Machine learning models for subtropical forest aboveground biomass mapping using combined SAR and optical satellite imagery

Abstract

Forest biomass assessment is a critical element influencing the decisions of stakeholders involved in forest management. In tropical and subtropical biodiversity hotspots, accurately measuring aboveground biomass (AGB) is crucial for ecosystem sustainability. However, estimating AGB in these forests is challenging due to the complex vegetation species, necessitating the integration of data from various sources. Therefore, this study aims to investigate the feasibility of integrating ground-based measurements with SAR and optical remote sensing data for estimating AGB in the subtropical forest of Hong Kong and compare various modeling approaches - Stepwise linear regression (SLR), K-nearest neighbors' regression (KNN), and Gradient boosted regression trees (GBRT) - in terms of their effectiveness for AGB mapping. Extensive field data were collected and then converted into biomass values per plot using a locally developed allometric model, designed to facilitate aboveground biomass (AGB) mapping. From the results, we observed that the combination of Sentinel-1 and Sentinel-2 datasets significantly enhanced our model's performance with the GBRT model (R² = 0.84, RMSE = 26.50 tons/ha), outperforming the KNN (R² = 0.67, RMSE = 38.33 tons/ha) and SLR (R² = 0.57, RMSE = 43.88 tons/ha). Furthermore, the GBRT modelling approach demonstrated fewer deviations, with residuals exhibiting less variability in the AGB predictions from the combined dataset, followed by the Sentinel-2 dataset and then the Sentinel-1 dataset. Seasonal analysis revealed a strong correlation between AGB and NDVI, with band ratios involving Sentinel-2 vegetation red-edge bands (SR74, SR85) serving as influential predictors for biomass estimation. In contrast, Sentinel-1 radar backscatter predictors demonstrated a weaker impact on biomass estimation. This research highlights the potential of machine learning approaches in conjunction with satellite remote sensing for accurate AGB mapping in subtropical forests, providing valuable insights for forest management and conservation. The findings not only contribute to the growing field of remote sensing applications but also align with Sustainable Development Goals (SDG) 13 by addressing climate change and SDG 11 by promoting urban sustainability and mitigating environmental risks.