The combination of UAV-LiDAR and LiDAR-SLAM (Simultaneous Localization and Mapping) technology can overcome the scanning limitations of different platforms and obtain comprehensive 3D structural information of forest stands. To address the challenges of the traditional registration algorithms, such as high initial value requirements and susceptibility to local optima, in this paper, we propose a high-precision, robust, NDT-VGICP registration method that integrates voxel features to register UAV-LiDAR and LiDAR-SLAM point clouds at the forest stand scale. First, the point clouds are voxelized, and their normal vectors and normal distribution models are computed, then the initial transformation matrix is quickly estimated based on the point pair distribution characteristics to achieve preliminary alignment. Second, high-dimensional feature weighting is introduced, and the iterative closest point (ICP) algorithm is used to optimize the distance between the matching point pairs, adjusting the transformation matrix to reduce the registration errors iteratively. Finally, the algorithm converges when the iterative conditions are met, yielding an optimal transformation matrix and achieving precise point cloud registration. The results show that the algorithm performs well in Chinese fir forest stands of different age groups (average RMSE—horizontal: 4.27 cm; vertical: 3.86 cm) and achieves high accuracy in single-tree crown vertex detection and tree height estimation (average F-score: 0.90; R2 for tree height estimation: 0.88). This study demonstrates that the NDT-VGICP algorithm can effectively fuse and collaboratively apply multi-platform LiDAR data, providing a methodological reference for accurately quantifying individual tree parameters and efficiently monitoring 3D forest stand structures.
Operational river monitoring increasingly combines video and radar, yet conventional fusion methods seldom accommodate source-specific scaling, correlated errors, outliers, and changing sensor availability. We developed an adaptive dynamic Bayesian fusion model (A-DRBF) combining affine calibration, correlated Student-t errors, dynamic state estimation, and a sensor-inclusion mask. Three video sources (June–December 2023) were evaluated by leave-one-month-out validation; radar was tested in bidirectional July–October comparisons. Across all predictions, A-DRBF yielded the lowest RMSE (0.014 m/s), an MAE of 0.011 m/s, and an 89.1% empirical coverage of a training-calibrated 90% prediction interval. Relative to the three strongest comparators, A-DRBF produced RMSE differences of 0.1–2.5%, with substantially overlapping bootstrap intervals. Radar increased July RMSE from 0.013934 to 0.014315 m/s but reduced October RMSE from 0.015035 to 0.014286 m/s, indicating that its incremental value depends on the hydrodynamic and error regime. Removing affine calibration increased RMSE to 0.165 m/s, while equal-precision weighting increased it to 0.044 m/s. In simulations, A-DRBF reduced RMSE by 45.0% under random outliers and 23.5% under compound disturbance relative to Gaussian dynamics. The model therefore provides accurate estimates and calibrated uncertainty under changing sensor configurations, with its main advantage emerging under contaminated or incomplete observations.