森林地上生物量估算的發展、問題與未來解方之回顧Development of forest aboveground biomass estimation, its problems and future solutions: A review

How have forest aboveground biomass estimation methods developed, which often-overlooked problems significantly affect their accuracy, and what technical pathway can address these problems?

The text confirms that the authors searched the Web of Science Core Collection in May 2023 using the keywords 'forest aboveground biomass' and 'remote sensing', covering the 23-year period from 1999 to 2022, yielding 2191 publications, and summarized annual publication volume, distribution by first authors' country, and journal distribution (Fig. 1).

The text confirms this is a targeted literature review. It first organizes AGB estimation into six method categories (individual tree, sample plot, optical remote sensing, SAR, LiDAR, and multi-source fusion) and uses Fig. 2 to show their evolution and pros and cons, while Tables 1 to 4 summarize representative studies and accuracy for optical, SAR, LiDAR, and multi-source approaches. The discussion then raises two often-overlooked problems and proposes a three-step technical solution: build a three-dimensional tree simulation system that generates simulated trees with known true heights, use the simulated tree height as ground truth to compare manual versus UAV LiDAR survey accuracy, and then use UAV LiDAR instead of manual data to study biomass estimation uncertainty under different plot-pixel overlap levels from 50% to 100%.

The text confirms that AGB-related research accelerated markedly between 2013 and 2022, with the United States and China ranking first and second at 805 and 542 papers (Fig. 1b). The authors summarize the strengths and weaknesses of six method categories and note that the current mainstream is to combine the advantages of multiple sensors through multi-source fusion. The central argument is identifying two often-overlooked yet critical problems: first, the unreliability of manually measured tree height, with the text noting that even the same person measuring at different positions and angles can differ by about 1 m on flat ground; second, that varying degrees of overlap between ground plots and remote sensing pixels introduce spatial-mismatch noise that reduces model accuracy. The authors propose a three-step solution combining three-dimensional tree simulation with UAV LiDAR and list six future research directions.

The text confirms the authors state that their three-step approach is a potentially effective solution rather than a validated system, requiring broad collaboration among researchers to verify the feasibility of replacing manual ground data with UAV LiDAR. Moreover, this is a targeted rather than systematic review, with literature retrieval based on two keywords in a single primary database, and Tables 1 to 4 each list only a few representative studies whose accuracy figures come from different species, regions, and sensors and should not be directly compared or extrapolated to Taiwan.