基于神经网络的印度洋长鳍金枪鱼(Thunnus alalunga)时空分布与海洋环境关系研究

长鳍金枪鱼(Thunnusalalunga)是主要的经济性金枪鱼鱼种之一,其空间分布与环境因子存在着密切联系。利用2012—2019年印度洋长鳍金枪鱼生产数据和海洋环境数据,包括海表面温度(sea surface temperature, SST)、叶绿素浓度(chlorophyll a, chl a)和海表面盐度(sea surface salinity, SSS)构建印度洋长鳍金枪鱼时空分布神经网络模型。以空间(经度,纬度)、环境因子(SST, chl a, SSS)为解释变量,局部渔获量为因变量,变化隐含层节点数,构建了18个BP空间分布模型,并采用10×10交叉验证模型稳定性,以均方误差(meansquareerror,MSE)、平均相对方差(averagerelativevariance,ARV)以及拟合优度(R~2)作为不同模型精度与稳定性的评判标准,最终选取5-18-1(隐含层节点18)模型为最佳模型,其平均MSE值为0.02232,平均ARV值为0.511。利用最优模型预测结果与同期实际捕捞产量进行叠加对比发现两者具有一致性。环境因子敏感性分析表明海表温度显著影响印度洋长鳍金枪鱼渔场分布,其贡献率达到0.2。印度洋长鳍金枪鱼高精度BP神经网络时空分布模型为其资源的可持续开发与动态管理提供了一种新思路。     

Genealogical concordance,comparative species delimitation,and the specific status of the Caspian pipefish Syngnathus caspius (Teleostei: Syngnathidae)

To evaluate the specific validity of the Caspian pipefish Syngnathus caspius, we used a comparative molecular species delimitation method on a COI barcode library of Syngnathus, as well as principles of genealogical concordance. Comparative species delimitation allowed us to delineate putative species without a priori assignment of individuals to nominal species, while genealogical concordance extended our species delimitation results to multiple genes, multiple codistributed species, and comparisons with biogeographic evidence. All species delimitation analyses including two topology‐based, one network‐based, and one distance‐based analysis showed genetically isolated lineages of pipefish in the Black and Caspian Sea, corresponding to S. abaster and S. caspius, respectively. Mean evolutionary divergence between the two lineages (0.029) was within the range separating species of Syngnathus (0.024–0.217). The interclade/intraclade ratio of variation was comparable to the operational criterion of divergence between clades greater or equal to 10 × the level within clades to recognize separate species. Our argument on taxonomic validity of S. caspius is also supported by the principles of genealogical concordance as a conceptual basis for recognition of biological species. As a second objective, using a limited number of S. caspius specimens from two semi‐confined water bodies along the Caspian Sea south coastal zone (i.e., Anzali Wetland in the west and Gorgan Bay in the east), we searched for a possible matrilineal structure. The retrieved phylogeographic pattern was characterized by a shallow genealogy and lineage distributions varied, most probably caused by low to modest contemporary gene flow between populations of S. caspius across the southern Caspian Sea that are linked tightly through history.     

Exploring the stemflow dynamics and driving factors at both inter- and intra-event scales in a typical subtropical deciduous forest

Numerous efforts have been made to understand stemflow dynamics under different types of vegetation at the inter-event scale, but few studies have explored the stemflow characteristics and corresponding influencing factors at the intra-event scale. An in-depth investigation of the inter- and intra-event dynamics of stemflow is important for understanding the ecohydrological processes in forest ecosystems. In this study, stemflow volume (F_V), stemflow funnelling ratio (F_R), and stemflow ratio (F_%) from Quercus acutissima and Broussonetia papyrifera trees were measured at both inter- and intra-event scales in a subtropical deciduous forest, and the driving factors, including tree species and meteorological factors were further explored. Specifically, the F_V, F_R and F_% of Q. acutissima (52.3 L, 47.2, 9.6%) were lower than those of B. papyrifera (85.1 L, 91.2, 12.4%). The effect of tree species on F_V and F_% was more obvious under low intensity rainfall types. At the inter-event scale, F_V had a strong positive linear correlation with rainfall amount (G_P) and event duration (D_E) for both tree species, whereas F_R and F_% had a positive logarithmic correlation with G_P and D_E only under high-intensity, short-duration rainfall type. F_R and F_% were mainly affected by wind speed and the maximum 30-min rainfall intensity under low-intensity, long-duration rainfall type. At the intra-event scale, for both tree species, the mean lag time between the start of rainfall and stemflow was the shortest under high-intensity, short-duration rainfall type, while the mean duration and amount of stemflow after rain cessation were the greatest under high-amount, long-duration rainfall type. The relationship between stemflow intensity and rainfall intensity at the 5-min interval scale also depended greatly on rainfall type. These findings can help clarify stemflow dynamics and driving factors at both inter- and intra-event scales, and also provide abundant data and parameters for ecohydrological simulations in subtropical forests.     

MARC: a robust method for multiple-aspect trajectory classification via space,time, and semantic embeddings

ABSTRACT

The increasing popularity of Location-Based Social Networks (LBSNs) and the semantic enrichment of mobility data in several contexts in the last years has led to the generation of large volumes of trajectory data. In contrast to GPS-based trajectories, LBSN and context-aware trajectories are more complex data, having several semantic textual dimensions besides space and time, which may reveal interesting mobility patterns. For instance, people may visit different places or perform different activities depending on the weather conditions. These new semantically rich data, known as multiple-aspect trajectories, pose new challenges in trajectory classification, which is the problem that we address in this paper. Existing methods for trajectory classification cannot deal with the complexity of heterogeneous data dimensions or the sequential aspect that characterizes movement. In this paper we propose MARC, an approach based on attribute embedding and Recurrent Neural Networks (RNNs) for classifying multiple-aspect trajectories, that tackles all trajectory properties: space, time, semantics, and sequence. We highlight that MARC exhibits good performance especially when trajectories are described by several textual/categorical attributes. Experiments performed over four publicly available datasets considering the Trajectory-User Linking (TUL) problem show that MARC outperformed all competitors, with respect to accuracy, precision, recall, and F1-score.     

传统聚落景观基因的地理信息特征及其理解

传统聚落是民族传统文化遗产的重要组成部分,对社会经济和文化建设具有重要的价值。针对现有研究缺乏探讨传统聚落景观基因（简称景观基因）蕴藏的地理信息特征及理解方法,论文从以下方面开展了深入分析：① 地理信息包含了语义描述、几何形态、属性特征、维度、时空框架、尺度、要素相互关系（空间关联）、演化过程（存在状态）等属性;② 景观基因是一种特殊的文化因子,蕴含着丰富的哲理,是认识传统聚落特征的分析方法,也是文化符号的集合;③ 景观基因包含了空间定位、载体特征、历史与文化特征等丰富的地理信息。根据前述分析结果,论文从符号化、数据挖掘、谱系分析与空间格局制图等途径系统地探讨了景观基因地理信息特征的理解方法。论文认为：在地理时空大数据与地理服务日益深化发展的背景下,结合GIS原理探讨景观基因的地理信息特征对于延伸地理信息科学的内涵,深化传统聚落的地学认识,促进人文GIS的发展具有积极的意义,今后应该继续加强相关方法探索。     

Inference of in situ stress from thermoporoelastic borehole breakouts based on artificial neural network

The determination of in situ stresses is very important in petroleum engineering. Hydraulic fracturing is a widely accepted technique for the determination of in situ stresses nowadays. Unfortunately, the hydraulic fracturing test is time-consuming and expensive. Taking advantage of the shape of borehole breakouts measured from widely available caliper and image logs to determine in situ stress in petroleum engineering is highly attractive. By finite element modeling of borehole breakouts considering thermoporoelasticity, the authors simulate the process of borehole breakouts in terms of initiation, development, and stabilization under Mogi-Coulomb criterion and end up with the shape of borehole breakouts. Artificial neural network provides such a tool to establish the relationship between in situ stress and shape of borehole breakouts, which can be used to determine in situ stress based on different shape of borehole breakouts by inverse analysis. In this paper, two steps are taken to determine in situ stress by inverse analysis. First, sets of finite element modeling provide sets of data on in situ stress and borehole breakout measures considering the influence of drilling fluid temperature and pore pressure, which will be used to train an artificial neural network that can eventually represent the relationship between the in situ stress and borehole breakout measures. Second, for a given measure of borehole breakouts in a certain drilling fluid temperature, the trained artificial neural network will be used to predict the corresponding in situ stress. Results of numerical experiments show that the inverse analysis based on finite element modeling of borehole breakouts and artificial neural network is a promising method to determine in situ stress.     

Impacts of small cascaded hydropower plants on river discharge in a basin in Southern China

The run‐off volume altered by the construction of hydropower plants affects ecohydrological processes in catchments. Although the impacts of large hydropower plants have been well documented in the literature, few studies have been conducted on the impacts of small cascaded hydropower plants (SCHPs). To evaluate the impacts of SCHPs on river flow, we chose a representative basin affected by hydropower projects and, to a lesser degree, by other human activities, that is, the Qiuxiang River basin in Southern China. The observed river discharge and climate data during the period of 1958–2016 were investigated. The datasets were divided into a low‐impact period and a high‐impact period based on the number of SCHPs and the capacities of the reservoirs. The daily river discharge alteration was assessed by applying the Indicators of Hydrologic Alteration. To separate the impact of the SCHPs on the local river discharge from that of climate‐related precipitation, the back‐propagation neural network was used to simulate the monthly average river discharge process. An abnormal result was found: Unlike large reservoirs in large watersheds, the SCHPs regulated the flows during the flood season but were not able to mitigate the droughts during the dry season due to their limited storage and the commonly occurring inappropriate interregulations of the SCHPs. The SCHPs also reduced the annual average river discharge in the research basin. The contribution of the SCHPs to the river discharge changes was 85.37%, much higher than the contributions of climate change (13.43%) and other human activities (1.20%). The results demonstrated that the impacts of the SCHPs were different from those of large dams and reservoirs that regulate floods and relieve droughts. It is necessary to raise the awareness of the impacts of these river barriers.     

基于双分支卷积神经网络的SAR与多光谱图像融合实验

针对合成孔径雷达(Synthetic Aperture Radar,SAR)和多光谱(Multi-Spectral,MS)融合图像中存在的空间细节模糊和颜色失真问题,该文兼顾光谱监督和空间细节监督,设计光谱损失函数和空间细节损失函数,提出一种基于双分支卷积神经网络(Convolution Neural Network,CNN)的SAR和MS图像融合算法。该算法网络框架包含光谱保持和细节提升两个分支:光谱保持分支通过上采样MS图像连接到网络的输出,直接将光谱信息传递到融合图像中;细节提升分支对SAR和MS图像通过高通滤波提取高频细节信息,然后应用CNN对细节信息进行特征提取、特征融合及重建,最后将重建的细节信息叠加到上采样的MS图像,得到融合结果。以哨兵-1B GRD级别的SAR图像和Landsat8卫星多光谱图像为实验数据,通过与传统融合算法和深度学习算法RSIFNN进行对比,结果表明,该文算法在定性和定量评价方面效果更好,能够在保持光谱信息的基础上增强多光谱图像的空间细节信息,有利于后续地物分类和目标识别等工作的开展。     

深度学习预测GPS时间序列在探索门源MS6.4地震前兆中的应用

以2016-01-21门源M_S6.4地震为例,提出用深度学习预测的GPS时间序列研究地震前兆。用震中附近门源台（QHME）、民乐台（GSML）及古浪台（GSGL）无震时的GPS时间序列训练LSTM神经网络,得到高精度的GPS时间序列预测模型,再分别对该地区无震时和地震前一段时间的GPS时间序列进行回溯性预测。对比预测时间序列与真实时间序列发现,震前2条时间序列大部分的相似性指标比无震时低,说明震前预测时间序列与真实时间序列差异明显,同时考虑震前时间序列的趋势异常,认为出现了异常时段;3个台站分别在E、N、U方向出现多个异常日期,且不同台站具有相同的异常日期,说明探索到了地震前兆。