基于机器学习算法的县域台风灾害损失风险评估

基于2008—2019年我国台风县(区)灾情的直接经济损失数据,根据经济损失率将台风灾害经济损失风险分为五类,考虑台风灾害的致灾因子和孕灾环境因子共选取10个解释变量,采用五种经典的机器学习算法,包括支持向量机(Support Vector Machine,SVM)、随机森林(Random Forest,RF)、AdaBoost、XGBoost(Extreme Gradient Boosting Machine)和LightGBM(Light Gradient Boosting Machine),分别构建台风灾害经济损失风险评估模型,选出准确率最高的模型,应用于经典台风过程并进行检验评估。结果表明:基于RF算法的台风灾害经济损失风险模型的准确率最高;利用RF、XGBoost、LightGBM、AdaBoost和SVM算法构建模型的准确率依次为0.69、0.63、0.62、0.45和0.41。选择RF算法构建的台风灾害经济损失风险模型的解释变量表明,致灾因子是最主要的解释变量,其中,降雨导致损失的重要性超过风速。该模型在训练集和测试集上对风险分类的TS评分为0.55和0.51,但对每种风险类别的辨别能力存在差异,对于最低风险和最高风险的分类效果较好,对于较高风险和中等风险的分类能力不足。利用该模型对2017年第13号台风“天鸽”的经济损失进行检验评估,评估结果与实际台风灾害经济损失的风险等级较一致,各风险等级的准确率均达到0.7以上,TS评分在0.58以上,空报率和漏报率分别在0.31和0.25以下。      

Benefits and drawbacks of two modelling approaches for a generalist carnivore: can models predict where Wile E. Coyote will turn up next?

Wildlife ecologists frequently make use of limited information on locations of a species of interest in combination with readily available GIS data to build models to predict space use. In addition to a wide range of statistical data models that are more commonly used, machine learning approaches provide another means to develop predictive spatial models. However, comparison of output from these two families of models for the same data set is not often carried out. It is important that wildlife managers understand the pitfalls and limitations when a single set of models is used with limited GIS data to try to predict and understand species distribution. To illustrate this, we carried out two sets of models (generalized linear mixed models (GLMMs) and boosted regression trees (BRTs)) to predict geographic occupancy of the eastern coyote (Canis latrans) on the island of Newfoundland, Canada. This exercise is illustrative of common spatial questions in wildlife research and management. Our results show that models vary depending on the approach (GLMM vs. BRT) and that, overall, BRT had higher predictive ability. Although machine learning has been criticized because it is not explicitly hypothesis-driven, it has been used in other areas of spatial modelling with success. Here, we demonstrate that it may be a useful approach for predicting wildlife space use and to generate hypotheses when data are limited. The results of this comparison can help to improve other models for species distributions and also guide future sampling and modelling initiatives.     

耦合注意力机制DNN的PM2.5估算及时空特征分析

PM_2.5作为指示环境质量的重要因子之一,不仅影响着灰霾天气的发生,还与公众健康息息相关,近年来受到广泛的关注。尽管PM_2.5地面观测站点在不断地扩张,其覆盖范围依旧有限,难以反映全域PM_2.5浓度的时空异质性。本研究运用卫星遥感气溶胶光学厚度数据,辅助因子除常规的气象因子等以外,还加入了针对中国人民生产生活习惯的农历日因子,提出一种耦合注意力机制的深度神经网络模型,对长三角区域2015年—2020年PM_2.5浓度进行了逐日的高精度估算。模型交叉验证结果显示决定系数R²高达0.85,斜率0.86,与地面站点观测值有较高的一致性,优于多元线性回归和随机森林模型。长三角区域PM_2.5浓度时空特征分析结果表明,PM_2.5浓度在空间上呈现北高南低的趋势;季节特征以冬季浓度最高,夏季浓度最低,春秋过渡。此外,长三角区域2015年—2020年整体PM_2.5浓度呈下降趋势,其中以上海市最为明显,下降速率为3.30μg/(m     

Integration of Geospatial Science in Teacher Education

Abstract

One of the primary missions of our university is to train future primary and secondary teachers. Geospatial sciences, including GIS, have long been excluded from teacher education curriculum. This article explains the curriculum revisions undertaken to increase the geospatial technology education of future teachers. A general education class introducing geospatial technology to the general student body has been developed, a cartography class has been modified to provide applied geospatial experience explicitly for future teachers, and a service learning partnership with local K–12 schools has been established where students are working with teachers to integrate geospatial sciences in their academic programs.     

基于多源数据和集成学习的城市住宅地价分布模拟——以武汉市为例

精准刻画城市住宅地价分布特征,对于科学引导城市空间布局规划、有效实现城市精明增长等具有重要意义。而城市住宅地价与其潜在影响因素之间的复杂非线性关系,给地价分布精细模拟带来了挑战。论文旨在探索基于地理大数据和集成学习的城市住宅地价分布模拟方法体系,以满足快速、精准监测地价动态变化的需要。选取武汉市为典型区,以住宅用地交易样点、兴趣点(points of interest, POI)和夜间灯光影像为数据源,以500 m分辨率网格为估价单元,提取POI核密度和夜间灯光强度作为住宅地价预测变量,采用机器学习算法和bagging、stacking集成方法构建住宅地价预测模型,并对比分析其精度。研究发现：① 单个机器学习算法中,支持向量回归(support vector regression, SVR)预测精度最高,接下来依次是k最近邻算法(k-nearest neighbor algorithm, k-NN)、高斯过程回归(Gaussian process regression, GPR)和BP神经网络(back propagation neural networks, BP-NN);② 在提升单个算法预测精度方面,stacking方法的性能优于bagging方法,使用stacking集成SVR和k-NN的地价预测模型精度最高,其平均绝对百分误差仅为8.29%,拟合优度R²达0.814;③ 基于论文所构建模型生成的城市住宅地价分布图能有效表征价格圈层分布特征和局部奇异性。研究结果可为城市住宅地价评估提供新的思路和方法借鉴。     

Climate-governance entrepreneurship,higher-order learning,and sustainable consumption: the case of the state of Oregon,United States

The ongoing devolution of climate policy-making to sub-national levels has prompted growing interest in policy entrepreneurship by individuals who are politically and technically creative and institutionally resourceful. This paper investigates the case of the materials-management programme in the Oregon Department of Environmental Quality which has emerged as a national and international leader by focusing on the role of household consumption in greenhouse gas (GHG) emissions. Two noteworthy innovations involve the development of a consumption-based GHG emissions inventory and introduction of policies aimed at facilitating construction of small homes (so-called Accessory Dwelling Units, ADU). The case traces over several decades the higher order learning processes within the group and their entrepreneurship toward affecting broader changes in emission accounting and climate policies in Oregon. The paper identifies the enabling factors for these innovations, and considers: how to create the conditions for learning, experimentation, and policy entrepreneurship; how to reproduce these conditions in different locales; and how to recognize and foster innovations that arise outside the established mainstream ‘climate community’. It also stresses the benefits of breaking down the barriers between science-based analysis and policy. The two questions frequently raised in the climate policy debate – how to bring researchers and practitioners together to develop efficacious policies; and how to replicate successful programmes and policies across different communities, jurisdictions, and locations – should be re-examined. It may be more appropriate to ask instead: How to create conditions for learning, experimentation, and policy entrepreneurship; and how to reproduce these conditions in different locales.

Key policy insights

• Using a consumption-based greenhouse gas emission inventory instead of a sector-based inventory radically changes climate policy priorities, shifting the emphasis from technological fixes to curbing household consumption.

• Policy innovations thrive in teams that combine technical and scientific competencies with: a commitment to addressing societal problems; interest in inquiry, experimentation, and learning; entrepreneurship; and strategic and political savvy.

• These qualities require breaking down artificial barriers between science and policy.

• Transformative policy ideas can originate within institutional nodes that operate outside of an established community of expertise and authority; and these should be identified and fostered.

     

Simulating alpine vegetation net primary productivity by remote sensing in Qinghai Province,China

Primary productivity of ecosystem is important indicator about ecological assessment. Remote sensing technology has been used to monitor net primary productivity （NPP） of ecological system for several years. In this paper, the remotely sensed NPP simulation model of alpine vegetation in Qinghai Province of Tibet Plateau was set up based on the theory of light use efficiency. Firstly a new approach based on mixed pixels and Support Vector Machine （SVM） algorithm were used to correct simulated NPP values derived from Moderate Resolution Imaging Spectroradiometer （MODIS） data. Finally, spatial distribution and monthly variation characteristics of NPP in Qinghai Province detail. The result showed in 2006 were analyzed in that NPP of vegetation in Qinghai Province in 2006 ranged from o to 422 gC/m2/a and the average NPP was 151 gC/m2/a. NPP gradually increased from northwest to southeast. NPP of different vegetation types were obviously different. The average NPP of broad-leaved forest was the largest （314 gC/m2/a）, and sparse shrub was the smallest （101 gC/m2/a）. NPP in Qinghai Province significantly changed with seasonal variation. The accumulation of NPP was primarily in the period （from April to September） with better moist and heat conditions. In July, the average NPP of vegetation reached the maximum value （43 gC/m2）. In our model, the advantage of traditional LUE models was adopted, and our study fully considered typicalcharacteristics of alpine vegetation light use efficiency and environmental factors in the study area. Alpine vegetation is the most important ecological resource of Tibet Plateau, exactly monitoring its NPP value by remote sensing is an effective protection measure.     

利用深度学习模型智能识别地下排水管道缺陷

排水管道健康状况直接影响整个城市的排水效果,CCTV检测作为目前最为常见的排水管道健康状况检测方法,仍存在自动化程度不高、工作效率低下、严重依赖人工经验等问题。为解决以上问题,本文将当前先进的深度学习技术与地下排水管道缺陷检测技术相结合,提出了一种基于深度学习模型的地下排水管道缺陷智能识别技术。同时,将管道缺陷智能识别与管道检测工作流程紧密结合,实现城市排水管道检测报告的快速自动生成等功能,从而大大提高排水管道缺陷的检测效率。     

利用深度学习进行GF-6影像枣园检测识别

针对新疆南疆大规模枣园的检测识别,本文提出了一种基于泛化迁移深度学习的枣园目标检测识别方法。以GF-6卫星影像数据为基础制作了Jujube数据集,并将其泛化扩充增强;以Faster R-CNN体系为基础,利用多态协同模式实现数据集的有效关联和优化重构,进行检测识别模型的迁移深度学习以提高对目标对象检测识别的泛化能力。结果表明,模型算法的验证识别精确率、召回率和调和平均值分别达0.979、0.952和0.965,在应用测试中,3个指标平均值均大于0.929,优于传统检测方法,且本文模型方法总体分类精度为0.97,Kappa系数为0.93,均高于面向对象最邻近法,能够有效地满足研究区规模化枣园目标检测识别的精度和效率的要求,为精细化枣园田间管理提供基础依据。     

面向多源数据地物提取的遥感知识感知与多尺度特征融合网络

房屋等建筑物的分布情况是衡量地区发展的重要指标,利用遥感影像实现建筑物的自动高精度提取在指导城乡规划和市镇建设等方面具有重要意义。已有方法大多忽略了像素数较少的小面积建筑和边缘等细节信息的处理,针对此问题,提出了一种双路细节关注网络,将语义特征与细节关注特征双路并行优化,进一步提高了遥感影像中建筑物的提取精度。所提方法首先使用双路特征提取模块获取语义特征与细节关注特征,并在解码过程中进行双向优化,增强语义特征细节的同时提高细节关注特征的连续性与类别准确性,然后对二者进行融合,结合细节关注损失的监督,实现建筑物的高精度提取。在WHU建筑物数据集、ISPRS Vaihingen数据集与某地区国产高分数据集上,将所提方法与多种主流方法进行了对比验证,所提方法的F1分数和交并比均高于对比方法,且提取的建筑物完整性更好,小面积建筑漏检、误检率更低。