合成開口レーダー・マルチソースデータ融合・機械学習による地上バイオマスおよび炭素蓄積量推定：体系的レビューと計量書誌分析

Unofficial AI-generated summary based on the public title and abstract. Not an official translation.

🔬研究者:SAR・光学・LiDAR融合とアンサンブル/深層学習の比較、検証手法、都市域の研究ギャップを俯瞰できる。

🏢実務担当者:森林・土地利用由来の炭素クレジットや自然資本開示のMRV設計で、どのセンサー融合とモデルが実績あるかを判断する材料になる。

🏛政策担当者:国家GHGインベントリや森林炭素モニタリング制度におけるリモートセンシング・AI活用の標準化と検証要件を検討する際の参考になる。

Abstract Accurate estimation of aboveground biomass (AGB) and carbon stocks is fundamental for sustainable forest management, climate change mitigation, and ecosystem monitoring. Advances in synthetic aperture radar (SAR), multi-source remote sensing, and artificial intelligence have substantially improved biomass estimation; however, existing research remains methodologically fragmented, with limited synthesis of sensor fusion strategies, machine learning, and deep learning approaches, and their applicability to urban tree carbon estimation. This study presents a systematic review and bibliometric analysis of peer-reviewed literature published between 2004 and 2026 following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA 2020) guidelines. Literature was retrieved from Scopus, Web of Science, and Google Scholar, screened using Rayyan, and 68 eligible studies were identified for qualitative synthesis. The review examined publication trends, geographical distribution, remote sensing platforms, data fusion strategies, machine learning and deep learning algorithms, validation approaches, and emerging research gaps. The findings reveal a substantial increase in publications after 2018, driven by the widespread availability of Sentinel-1, Sentinel-2, LiDAR, cloud-based geospatial platforms, and advances in artificial intelligence. The reviewed literature demonstrates a clear transition from single-sensor approaches towards feature-level fusion of SAR, optical, and LiDAR data, which consistently improved biomass and carbon estimation compared with individual sensors. Among predictive models, ensemble learning algorithms, particularly Random Forest and Extreme Gradient Boosting, remained the most robust and widely adopted, while deep learning approaches showed increasing potential for high-resolution applications where sufficient training data were available. Despite these advances, research focusing specifically on urban tree biomass and carbon estimation remains limited, highlighting challenges related to heterogeneous urban environments, limited reference datasets, SAR integration, and model transferability. This review provides a comprehensive synthesis of current methodological developments, identifies critical knowledge gaps, and proposes future research directions to develop robust, scalable, interpretable, and transferable frameworks for biomass and carbon estimation across natural and urban ecosystems.

• openalex https://doi.org/10.21203/rs.3.rs-11364089/v1first seen 2026-11-02 04:40:21 · last seen 2026-11-06 04:36:39