Accurate estimation of mangrove aboveground biomass density (AGBD) is crucial for assessing ecosystem productivity and blue carbon sequestration. However, frequent cloud cover, complex canopy structures, and scarce ground samples pose critical challenges for coastal AGBD mapping. We developed an annual Mangrove AGBD estimation Model (MAGB-STM) by considering three methodological aspects: (i) integrating spectral-temporal metrics (STM) derived from Sentinel-1 radar polarization and Sentinel-2 spectral and vegetation index features to capture mangrove dynamics; (ii) leveraging a random observation selection (ROS) strategy to simulate varying clear sky observation (CSO) densities across the 2017–2024 time series for transferability assessment; and (iii) evaluating synergistic contributions by comparing the proposed single-year approach against single-source models (Sentinel-1 or Sentinel-2 alone) and multi-year models trained using combined 2017 to 2024 data. We validated the model in the Zhanjiang Mangrove National Nature Reserve (ZMNNR) using 2024 UAV-LiDAR and field samples, and further evaluated it in three other mangrove regions using multi-year UAV-LiDAR data, with reference AGBD derived from point cloud structural features. Results showed that: (1) The MAGB-STM model, integrating optical and SAR data through STM, achieved superior performance (R2 = 0.76, RMSE = 14.81 Mg·ha−1, MAE = 12.39 Mg·ha−1). This improvement, corresponding to an R2 increase of 0.24-0.39 over single-source models, demonstrates the complementary characteristics of the SAR backscatter signal and optical STM in mangrove structural components and biomass growth. (2) The temporal density of per-pixel CSO and their STM statistics were identified as key variables for a single-year MAGB-STM model; the ROS strategy achieved an approximately 8% relative increase in mean R2 against the 0.76 baseline across years by mitigating the impact of CSO density fluctuations. (3) The ROS strategy delineated optimal operational thresholds, with an 80% CSO density (31 ± 2 CSOs per pixel) mitigating overfitting in high CSO density years, while a 60% density prevented underfitting in low-density years, ensuring robust inter-annual estimation in cloudy regions. Validation against in situ and UAV-LiDAR data revealed that the AGBD estimates from 2017 to 2024 achieved a mean accuracy of R2 = 0.81, RMSE = 12.03 Mg·ha−1, and MAE = 11.31 Mg·ha−1, with a mean annual AGBD of approximately 71.31 Mg·ha−1 in ZMNNR, Guangdong Province, South China. This study provides a robust solution for mangrove biomass estimation in cloudy coastal areas and suggests the transferability of the proposed synergistic approach to other tropical forests with persistent cloud cover.
Xue et al. (Tue,) studied this question.