Assessing and quantifying the saturation effect of forest aboveground biomass mapping using Landsat 8, Sentinel 1/2

The application of remote sensing to forest aboveground biomass (AGB) mapping has received increasing attention, and substantial progress has been achieved. Nevertheless, saturation of remotely sensed signals remains a major challenge, often leading to the underestimation of AGB in high-density forests. In structurally complex forests, saturation in forest AGB mapping is influenced by multiple factors, including data sources, tree species, and modeling approaches. Although numerous methods have been proposed to mitigate saturation, a standardized metric for quantifying saturation and evaluating the effectiveness of these methods is still lacking. In this study, by separating saturation effects from mapping errors, SAT is proposed as a quantitative metric for assessing saturation in forest AGB mapping. The saturation levels of widely used and freely available Landsat 8 and Sentinel-1/2 remote sensing data were evaluated for forests dominated by ten common tree species in southern China. In addition, a quantitative analysis was conducted to examine the effects of multi-source data integration and nine different machine learning models on saturation levels. The results showed that tree species had a pronounced influence on estimation performance and saturation levels, with SAT values ranging from 48.15 Mg/ha to 147.93 Mg/ha. Integrated decision tree-based ensemble models produced more stable estimation results and higher saturation levels than single models. Moreover, texture features derived from optical images and Sentinel-1 data improved both the accuracy and saturation level of forest AGB estimation, with Sentinel-1 features showing a stronger contribution than texture features alone. Overall, the proposed SAT goes beyond conventional accuracy assessment by explicitly quantifying saturation-induced underestimation in high-biomass forests. This provides practical support for comparing sensor combinations and modeling strategies, and improves the reliability and interpretability of remote-sensing-based forest AGB products in dense forests.