Mean Square Error (RMSE). The RF model, using field-measured tree height, Ratio Vegetation Index (RVI), and Transformed Soil-Adjusted Vegetation Index (TSAVI), achieved the best performance ( R² training = 0.93, R² testing = 0.84, 5-fold CV RMSE = 12.20 Mg C ha⁻¹). Predicted AGC ranged from 5.39 to 57.18 Mg C ha⁻¹ (mean ± Standard Deviation (SD) = 30.43 ± 16.09 Mg C ha⁻¹) and showed improved accuracy compared to the global mangrove biomass dataset of (Simard et al., 2019). A key contribution of this study is the integration of field-measured tree height within a satellite-based ML framework, which enhances the accuracy and ecological relevance of AGC estimation compared to approaches relying solely on spectral predictors or remotely sensed canopy height products, offering a practical and cost-effective alternative for sites where UAV or LiDAR data are unavailable. This approach provides a practical method for regional mangrove carbon monitoring, national carbon accounting and supports climate change mitigation efforts.

Plant phenotyping relevance

マングローブの樹高・樹冠情報と衛星データを統合し、機械学習で個体・プロットレベルの地上部炭素量という植物状態を推定する手法を開発・交差検証しており、単なる生態系測定ではなく表現型取得手法が中心である。

Code and data availability

The authors state that all analysis code (model development, hyperparameter configuration, diagnostics, accuracy assessment) is publicly available in their GitHub repository Mangroves-AGC-Baluran, which reproduces this paper's mangrove AGC machine-learning analysis.