2026-04-24 北海道大学,国立環境研究所
平均森林地上部成長速度（AGR：Average Growth Rate、1ピクセルあたり2m x 2m、単位Mgバイオマスha-1yr-1）。(a)2004-2022、(b)2004-2014、(c)2014-2022、(d) cとbの差分。(d)の赤領域は1期目(2004-2014)より2期目(2014-2022)の方が、成長速度が大きい「遅延加速型回復」を示す。
＜関連情報＞
- https://www.hokudai.ac.jp/news/2026/04/post-2265.html
- https://www.hokudai.ac.jp/news/pdf/260424_pr3.pdf
- https://www.sciencedirect.com/science/article/abs/pii/S0378112726001908
北日本の冷温帯林における複数時期のLidarと高解像度画像及び深層学習を用いた台風かく乱によるバイオマス変化の解明 Characterizing typhoon-disturbed biomass dynamics using multitemporal LiDAR, high-resolution imagery, and deep learning in cool-temperate forest, northern Japan
Ang Li, Tomomichi Kato, Hantao Li, Long Duc Nguyen, Masato Hayashi, Ram Avtar, Tatsuro Nakaji
Forest Ecology and Management Available online: 12 March 2026
DOI:https://doi.org/10.1016/j.foreco.2026.123692
Highlights
- Multitemporal lidar and imagery map landscape-scale post-typhoon biomass recovery.
- Field-plot calibration links canopy height models to aboveground biomass.
- Deep learning maps windthrow disturbance from high-resolution images.
- Deep learning classifies needleleaf and broadleaf at individual tree level.
- Disturbed stands gain biomass faster than undisturbed after an early recovery lag.
Abstract
Temperate forests are key carbon sinks, yet their post-cyclone biomass recovery remains poorly quantified at high spatial and temporal resolution. We combined multi-temporal LiDAR and field measurements to track aboveground biomass (AGB) dynamics following Typhoon Songda (2004) in the Tomakomai Experimental Forest in northern Japan. Harmonized 2 m canopy height models (CHMs) were generated from airborne and UAV LiDAR acquired in 2004, 2014 and 2022, after point-density normalization. Plot inventories were used to calibrate an allometric link between CHM and AGB. Windthrow-disturbed and undisturbed stands were mapped from 2004 color-infrared aerial imagery with a deep learning model (balanced accuracy on the test set = 92.2%), and needleleaf vs. broadleaf crowns were classified from 6 cm UAV RGB imagery (balanced accuracy on the test set = 93.5%). Across 2,516 ha, mean CHM and AGB growth rates over 2004–2022 were 0.13 m yr−1 and 1.30 Mg ha−1 yr−1, respectively. Disturbed stands accumulated biomass faster than undisturbed stands (1.64 vs. 1.29 Mg ha−1 yr−1). Growth in disturbed stands accelerated from 0.97 Mg ha−1 yr−1 (2004–2014) to 2.48 Mg ha−1 yr−1 (2014–2022), revealing an approximate decade-long lag before recovery intensifies. The mean biomass growth rate in disturbed stands exceeded the median during 2014–2022, which may be consistent with gains driven by pioneer species. Our results demonstrate the value of harmonized multi-temporal LiDAR for landscape-scale monitoring of biomass recovery and highlight delayed but substantial post-typhoon carbon gains in cool-temperate forests. The resulting maps and growth rates provide actionable baselines for restoration planning and carbon accounting. They also enable integrating species composition and disturbance history to improve understanding of AGB growth dynamics.
