In recent years, a rise in interest in using Unmanned Aerial Vehicles (UAV) with LiDAR (Light Detection and Ranging) to capture the 3D structure of forests for forestry and ecosystem monitoring applications has been witnessed. Since the terrain is an essential basis for the vertical structure modeling of a forest, the point cloud filtering delivering a highly accurate Digital Terrain Model (DTM) contributes significantly to forest studies. Conventional point cloud filtering algorithms require users to select suitable parameters according to the knowledge of the algorithm and the characteristics of scanned scenes, which are normally empirical and time-consuming. Deep learning offers a novel method in classifying and segmenting LiDAR point cloud, while there are only few studies reported on utilizing deep learning to filter non-ground LiDAR points of forested environments. In this study, we proposed an end-to-end and highly-efficient network named Terrain-net which combines the 3D point convolution operator and self-attention mechanism to capture local and global features for UAV point cloud ground filtering. The network was trained with over 15 million labeled points of 70 forest sites and was evaluated at 17 sites covering various forested environments. Terrain-net was compared with four classical filtering algorithms and one of the most well-recognized point convolution-based deep learning methods (KP-FCNN). Results indicated that Terrain-net achieved the best performance in respect of the Kappa coefficient (0.93), MIoU (0.933) and overall accuracy (98.0%). Terrain-net also performed well in transferring to an additional third-party open dataset for ground filtering in large-scale scenes and other vegetated environments. No parameters need to be tuned in transferring predictions. Terrain-net will hopefully be widely applied as a new highly-efficient, parameter-free, and easy-to-use tool for LiDAR data ground filtering in varying forest environments.

Efficient and accurate acquisition of tree distribution and three-dimensional geometric information in forest scenes, along with three-dimensional reconstructions of entire forest environments, hold significant application value in precision forestry and forestry digital twins. However, due to complex vegetation structures, fine geometric details, and severe occlusions in forest environments, existing methods—whether vision-based or LiDAR-based—still face challenges such as high data acquisition costs, feature extraction difficulties, and limited reconstruction accuracy. This study focuses on reconstructing tree distribution and extracting key individual tree parameters, and it proposes a forest 3D reconstruction framework based on high-resolution remote sensing images. Firstly, an optimized Mask R-CNN model was employed to segment individual tree crowns and extract distribution information. Then, a Tree Parameter and Reconstruction Network (TPRN) was constructed to directly estimate key structural parameters (height, DBH etc.) from crown images and generate tree 3D models. Subsequently, the 3D forest scene could be reconstructed by combining the distribution information and tree 3D models. In addition, to address the data scarcity, a hybrid training strategy integrating virtual and real data was proposed for crown segmentation and individual tree parameter estimation. Experimental results demonstrated that the proposed method could reconstruct an entire forest scene within seconds while accurately preserving tree distribution and individual tree attributes. In two real-world plots, the tree counting accuracy exceeded 90%, with an average tree localization error under 0.2 m. The TPRN achieved parameter extraction accuracies of 92.7% and 96% for tree height, and 95.4% and 94.1% for DBH. Furthermore, the generated individual tree models achieved average Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index (SSIM) scores of 11.24 and 0.53, respectively, validating the quality of the reconstruction. This approach enables fast and effective large-scale forest scene reconstruction using only a single remote sensing image as input, demonstrating significant potential for applications in both dynamic forest resource monitoring and forestry-oriented digital twin systems.

The ecology and life history of bird species inhabiting limestone forests, which are under major conservation threats, is currently poorly known. To cover this gap of knowledge, in this study we report for the first time on several aspects of the breeding ecology of the Blue-rumped Pitta (Pitta soror) and the Fairy Pitta (P. nympha) inhabiting two typical limestone forests of south China. The mean density of Fairy Pittas in our study locations was 3.13 ± 2.82 and 1.05 ± 2.09 individuals/km². The Blue-rumped Pitta was common in Nonggang and showed a mean density of 4.67 ± 2.44 individuals/km², yet was absent in the other. We found nine nests between February 2009 and June 2015, including five of the Blue-rumped Pitta and four of the Fairy Pitta. Blue-rumped Pittas laid 4.8 ± 0.4 eggs with a mean fresh mass of 8.10 ± 0.40 g. Fairy Pittas had a clutch size of 5.2 ± 0.45 eggs with a mean fresh mass of 6.03 ± 0.22 g. Blue-rumped and Fairy Pitta parents fed their nestlings 4.0 ± 1.2 times and 3.9 ± 1.5 times per hour, respectively. Earthworms were the most common food item delivered to nestlings by Blue-rumped (93.6%) and Fairy Pitta (91.2%) parents. Blue-rumped Pittas bred successfully in 40% of nests (two of five), whilst in the case of Fairy Pittas nest success reached 75% (three of four). The two Pittas had larger clutch sizes than in southern populations of the same species and also than most other bird species inhabiting limestone forests. This would imply that Pittas show differences in life history traits within limestone ecosystems.

Rapid and accurate tree crown detection is significant to forestry management and precision forestry. However, the variability of the data is challenging for traditional deep learning-based methods in cross-regional scenarios. To solve the problem of low accuracy of existing methods due to the distribution characteristics of tree crowns and the special background in which they are located, we propose a multi-adversarial domain adaptive (DA) crown detection model, which saves much of labeling efforts needed for covering different features in different regions. First, we embed a multilevel transferable attention mechanism to extract the multiscale transferable features of the tree crown and combat the negative transfer caused by the background. Second, a multi-adversarial instance alignment module with entropy regularization and a weighting factor is designed to match the cross-regional data distributions to reduce the misdetection and misidentification of complex backgrounds. The above-mentioned adaptations are integrated into a two-stage object detector to realize end-to-end training. We conduct comprehensive experiments with four transfer tasks. Our method improves F1-score to 95.68%, 83.12%, 78.56%, and 66.89% and performs 3.61% and 15.38% better than the state-of-the-art domain adaptation approaches without extra inference costs. The results demonstrate the great potential of our method for DA tree crown detection, which provides a reference for further research on forestry detection.