A Development of Deep Learning Approach to Estimate Blue Carbon Stock on Mangrove Habitats

Deep learning approaches have been widely used to detect objects through aerial image data. The common utilization of deep learning is to detect forestry objects through primary data (e.g., drone-based data acquisition) or secondary data (e.g., satellite-based images). This study developed a novel method to detect the 3-dimensional attribute (height and canopy - i.e., length and width) of mangroves and use them to estimate the potential blue carbon stock at the coast of Karawang region, Indonesia. The mangrove detection is conducted through a deep learning approach called YOLOv8 based on drone-based aerial images. Meanwhile, the blue carbon stock is estimated according to the canopy height model. The results indicated that YOLOv8 has a better performance in detecting mangroves compared to other similar methods based on the Mean Average Precision (mAP) using datasets in this study. Finally, the potential blue carbon stock in the Karawang region coastal area is estimated at 1,569,733.46 Mg/ha.