Species information in multi-temporal Sentinel-2 data improves forest canopy height estimation

作者：Changhyun Choi, Youngryel Ryu, Benjamin Dechant, Tackang Yang, Liang Wan, Sungchan Jeong, Jin Wu · 发表于：Agricultural and Forest Meteorology · 年份：2026 · DOI：10.1016/j.agrformet.2026.111114 · 被引用次数：1 · 研究领域：Remote Sensing and LiDAR Applications、Remote Sensing in Agriculture、Plant Water Relations and Carbon Dynamics

• Species information largely explains CHM improvements from multi-temporal data. • Multi-temporal data improves CHM across broadleaf, needleleaf, and mixed forests. • Value of species maps for CHM peaks in species-mixed pixels, drops in pure stands. • CHM accuracy depends on number and timing of Sentinel-2 images. The integration of Sentinel-2 (S2) multispectral data and LiDAR measurements using convolutional neural networks (CNNs) has been extensively explored for large-scale canopy height mapping. Recent studies reported multi-temporal S2 imagery improves canopy height mapping, but the ecological mechanism underlying this improvement remains poorly understood. Here, we hypothesize that this improvement in canopy height estimation is primarily because multi-temporal data contain information relevant to tree species composition. Each tree species likely exhibits a unique reflectance–height relationship, suggesting that incorporating species information could improve canopy height estimates. We further hypothesize that simple plant functional type maps can improve canopy height estimation but less improvement than detailed species-level information. To test these hypotheses, we trained CNN models using airborne LiDAR canopy height at nine National Ecological Observatory Network sites including broadleaf, needleleaf, and mixed forests, using four input combinations: [1] single acquisition only, [2] multi-temporal data only, [3] single acquisition with species information, and ...