Individual tree segmentation and diameter at breast height (DBH) estimation are fundamental to precision forest inventory. Light Detection and Ranging (LiDAR) technology has become an essential tool for achieving these objectives at the single-tree level. Different LiDAR platforms—notably unmanned aerial vehicle (UAV) and Mobile Laser Scanning (MLS)—each offer distinct advantages in capturing forest structural information. Multi-source LiDAR fusion has been proposed as a strategy to combine these complementary strengths. However, how to effectively select segmentation algorithms across different LiDAR data sources remains insufficiently understood, particularly for subtropical coniferous plantations with heterogeneous canopy structure. This study systematically compared four individual tree segmentation algorithms (Donager2021, Dalponte2016, Silva2016, and marker-controlled watershed segmentation [MCWS]) across three LiDAR data sources (UAV-only, MLS-only, and fused UAV–MLS) in Pinus massoniana plantations in subtropical China. DBH estimation models were then developed based on the best-performing segmentation results to examine whether data fusion simultaneously improves both detection and DBH retrieval accuracy. The main findings are as follows: (1) the three canopy height model (CHM)-based algorithms achieved a mean overall accuracy (OA) for individual-tree detection of approximately 64% on UAV data but fell below 30% on MLS data, failing to support effective detection; (2) The Donager2021 algorithm, which directly exploits trunk structure from three-dimensional point clouds, achieved the highest OA of 93.15% with MLS data and further improved to 94.82% with fused data; (3) DBH estimation reached a mean R² of 0.96 for both MLS and fused datasets, yet MLS LiDAR alone produced a lower RMSE (2.31 cm; rRMSE = 5.91%) than fused LiDAR (RMSE = 2.40 cm; rRMSE = 6.24%); and (4) higher detection accuracy did not necessarily lead to better DBH estimation, revealing a trade-off between the two objectives. These findings indicate that fusion does not universally improve all downstream tasks, and that the choice of LiDAR configuration and segmentation algorithm should be guided by specific inventory objectives.

Accurate quantification of campus vegetation carbon stocks is essential for advancing carbon neutrality goals and refining urban carbon management strategies. This study pioneers the integration of drone and backpack LiDAR data to overcome limitations in conventional carbon estimation approaches. The Comparative Shortest-Path (CSP) algorithm was originally developed to segment tree crowns from point cloud data, with its design informed by metabolic ecology theory—specifically, that vascular plants tend to minimize the transport distance to their roots. In this study, we deployed the Comparative Shortest-Path (CSP) algorithm for individual tree recognition across 897 campus trees, achieving 88.52% recall, 72.45% precision, and 79.68% F-score—with 100% accuracy for eight dominant species. Diameter at breast height (DBH) was extracted via least-squares circle fitting, attaining >95% accuracy for key species such as Magnolia grandiflora and Triadica sebifera. Carbon storage was calculated through species-specific allometric models integrated with field inventory data, revealing a total stock of 163,601 kg (mean 182.4 kg/tree). Four dominant species—Cinnamomum camphora, Liriodendron chinense, Salix babylonica, and Metasequoia glyptostroboides—collectively contributed 84.3% of total storage. As the first integrated application of multi-platform LiDAR for campus-scale carbon mapping, this work establishes a replicable framework for precision urban carbon sink assessment, supporting data-driven campus greening strategies and climate action planning.

Phyllostachys edulis (Moso bamboo) is a subtropical bamboo species endemic to China and holds substantial economic and ecological value. Ongoing climate change is reshaping the geographic distribution of its suitable habitats. To reveal these climate-driven dynamics, we calibrated the maximum entropy (MaxEnt) species distribution model using the ENMeval R package version 4.5.2 and evaluated model robustness via spatially independent block cross-validation. We identified nine critical environmental predictors to model spatiotemporal variations in suitable habitats for Moso bamboo, examining both present-day climate and four future periods under four Shared Socioeconomic Pathway (SSP) scenarios. Model performance was optimized at regularization multiplier (RM) of 1.5 with the feature class combination LQHPT, yielding a spatial validation AUC of 0.9104 and robust predictive capacity. Environmental controls were dominated by mean temperature of the coldest quarter (bio11, 63.9% relative contribution) and precipitation of the driest quarter (bio17, 31.2%), while soil clay and organic carbon content modulated suitability at local scales. Under current conditions, the total suitable habitat area encompasses approximately 218.88 × 10⁴ km², with the majority distributed throughout the subtropical zone of southeastern China. Under nearly all future scenarios, high-suitability habitats showed a general contracting trend of 20.94%–95.29% relative to the current baseline, while moderate-suitability habitats exhibited a fluctuating expansion trend, reaching a peak increase of 47.90% in the 2030s under SSP585, but contracted by 18.65% in the 2090s. Meanwhile, habitat centroids underwent small-scale oscillatory shifts in multiple directions and remained entirely within Hunan Province across all periods. MESS analysis confirmed that 90.60%–99.82% of China’s land area fell within climate-analogous ranges, indicating low extrapolation risk for core habitat projections. Core climate refugia with high cross-scenario stability were primarily concentrated in the traditional bamboo-producing regions of southeastern China. This study provides a scientific basis for the sustainable management of Moso bamboo resources, biodiversity conservation, and ecological risk prevention in the context of ongoing climate change.