地质矿产术语分类代码在地矿点源信息系统中的应用

论述了如何利用地质矿产术语分类代码来建立有关数据字典,如何在数据模式设计以及地质数据描述上采用这套代码,实现地质信息共享。重点论述了地在太点源信息系统中的方法和所要解决的问题。     

Air pollution hazard assessment using decision tree algorithms and bivariate probability cluster polar function: evaluating inter-correlation clusters of PM10 and other air pollutants

ABSTRACT

The automated classification of ambient air pollutants is an important task in air pollution hazard assessment and life quality research. In the current study, machine learning (ML) algorithms are used to identify the inter-correlation between dominant air pollution index (API) for PM10 percentile values and other major air pollutants in order to detect the vital pollutants’ clusters in ambient monitoring data around the study area. Two air quality stations, CA0016 and CA0054, were selected for this research due to their strategic locations. Non-linear RPart and Tree model of Decision Tree (DT) algorithm within the R programming environment were adopted for classification analysis. The pollutants’ respective significance to PM10 occurrence was evaluated using Random forest (RF) of DT algorithms and K means polar cluster function identified and grouped similar features, and also detected vital clusters in ambient monitoring data around the industrial areas. Results show increase in the number of clusters did not significantly alter results. PM10 generally shows a reduction in trend, especially in SW direction and an overall minimal reduction in the pollutants’ concentration in all directions is observed (less than 1). Fluctuations were observed in the behaviors of CO and NO_x during the day while NOx displayed relative stability. Results also show that a direct and positive linear relationship exists between the PM10 (target pollutant) and CO, SO₂, which suggests that these pollutants originate from the same sources. A semi-linear relationship is observed between the PM10 and others (O₃ and NO_x) while humidity shows a negative linearity with PM10. We conclude that most of the major pollutants show a positive trend toward the industrial areas in both stations while tra?c emissions dominate this site (CA0016) for CO and NOx. Potential applications of nuggets of information derived from these results in reducing air pollution and ensuring sustainability within the city are also discussed. Results from this study are expected to provide valuable information to decision makers to implement viable strategies capable of mitigating air pollution effects.     

Multi-source spatial data-based invasion risk modeling of Striga (Striga asiatica) in Zimbabwe

ABSTRACT

Monitoring of destructive invasive weeds such as those from the genus Striga requires accurate, near real-time predictions and integrated assessment techniques to enable better surveillance and consistent assessment initiatives. Thus, in this study, we predicted the potential ecological niche of Striga (Striga asiatica) weed in Zimbabwe, to identify and understand its propagation and map potentially vulnerable cropping areas. Vegetation phenology from remote sensing, bioclimatic and other environmental variables (i.e. cropping system, edaphic, land surface temperature, and terrain) were used as predictors. Six machine learning modeling techniques and the ensemble model were evaluated on their suitability to predict current and future Striga weed distributional patterns. The mentioned predictors (n = 40) were integrated into six models with “presence-only” training and evaluation data, collected in Zimbabwe over the period between the 12^th and 28^th of March 2018. The area under the curve (AUC) and true skill statistic (TSS) were used to measure the performance of the Striga modeling framework. The results showed that the ensemble model had the strongest Striga occurrence predictive power (AUC = 0.98; TSS = 0.93) when compared to the other modeling algorithms. Temperature seasonality (Bio4), the maximum temperature of the warmest month (Bio5) and precipitation seasonality (Bio15) were determined to be the most dominant bioclimatic variables influencing Striga occurrence. “Start of the season” and “season minimum value” of the “Enhanced Vegetation Index base value” were the most relevant remote sensing-based variables. Based on projected climate change scenarios, the study showed that up to 2050, the suitable area for Striga propagation will increase by ~ 0.73% in Zimbabwe. The present work demonstrated the importance of integrating multi-source data in predicting possible crop production restraints due to weed propagation. The results can enhance national preparedness and management strategies, specifically, if the current and future risk areas can be identified for early intervention and containment     

Increasing the detail of European land use/cover data by combining heterogeneous data sets

ABSTRACT

Data on land use and land cover (LULC) are a vital input for policy-relevant research, such as modelling of the human population, socioeconomic activities, transportation, environment, and their interactions. In Europe, CORINE Land Cover has been the only data set covering the entire continent consistently, but with rather limited spatial detail. Other data sets have provided much better detail, but either have covered only a fraction of Europe (e.g. Urban Atlas) or have been thematically restricted (e.g. Copernicus High Resolution Layers). In this study, we processed and combined diverse LULC data to create a harmonised, ready-to-use map covering 41 countries. By doing so, we increased the spatial detail (from 25 to one hectare) and the thematic detail (by seven additional LULC classes) compared to the CORINE Land Cover. Importantly, we decomposed the class ‘Industrial and commercial units’ into ‘Production facilities’, ‘Commercial/service facilities’ and ‘Public facilities’ using machine learning to exploit a large database of points of interest. The overall accuracy of this thematic breakdown was 74%, despite the confusion between the production and commercial land uses, often attributable to noisy training data or mixed land uses. Lessons learnt from this exercise are discussed, and further research direction is proposed.     

基于Sentinel-1A数据的多种机器学习算法识别冰山的比较

冰山识别对于海洋环境监测和船只安全运行等具有重要的意义,是北极航道开通和北极开发过程中的重要内容。采用合成孔径雷达(SAR)影像进行冰山识别具有独特的优势,多种机器学习算法均可用于SAR影像的冰山识别中。为了最大限度地发挥机器学习算法的性能,有必要对不同机器学习算法及其搭配使用的特征与特征标准化方法进行评估,从而进行最优冰山识别方法的选择。因此,本文基于Sentinel-1A SAR影像,采用多种机器学习方法、多种特征组合及多种特征标准化方法进行冰山识别,并比较各流程方法的识别性能差异。采用的机器学习算法包括贝叶斯分类器(Bayes)、反向神经网络(BPNN)、线性判别分析(LDA)、随机森林(RF)以及支持向量机(SVM);特征标准化方法包括Min-max标准化、Z-score标准化及log函数标准化;数据集是含有12个SAR影像特征的969个冰山与非冰山样本,样本主要位于格陵兰岛东海岸。分类效果采用接收者操作特性(ROC)曲线下的面积(AUC)进行衡量。结果显示,最佳搭配下的RF的AUC值最高,达到了0.945,比最差的Bayes高出0.09。从识别率上来看,RF在冰山查全率为80%的情况下非冰山查全率达到92.6%,效果最好,比第2位的BPNN高出1.4%,比最差的Bayes高出2.6%;BPNN在冰山查全率为90%的情况下非冰山查全率达到87.4%,比第2位的RF高出0.8%,比最差的Bayes高出2.7%。上述结果表明,对冰山识别而言,选择最优的机器学习算法和最佳的特征与特征标准化方法都是十分重要的。     

一种高分辨率遥感影像建筑物自动检测方法

高分辨率遥感影像具有丰富的空间信息、地物几何结构和纹理信息,有助于对地物目标进行认知和解译。而建筑物目标在人类活动区域内占据重要地位,对高分辨率遥感影像中的建筑物进行自动检测具有重大意义。提出了一种基于全卷积神经网络的建筑物自动检测方法,并制作了建筑物样本数据集,利用基于区域的全卷积神经网络和特征检测网络进行建筑物检测模型的参数训练,对待检测影像进行预处理之后利用模型进行建筑物检测,得到影像中的建筑物目标的具体位置和类别置信度。实验证明,提出的检测方法具有更好的效果和更快的速度。检测召回率达到92%,检测准确率达到98%,证明了该方法针对建筑物检测具有较高的精度和较强的稳定性。     

成分数据理论和无监督聚类K- means方法提取背景和异常信息——以安徽省兆吉口铅锌矿床为例

地球化学数据是应用地球化学研究的重要组成部分，是化学勘查工作的基础成果。勘查地球化学数据基本上以元素的质量百分浓度 (简称浓度)的形式表达，是典型的成分数据。其表达的是“组分/总体”相对质量贡献信息，而不是绝对的质量变化信息。浓度数据分布在单纯形空间，而不是整个欧式空间。对成分数据进行处理之前，进行适当的对数比值转换处理可以提高其信息表达。本文以安徽省兆吉口铅锌矿床土壤中Pb数据为示范案例，通过对数比值转换方法优化浓度数据的结构以提高相对信息的表达，并利用无监督学习K- means聚类方法根据对数比值转换数据分布空间质心的距离识别背景和异常信息，最后对K- means聚类方法识别的背景和异常与迭代2倍标准差法和浓度- 面积分形分析法进行比较以衡量其表现。结果表明：浓度数据表达的是相对质量信息，而不是绝对质量关系，不同样品间不能通过比较浓度高低推断出质量的多寡关系。对数比值方法可以有效地提高浓度数据的结构和信息表达，K- means方法能够准确识别对数比值转换数据的背景和异常信息，其效果类似浓度- 面积分形分析方法，比迭代2倍标准差法好。     

Mapping Soils,Vegetation, and Landforms: An Integrative Physical Geography Field Experience*

Students in a graduate seminar at Michigan State University produced a series of detailed vegetation, soils, and landform maps of a 1.5‐square‐mile (3.9 km²) study area in southwest Lower Michigan. The learning outcomes (maps) and skill development objectives (sampling strategies and various GIS applications) of this field‐intensive mapping experience were driven by the assumption that students learn and understand relationships among physical landscape variables better by mapping them than they would in a classroom‐based experience. The group‐based, problem‐solving format was also intended to foster collaboration and camaraderie. The study area lies within a complex, interlobate moraine. Fieldwork involved mapping in groups of two or three, as well as soil and vegetation sampling. Spatial data products assembled and used in the project included topographic maps, a digital elevation model (DEM), aerial photographs, and NRCS (National Resource Conservation Service) soil maps. Most of the soils are dry and sandy, with the main differentiating characteristic being the amount of, and depth to, subsurface clay bands (lamellae) or gravelly zones. The presettlement (early 1830s) vegetation of the area was oak forest, oak savanna, and black oak “barrens.” Upland sites currently support closed forests of white, black, and red oak, with a red maple, dogwood, and sassafras understory. Ecological data suggest that these oak forests will, barring major disturbance, become increasingly dominated by red maple. This group‐based, problem‐solving approach to physical geography education has several advantages over traditional classroom‐based teaching and could also be successfully applied in other, field‐related disciplines.     

贝叶斯网络结构学习及其应用研究

阐述了贝叶斯网络结构学习的内容与方法 ,提出一种基于条件独立性 (CI)测试的启发式算法。从完全潜在图出发 ,融入专家知识和先验常识 ,有效地减少网络结构的搜索空间 ,通过变量之间的CI测试 ,将全连接无向图修剪成最优的潜在图 ,近似于有向无环图的无向版。通过汽车故障诊断实例 ,验证了该算法的可行性与有效性。     

A hybrid genetic—instance based learning algorithm for CE-QUAL-W2 calibration

This paper presents a calibration model for CE-QUAL-W2. CE-QUAL-W2 is a two-dimensional (2D) longitudinal/vertical hydrodynamic and water quality model for surface water bodies, modeling eutrophication processes such as temperature–nutrient–algae–dissolved oxygen–organic matter and sediment relationships. The proposed methodology is a combination of a ‘hurdle-race’ and a hybrid Genetic-k-Nearest Neighbor algorithm (GA-kNN). The ‘hurdle race’ is formulated for accepting–rejecting a proposed set of parameters during a CE-QUAL-W2 simulation; the k-Nearest Neighbor algorithm (kNN)—for approximating the objective function response surface; and the Genetic Algorithm (GA)—for linking both. The proposed methodology overcomes the high, non-applicable, computational efforts required if a conventional calibration search technique was used, while retaining the quality of the final calibration results. Base runs and sensitivity analysis are demonstrated on two example applications: a synthetic hypothetical example calibrated for temperature, serving for tuning the GA-kNN parameters; and the Lower Columbia Slough case study in Oregon US calibrated for temperature and dissolved oxygen. The GA-kNN algorithm was found to be robust and reliable, producing similar results to those of a pure GA, while reducing running times and computational efforts significantly, and adding additional insights and flexibilities to the calibration process.