Disturbances from wind, bark beetles, and wildfires have increased in Europe's forests throughout the 20 century. Climatic changes were identified as a main driver behind this increase, yet how the expected continuation of climate change will affect Europe's forest disturbance regime remains unresolved. Increasing disturbances could strongly impact the forest carbon budget, and are hypothesized to contribute to the recently observed carbon sink saturation in Europe's forests. Here we show that forest disturbance damage in Europe has continued to increase in the first decade of the 21 century. Based on an ensemble of climate change scenarios we find that damage from wind, bark beetles, and forest fires is likely to increase further in coming decades, and estimate the rate of increase to +0.91·10 m of timber per year until 2030. We show that this intensification can offset the effect of management strategies aiming to increase the forest carbon sink, and calculate the disturbance-related reduction of the carbon storage potential in Europe's forests to be 503.4 Tg C in 2021-2030. Our results highlight the considerable carbon cycle feedbacks of changing disturbance regimes, and underline that future forest policy and management will require a stronger focus on disturbance risk and resilience.

为探究不同机器学习模型在我国南方典型湿润山区的植被类型分类效果，基于无人机遥感影像、实地观测数据、数字高程模型建立遥感特征，选用决策树、随机森林、支持向量机和AdaBoost模型在安徽屯溪流域构建植被类型遥感分类模型；选择总体精度、Kappa系数、均方误差、用户精度和生产者精度等评价指标，分析对比4种机器学习模型在典型小流域的适用性。结果表明：对于林地类型，AdaBoost模型分类精度最高，表明AdaBoost模型在林地分类中有明显的优势；对于非林地类型，模型之间精度差异较大，随机森林模型精度最高；整体而言，4种模型在南方典型湿润山区典型小流域均可获得较好的分类效果，其中AdaBoost模型总体精度为95.55%、Kappa系数为0.9419，均为最高，支持向量机模型表现均最低。地形因子、纹理特征等辅助特征为分类过程提供了重要信息，有助于提高分类精度。

山地因其较高的异质性和特殊的环境特征给遥感科学及其应用带来了诸多问题和挑战。为实现山地植被信息的精准提取,本研究选择部分滇西北山地区域作为研究区开展方法实验,利用高分辨率遥感影像数据和数字高程模型,结合分区分层感知思想,提出一种基于不确定性理论的山地植被型组分类制图方法。首先结合地形对研究区影像进行多尺度分割制作图斑;然后根据图斑特征使用随机森林方法进行分类,将分类结果与对应类别样本间的相似性作为优化目标, 并构建混合熵模型定量计算图斑推测类型的不确定性,据此进行针对性的样本补充和分类模型的迭代优化。实验总体分类精度达90.8%,较迭代前提升了29.4%,Kappa系数达到0.875。在高不确定性区域,该方法相比使用一次性补样和随机补样方法的分类结果,精度分别提高了17%和13%。研究结果表明,通过人机交互的方式,基于不确定性理论为样本库融入增量信息的迭代优化方法能够有效提高植被型组分类的精度,相较于传统的样本选择方法具有更高的效率和更低的不确定性。

植被分类是森林资源调查与动态监测的基础与前提。当前植被分类研究大都利用光学遥感影像,然而,光学遥感成像易受到云雨覆盖的影响,难以构建完整时间序列,植被分类精度有限。微波遥感具有全天时全天候、时间序列完整的优势,在植被调查与分析中具有巨大的应用潜力。本文利用2018年Sentinel-1A微波遥感时间序列数据和深度循环网络方法,对秦岭太白山区的森林植被进行分类制图。首先利用Sentinel-2光学影像与数字高程数据对研究区进行多尺度分割;然后将处理后的时间序列Sentinel-1A数据空间叠加到分割地块上,构建地块的多元时间序列曲线;最后利用深度循环网络提取与学习多元时间序列的时序特征并分类。实验结果表明：① 与传统机器学习方法（如RF、SVM）相比,本文提出的深度循环网络方法的分类精度提高10%以上;② 在Sentinel-1A微波极化特征组合中VV+VH表现最好,与VV+VH+VV/VH极化特征组合的精度相近;③ 使用全年的时间影像构建时间序列分类精度最高,达到82%。研究表明,利用深度循环网络与时间序列Sentinel-1A数据的方法能够有效提高植被分类的精度,从数据源与分类方法上为森林植被分类研究提供了新的思路。

实景三维是对人类生产、生活和生态空间进行真实、立体化反映和表达的数字虚拟空间，已作为基础数据广泛应用于智慧城市、可视化展览等领域。实景三维模型语义解译是充分发掘实景三维模型的数据价值以实现场景的自动理解与分析的基础。已有的三维解译方法中，存在被遮挡地物解译不完全，不同地物边界解译不准确等问题。针对该问题，本研究提出了一种基于多视角的实景三维模型解译方法。该方法首先模拟成像过程实现实景三维的多视角二维投影，进一步基于二维影像解译快速获取语义信息，最后将多个二维解译结果进行逆投影获得三维语义模型。本研究以深圳大学实景三维模型为实验数据，通过与基于正射影像的三维解译、面向对象的三维分层解译等方法对比，验证本文所提方法的有效性。结果显示本研究提出的基于多视角的解译方法获得了最高的分类精度（总体分类精度为96.69%， Kappa系数为0.942），在正射遮挡区域，以及不同地物边界区域有更好的解译效果。本方法充分利用实景三维数据的多角度信息，具有较高的理论和实践参考价值，为进一步促进实景三维建设及在自然资源监测领域的应用提供了方法支撑。

在摄影测量与遥感领域,三维网格数据是最终用户产品之一,已广泛应用于城市规划、导航等任务中。但针对以三维网格表示的复杂城市场景的智能化语义分割的研究较少。为此,本文提出复杂城市动态图卷积网络三维场景语义分割方法(3Dcity-net)。利用三维网格固有的三维空间坐标信息和纹理信息构建的复合特征向量来表示三维网格中的三角面片。为降低纹理信息中噪声和冗余信息对语义分割结果的影响,提出在3Dcity-net网络结构中嵌入主成分分析模块。为缓解样本数据不平衡引起的语义分割精度下降的问题,采用焦点损失函数替代交叉熵损失函数。利用Hessigheim三维网格数据进行了语义分割试验。试验结果表明,3Dcity-net能够获得具有竞争力的三维网格语义分割结果,其中总体精度OA、Kappa系数、平均准确率mP、平均召回率mR、平均F1值(F1 score)和平均交并比mIoU分别为81.5%、0.776、73.0%、58.4%、62.6%和49.8%。与先进方法相比,本文方法总体精度OA分别提高了0.9%和8.3%。

In photogrammetry and remote sensing community, 3D mesh is one of the final user products, which is widely applied in urban planning, navigation, etc. However, there are few works on semantic complex 3D mesh urban scene segmentation based on deep learning methods. Thus, a semantic segmentation method of 3D scenes using dynamic graph CNN for complex city (3Dcity-net) is proposed. By using mesh-inherent features containing 3D spatial information and texture information, a composite feature vector is proposed to represent each face in 3D mesh. To reduce the influence on semantic segmentation by the noise and redundant information in texture information, a principal component analysis (PCA) module is embedded in to the proposed 3D city-net. In order to alleviate the problem of semantic segmentation precision decrease caused by the unbalanced sample data, the focal loss function is used to replace the cross-entropy loss function. The Hessigheim 3D mesh data are utilized to perform experiments. The results of experiments show that the proposed method can obtain competitive semantic segmentation results on 3D mesh. The overall accuracy, Kappa coefficient, mean precision, mean recall, mean F1 score, and mean IoU is 81.5%, 0.776, 73.0%, 58.4%, 62.6%, and 49.8%, respectively. Comparing to two state-of-the-art methods, the overall accuracy increases by 0.9%, and 8.3%, respectively.

3D plant structure observation and characterization to get a comprehensive knowledge about the plant status still poses a challenge in Precision Agriculture (PA). The complex branching and self-hidden geometry in the plant canopy are some of the existing problems for the 3D reconstruction of vegetation. In this paper, we propose a novel application for the fusion of multispectral images and high-resolution point clouds of an olive orchard. Our methodology is based on a multi-temporal approach to study the evolution of olive trees. This process is fully automated and no human intervention is required to characterize the point cloud with the reflectance captured by multiple multispectral images. The main objective of this work is twofold: (1) the multispectral image mapping on a high-resolution point cloud and (2) the multi-temporal analysis of morphological and spectral traits in two flight campaigns. Initially, the study area is modeled by taking multiple overlapping RGB images with a high-resolution camera from an unmanned aerial vehicle (UAV). In addition, a UAV-based multispectral sensor is used to capture the reflectance for some narrow-bands (green, near-infrared, red, and red-edge). Then, the RGB point cloud with a high detailed geometry of olive trees is enriched by mapping the reflectance maps, which are generated for every multispectral image. Therefore, each 3D point is related to its corresponding pixel of the multispectral image, in which it is visible. As a result, the 3D models of olive trees are characterized by the observed reflectance in the plant canopy. These reflectance values are also combined to calculate several vegetation indices (NDVI, RVI, GRVI, and NDRE). According to the spectral and spatial relationships in the olive plantation, segmentation of individual olive trees is performed. On the one hand, plant morphology is studied by a voxel-based decomposition of its 3D structure to estimate the height and volume. On the other hand, the plant health is studied by the detection of meaningful spectral traits of olive trees. Moreover, the proposed methodology also allows the processing of multi-temporal data to study the variability of the studied features. Consequently, some relevant changes are detected and the development of each olive tree is analyzed by a visual-based and statistical approach. The interactive visualization and analysis of the enriched 3D plant structure with different spectral layers is an innovative method to inspect the plant health and ensure adequate plantation sustainability.

相比于传统的人工实地调查的方法，通过无人机倾斜摄影测量技术多角度拍摄提取单木树冠信息，具有高效、准确和低成本等优势。以内蒙古自治区赤峰市喀喇沁旗西南部旺业甸林场的落叶松近成熟林为研究对象，以倾斜摄影测量技术获取的无人机影像为数据源，提出一种结合可见光植被指数和分水岭算法的单木冠幅提取方法。首先通过数字正射影像计算可见光波段超绿超红差分指数；选择合理阈值对图像进行二值化，并利用中值滤波对树冠区域图进行去噪处理，分离出植被与非植被区域；以植被区域为掩膜范围对冠层高度模型进行掩膜；最后，利用分水岭分割算法提取单木树冠进行精度验证。在树冠区域提取过程中，基于超绿超红差分指数结合阈值法成功分离出植被与非植被区域；通过中值滤波有效地去除了因亮度不均、阴影及非植被区域的纹理等信息所造成的斑点和噪声，保证了树冠边缘的完整性及树冠内部的连通性，同时减少了分水岭算法的过分割现象。单木尺度上，树冠参数信息提取的准确率分别为88.72%和79.38%，召回率分别为93.29%和88.60%，F测度分别为90.59%和83.74%；样地尺度上，相对误差分别为15.45%和22.92%。研究结果表明：基于数字正射影像采用可见光植被指数提取植被区域很好地消除了林内裸地等背景因素的影响，同时基于冠层高度模型利用分水岭分割算法能够精确的区分单木信息。基于无人机倾斜摄影测量技术得到的两种数据源结合发挥了各自的优势，能够高效、准确地提取较高郁闭度林分的单木树冠信息。

Separating point clouds into ground and non-ground measurements is an essential step to generate digital terrain models (DTMs) from airborne LiDAR (light detection and ranging) data. However, most filtering algorithms need to carefully set up a number of complicated parameters to achieve high accuracy. In this paper, we present a new filtering method which only needs a few easy-to-set integer and Boolean parameters. Within the proposed approach, a LiDAR point cloud is inverted, and then a rigid cloth is used to cover the inverted surface. By analyzing the interactions between the cloth nodes and the corresponding LiDAR points, the locations of the cloth nodes can be determined to generate an approximation of the ground surface. Finally, the ground points can be extracted from the LiDAR point cloud by comparing the original LiDAR points and the generated surface. Benchmark datasets provided by ISPRS (International Society for Photogrammetry and Remote Sensing) working Group III/3 are used to validate the proposed filtering method, and the experimental results yield an average total error of 4.58%, which is comparable with most of the state-of-the-art filtering algorithms. The proposed easy-to-use filtering method may help the users without much experience to use LiDAR data and related technology in their own applications more easily.

为探索机载点云与无人机可见光影像在乔木树种识别与分类领域的应用潜力，提出了一种多模态特征与决策混合融合的无人机单木尺度树种分类识别方法。首先使用Kendall Rank相关系数法与排列重要性分析（Permutation Importance， PI）进行特征选择，采用高效低秩多模态融合算法（Low-rank Multimodal Fusion， LMF）融合点云与影像特征。再引入集成学习，将点云、影像及融合特征分别输入Stacking集成的极限梯度提升机（eXtreme Gradient Boosting， XGBoost）、轻型梯度提升机（Light Gradient Boosting Machine， LightGBM）与随机森林（Random Forest， RF）3个基分类器，最后采用元分类器—朴素贝叶斯进行决策融合。实验数据表明：所提方法独立测试精度达99.4%，较传统的特征串联融合随机森林分类器提升了22.58%，Kappa系数提升了0.285 4。与卷积神经网络（Convolutional Neural Network， CNN）对比实验证明：所提算法在小样本训练的优势明显，且具有更好的泛化能力。