Fine-scale carbon stocks mapping in the mangrove forests of Tumaco, Colombia using machine learning and remote sensing approaches

Mangroves play a critical role in mitigating climate change, sequestering up to five times more carbon than other forests. Accurate assessment of their carbon stocks is crucial for effective planning and management in climate change strategies. This study presents an innovative approach that integrates remote sensing with field data, utilizing high-resolution imagery and evaluating two machine learning algorithms: Random Forest and Support Vector Regression. Mangrove area was mapped using supervised classification, and both aboveground and belowground biomass, along with the carbon stored in these compartments, were quantified. The classification achieved an accuracy of 87%, and mean values of 192.50±102.78 Mg/ha for aboveground biomass, 79.95±56.85 Mg/ha for belowground biomass, and 127.43±73.49 CMg/ha for stored carbon. The Random Forest model performed best, with an RMSE of 140.68 and an R² of 0.78, surpassing global models. Additionally, spectral indices significantly enhanced the model’s ability to predict aboveground biomass.