基于A- CGAN的深反射地震数据随机噪声压制方法研究

基于深度学习的地震数据噪声压制方法是当前地震数据去噪处理的重要方向。深度学习方法突破了传统滤波处理的局限,在对常规地震数据的噪声压制中表现出效率高、信噪分离效果好的特点。但针对深部弱有效反射数据,当前的深度学习方法特征提取能力有限,难以取得较好的去噪效果。笔者等结合深反射地震数据特点,针对当前深度学习噪声压制方法在特征提取及对数据集依赖上的局限,提出了基于注意力循环生成对抗网络(Attention Cycle- Consistent Generative Adversarial Networks,A- CGAN)的深反射地震数据随机噪声压制方法。借助循环一致生成对抗网络(Cycle- Consistent Generative Adversarial Networks,Cycle- GAN)的域映射思想,降低对数据集的要求。为了构建适用于深反射地震数据的去噪网络,从3个方面对Cycle- GAN进行改进：在Cycle- GAN的生成器（去噪器）中加入残差结构和注意力机制,用于加深网络深度和提高其特征提取能力;在Cycle- GAN的鉴别器中使用块判决,提高鉴别精度和准确度;在损失函数部分加入感知一致性损失函数,提升网络模型恢复纹理细节信息的能力。通过合成地震数据和实际深反射地震数据测试,验证了优化算法的有效性,体现了良好的应用价值。     

基于光谱—空间特征的ASTER影像岩性分类研究——以甘肃北山白峡尼山地区为例

遥感岩性制图是地质填图中的重要工作,基于光谱特征的岩性分类易受到色调、纹理等因素影响导致精度不佳。前人进行岩性自动分类研究多关注影像的光谱特征,而忽略空间特征,笔者等基于甘肃北山白峡尼山地区ASTER影像,将支持向量机、极限学习机两种机器学习分类方法与基于空间特征的快速漂移算法相结合进行岩性分类。结果表明支持向量机分类总体精度为89.17%;极限学习机不但具有需调节参数少的优势,且分类精度和速度均优于支持向量机,分类总体精度达96.70%;利用快速漂移算法提取的影像空间特征可有效减少错分区,提升岩性分类效果。研究证实将基于光谱特征的极限学习机和基于空间特征的快速漂移算法结合的岩性分类方法具有客观、高效、高精度等优势,可为后续地质填图和找矿勘查工作提供可靠数据支撑,在遥感岩性分类领域具有较高的推广价值。     

基于机器学习的锆石成因分类研究

锆石是在自然界中多种温压条件下能够稳定保存,并记录原岩年龄信息的副矿物。锆石微量元素能完整记录地质演化过程信息。通过微量元素分析锆石成因的研究已久,通常利用Th-U图解和La_N-(Sm/La)_N图解等二元图解对锆石进行分类研究。然而,随着锆石研究的深入,以及二元图解无法呈现数据高维度信息的局限性,传统图解已经不能满足对锆石类型进行准确判别,且对已知类型的锆石出现判定偏差。因此,本文将地质大数据与机器学习相结合,训练出高维度锆石成因分类器。文中收集了3 498条不同成因类型的锆石微量元素数据,并通过测试和运用随机森林、支持向量机、人工神经网络和k近邻等4种机器学习算法,最终得出准确率为86.8%的线性支持向量机锆石成因分类器,用于锆石类型的判定与预测。这项工作为锆石分类研究提供了更高维度的判别手段,极大提高了微量元素分析成因结果的精度。将锆石微量元素数据与机器学习方法相结合,是大数据分析与机器学习技术在地球化学研究中的积极探索。     

基于机器学习模型的淮河流域中长期径流预报研究

准确可靠的中长期径流预报是支撑水资源科学调配、提高水资源利用效率的关键。本研究采用AdaBoost模型(AdB)、随机森林模型(RF)和支持向量机模型(SVM)进行淮河流域王家坝和蚌埠站当年11月至次年10月共12个月的中长期径流预报研究。采用置换准确度重要性度量法从130项气象-气候因子及前期降雨/流量构建的1 562个因子变量中筛选出影响各月径流的关键因子,构建了基于AdB、RF和SVM模型的各月径流预报模型,模型参数采用随机搜索技术并结合交叉验证方式确定。采用变幅误差合格率和等级(五级)预报合格率指标对模型的预报精度进行了评估。变幅误差合格率指标表明,王家坝12个月的平均合格率分别为99.8%(AdB)、96.6%(RF)和95.9%(SVM),蚌埠站分别为100%(AdB)、94.8%(RF)和93.8%(SVM);等级预报合格率指标表明,王家坝12个月的平均合格率分别为79.0%(AdB)、76.4%(RF)和79.9%(SVM),蚌埠站分别为81.0%(AdB)、75.6%(RF)和76.6%(SVM)。模型均具有较好的预报效果,但RF和SVM模型对于高流量值的预报存在偏低现象,AdB模型整体上优于RF和SVM模型。     

基于机器学习的表层土壤成矿元素空间预测: 以稀有金属铷元素为例

近些年随着土地质量地球化学调查工作的开展,获取了大量表层土壤样品数据。然而,这些数据也存在一个明显的缺陷,即1∶50 000大比例尺表层土壤数据往往缺少成矿元素。鉴于土壤成矿元素含量对于矿产资源勘查的重要指示作用,尝试基于现有数据对大比例尺表层土壤成矿元素含量提供一个补全方案。以稀有金属铷元素为例,采用随机森林算法把同一区域2 548组1∶250 000小比例尺表层土壤数据按照8∶2的比例随机分为两组,用80%的数据进行训练建模,20%的数据对模型进行验证。采用变量重要性度量排序和构建学习曲线的组合方法优选了8种元素(K、B、Ni、V、Zn、As、Co、Cu)作为预测变量,模型对训练数据和测试数据的拟合优度R²分别达到0.983 2和0.895 6,说明预测变量的优选方法是有效的。随后将1∶50 000表层土壤的上述预测变量数据作为输入变量导入模型中,得到预测的Rb元素含量,预测结果比较符合实际特征。本研究表明将大数据机器学习随机森林算法引入表层土壤地球化学元素含量空间定量预测具有可行性,可进一步拓展土地质量地球化学数据的服务应用维度。     

A Bayesian network approach for predicting seismic liquefaction based on interpretive structural modeling

The Bayesian network (BN) is a type of graphical network based on probabilistic inference that has been gradually applied to assessment of seismic liquefaction potential. However, how to construct a robust BN remains underexplored in this field. This paper aims to present an efficient hybrid approach combining domain knowledge and data to construct a BN that facilitates the integration of multiple factors and the quantification of uncertainties within a network model to assess seismic liquefaction. Initially, only using given domain knowledge, a naive network model can be constructed using interpretive structural modeling. Thereafter, some effective information about the naive model is provided to construct a robust model using structural learning of BN from historic data. Finally, the returning predictive results and the predictive results are compared to other methods including non-probabilistic and probabilistic models for seismic liquefaction using the metrics of the overall accuracy, the area under the curve of receiver operating characteristic, prediction, recall and F₁ score. The methodology proposed in this paper achieved better performance, and we discussed the power and value of the proposed approach at the end of this paper, which suggest that BN is a good alternative tool for seismic liquefaction prediction.     

Social capital and community preparation for urban flooding in China

Social capital can enhance community resilience to environmental change. Productive and trusted relations among social actors and effectual social norms can help local residents share resources, information and risks. The main objective of our study is to understand the ways in which social attributes and risk considerations influence adoption of resilient economic measures by individuals for reducing potential losses due to catastrophic rainstorm and flooding. This article provides evidence from China on how social capital contributes to anticipatory adaptation to environmental change. The inquiry is based on structured interviews with local residents of Tianjin, a flood-prone port city in China, and a standard regression analysis. Findings show that the intention to make preparation increases with the levels of social expectation, social relationship, and institutional trust. Perceived risk and damage experience, however, have no significant impacts. This suggests that building social capacity and trust will be more effective in enhancing community resilience than merely increasing awareness of hazard risks. We call for greater efforts on strengthening the capacity of formal and informal communal institutions. The structural changes required, however, are challenging.     

Environmental governance through “Speaking Like an Indigenous State” and respatializing resources: Ethical livelihood concepts in Bolivia as versatility or verisimilitude?

This research addresses recent environmental governance in Bolivia through its relations to indigeneity and respatializations. It introduces and develops the concept of “speaking like an indigenous state” to examine the Bolivian state’s recent use of a pair of indigenous linguistic concepts, Living Well and Earth Mother, representing the identities of citizens and their rights to resources and livelihoods. State relations to indigenous social movements highlight the use of Living Well and Earth Mother concepts through accommodation, resistance, and protaganism. Six active issues of environmental governance are examined: (1) climate change and justice movement; (2) agrarian reform, agrobiodiversity, and food justice; (3) water resources; (4) indigenous territories; (5) Protected Areas; and (6) extractive industries (mining, hydrocarbons). The usages of Living Well and Earth Mother show versatility as they have been mobilized in the respatializing of the politics and social-power dynamics of environmental issues at scales of the state, global and international institutions, and community and local levels. Analysis also reveals deployment of Living Well and Earth Mother that is discursively influential and yet conceptually reduced and unevenly applied, thus suggesting a characteristic of verisimilitude. My analysis determines that respatialization at various levels, including territorial transitions of sub-national regional spaces, are associated with the heightened articulation of environmental governance through indigeneity and “speaking like an indigenous state” amid resource nationalism. Linkages and logics operating within this conjuncture differ from the prevailing interpretation of the Bolivian state’s use of Living Well and Earth Mother as solely an unwitting contradiction or instrumentalist camouflage.     

Experimenting with spaces of encounter: Creative interventions to develop meaningful contact

To-date geographical research on encounters has primarily comprised observation of naturalistic settings (both micro publics and everyday public spaces) or narrative accounts of encounters generated by conventional methods, this paper focuses on a contrived spatial experiment to create meaningful contact across difference. Inspired by, and drawing on insights from, Architecture – a discipline relatively neglected by geographers in recent times – we use a creative form of spatial play to stimulate groups of people to explore their differences and to develop shared understandings. Whereas previous studies have identified the importance of material objects in art projects and home spaces in mediating relations between people, surprisingly little attention has been paid to the physical configuration of space in work on encounters, despite the fact that geography is a spatial discipline. Here, we engage directly with the role of the form and function of space. Our spatial experiments demonstrate how paying attention to the materiality of micro-spaces rather than social relations alone, can provide insights into the generation of positive interactions by contributing to a greater understanding of how meaningful encounters happen, and what can be done to facilitate them. Specifically, by exploring the size and configuration of space, issues of ownership and surveillance, the relationship between primary and secondary space, and aural architecture this research identifies how working together to create a shared private or intimate space – might facilitate a sense of empowerment and the production of social relations characterised by democracy and inclusion.     

Novel insights on intensity and typology of direct human-nature interactions in protected areas through passive crowdsourcing

Recent advances in geotagging, sharing and automatically analyzing online content from Social Networking Sites (SNS) offer unprecedented opportunities for the analysis of human-nature interactions. Previous studies in this field, however, offer limited insights regarding the benefits of automated content analysis especially at large scales, biases arising from the selection of SNS sources, and the predictive power of visitation models based on SNS data. We explore quantitative and qualitative aspects related to intensity, interests and sentiments associated with on-site experiences in 568 protected areas in Israel and the Palestinian Authority. We analyze counts and content of >100,000 photographs and tweets from four different SNSs, calibrate visitation models and predict visitation in unmonitored sites, cluster sites based on the typology of human-nature interactions reflected in online photographs, and characterize the polarity of sentiments associated with experiences in individual sites and clusters thereof. We find benefits in combining data from multiple sources and controlling for biases related to sites’ photogenicity and type of human-nature interactions. Our results suggest that current best estimates of visitation in unmonitored sites underestimate by 39% the actual number of visits. We discuss how the techniques and findings in this study are applicable in the broader context of the management and conservation of sites of environmental or cultural interest.