Estimating Above-Ground Blue Carbon in a Tropical Mangrove Using Multispectral Unmanned Aerial Vehicle Imagery

Technical Summary: Estimating Above-Ground Blue Carbon in a Tropical Mangrove Using Multispectral Unmanned Aerial Vehicle Imagery

Problem Statement
Mangrove forests are critical blue carbon ecosystems, storing significant amounts of carbon in their biomass. However, accurately quantifying this carbon stock is logistically challenging due to the difficult terrain of mangrove environments (unconsolidated mud, dense aerial roots) and the limitations of traditional field inventories, which are time-consuming and cover small areas. While satellite imagery and LiDAR-equipped drones offer alternatives, there is a scarcity of methodologies that utilize multispectral drone imagery to derive structural parameters (like canopy dimensions and height) for biomass estimation in complex, heterogeneous mangrove forests. Furthermore, existing global carbon estimates often lack the resolution to capture local variability, particularly in degraded or anthropogenically pressured systems like the Suape estuarine bay in Brazil.

Methodology
The study developed a reproducible workflow to estimate Above-Ground Biomass (AGB) and carbon stock using multispectral Unmanned Aerial Vehicle (UAV) imagery and Object-Based Image Analysis (OBIA).

• Data Acquisition: A DJI P4 Multispectral drone captured high-resolution imagery (RGB, Blue, Green, Red, Red Edge, and Near-Infrared) over a 31-hectare area in Suape Bay, Brazil. A Canopy Height Model (CHM) was derived by subtracting the Digital Terrain Model (DTM) from the Digital Elevation Model (DEM) using Structure from Motion (SfM) photogrammetry.
• Segmentation Workflow: A four-stage OBIA workflow was implemented to delineate individual tree-like objects (TLOs):
  1. Seed Generation: "VegITD" seeds (potential treetops) were generated using a local maxima algorithm on the CHM, filtered by NDVI. "NVeg" seeds (non-vegetated areas) were manually digitized.
  2. Pixel Clustering: Region-growing algorithms expanded these seeds into spectrally coherent objects using thresholds based on NDVI (for vegetation) and NDWI (for water/sediment).
  3. Crown Identification: Multiresolution segmentation identified treetops and canopy gaps using spectral thresholds in the Red Edge and Near-Infrared bands.
  4. Crown Shape Delineation: A second segmentation stage expanded confirmed treetops into full crown boundaries, constrained by geometric criteria (Shape Index < 2) and area limits to prevent merging of adjacent crowns.
• Biomass Estimation: Five allometric equations were evaluated. Three relied on parameters derivable from UAV data (Canopy Diameter, Canopy Area, Height, and Vegetation Indices: GNDVI, SAVI, SR). Two relied on Diameter at Breast Height (DBH), which was estimated via a linear regression between field-measured height and DBH.
• Validation: Ground truth data was collected in a 225 m² vegetation plot (P1) containing 58 individuals (mostly Rhizophora mangle). The UAV-derived TLOs within this plot were compared against field measurements of tree count and height.

Key Contributions

• Novel Segmentation Strategy: The paper introduces a hybrid OBIA approach that integrates structural seeds (from CHM local maxima) with spectral constraints (from multispectral reflectance) to delineate individual tree crowns in dense mangroves. This method addresses the "false negative" problem of standard treetop detection algorithms by identifying additional canopy tops based on spectral reflectance (VegDN) where structural peaks were not detected.
• Equation Selection: The study systematically evaluated five allometric equations. It determined that equations relying on DBH (estimated via regression) or vegetation indices alone were less suitable for this specific application. Instead, an equation based directly on canopy area and height (AGB2) provided the most plausible estimates for area-wide application, as these parameters are directly observable by the UAV without intermediate regression steps.
• High-Resolution Baseline: The research provides a high-resolution carbon stock baseline for a mangrove system subjected to intense anthropogenic pressure (port and industrial activities), demonstrating that even degraded systems retain significant carbon stocks.

Results

• Segmentation Performance: The workflow successfully delineated 5,363 TLOs over a 7.5-hectare vegetated area. Validation in the ground plot showed that the UAV detected 13 TLOs, corresponding closely to the 13 tallest trees in the plot (mean height difference of ~0.17 m), confirming the method's ability to map dominant canopy trees.
• Carbon Stock Estimates: The study estimated the above-ground carbon stock of the Suape mangrove at 79.8 MgC.ha⁻¹ using the selected AGB2 equation.
• Spatial Distribution: Carbon stocks were not uniform; they were stratified, with higher values concentrated in trees adjacent to estuarine channels and decreasing towards the upper mangrove and hypersaline tidal flats.
• Comparison: The estimated value (79.8 MgC.ha⁻¹) was higher than some regional estimates (e.g., 50 MgC.ha⁻¹ found 64 km north) but lower than global averages (109.3 MgC.ha⁻¹). The study noted that global datasets underestimated the local stock by approximately half, highlighting the necessity of local, high-resolution assessments.
• Limitations: The methodology inherently underestimates total biomass because it only detects trees above the canopy (top-down view), missing smaller individuals and sub-canopy trees. The DBH-based equations (AGB4, AGB5) yielded lower estimates (33.5–44.7 MgC.ha⁻¹), likely due to the underestimation of smaller trees in the field plot comparison and the uncertainties of the height-DBH regression.

Significance and Claims
The authors claim that this study successfully demonstrates a cost-effective, logistically feasible, and reproducible methodology for estimating blue carbon stocks in tropical mangroves. The significance lies in:

1. Scalability: The UAV approach can survey 31 hectares in the time it takes to survey a single 225 m² plot, offering a massive advantage for large-scale inventories.
2. Accuracy in Degraded Systems: The ability to generate accurate baselines in highly modified, anthropogenically pressured environments is crucial for restoration planning and carbon market eligibility (e.g., REDD+, NDCs).
3. Methodological Standardization: The workflow, once parameters are defined, is theoretically reproducible for other estuarine systems, potentially serving as a candidate protocol for standardized blue carbon inventories.
4. Complementarity: The study emphasizes that local, high-resolution assessments are essential to complement and correct broader global estimates, particularly for achieving Tier 3 IPCC assessment levels.

The paper concludes that while the method has limitations regarding sub-canopy detection, it provides a robust tool for generating local carbon baselines, which are vital for environmental damage assessment and ecosystem management. Future work is suggested to integrate multi-sensor data and refine species-specific equations, but the current study establishes a viable foundation for UAV-based blue carbon monitoring.