基于GTWR-XGBoost模型的四川省PM2.5小时浓度估算

卫星气溶胶光学厚度(AOD)和气象数据已被广泛用于估算空气动力学直径≤2.5μm的地表颗粒物(PM_2.5)浓度.研究高时间分辨率、高精度的PM_2.5浓度估算方法,对及时准确的空气质量预报和大气污染的预防及缓解具有重要意义.使用Himawari-8 AOD小时产品和ERA5气象再分析资料作为估算变量,提出GTWR-XGBoost组合模型,对四川省PM_2.5小时浓度进行估算.结果表明：(1)提出的组合模型运用于全数据集的性能优于KNN、 RF、 AdaBoost、 GTWR、 GTWR-KNN、 GTWR-RF和GTWR-AdaBoost模型,拟合精度指标R²、 MAE和RMSE分别为0.96、 3.43μg·m^-3和5.52μg·m^-3,验证精度指标R²、 MAE和RMSE分别为0.9、 4.98μg·m^-3和7.92μg·m^-3.(2)该模型作用于PM_2.5浓度小时估算...     

基于CNN-GRU的瓦斯浓度预测模型及应用*

为解决传统瓦斯浓度预测方法预测精度低和适用性不强等问题,提出运用卷积神经网络（CNN）提取瓦斯浓度时间序列的变化趋势及局部关联特征,应用门自适应矩估计（Adam）优化的控循环单元神经网络（GRU）,在关联特征基础上进行时序性预测的组合方法,并以铜川玉华煤矿监测数据为样本,对比CNN-GRU组合模型、传统机器学习模型LSTM和GRU模型的预测效果。研究结果表明:CNN-GRU模型的预测精度和收敛速度均优于LSTM和GRU模型;CNN-GRU平均绝对误差和均方根误差分别可降低至0.042,0.006,运行效率分别提高59.15%,35.04%,研究结果可为矿井瓦斯灾害防治提供依据。     

除草剂三氟羧草醚经口暴露对小鼠学习记忆能力的影响

目前,有关三氟羧草醚(Acifluorfen,AF)的神经毒性未见报道且亟待确定.为探讨AF经口暴露对小鼠学习记忆能力的影响及其可能机制,将30只雄性昆明小鼠随机分成生理盐水对照组、0.13、1.3、13和130 mg·kg~(-1)·d~(-1) AF染毒组共5组,灌胃染毒14 d,进行Morris水迷宫实验,观察脑海马病理切片,并检测脑组织中活性氧(ROS)、丙二醛(MDA)、还原型谷胱甘肽(GSH)、磷酸化cAMP反应元件结合蛋白(pCREB)和脑源性神经营养因子(BDNF)含量.结果显示,与对照组相比,AF染毒剂量分别为13和130 mg·kg~(-1)·d~(-1)时,小鼠行为学上学习记忆能力下降;海马细胞排列松散;出现氧化损伤,其中,ROS含量显著升高(p0.05),GSH含量显著减少(p0.05);神经保护能力减弱,其中,pCREB和BDNF水平显著(p0.05)或极显著(p0.01)下降.结果表明,AF能够导致小鼠学习记忆能力下降,氧化应激和CREB-BDNF通路级联下调可能是AF造成神经毒性的机制之一.     

基于周边站点优化选取的随机森林PM2.5小时浓度预测研究

空气中的PM_2.5是威胁人体健康的主要大气污染物,对其进行有效预测和及时预警具有重要意义.大量研究表明,纳入周边站点信息的随机森林模型在单站点PM_2.5预测中显示出良好的效果,但在周边站点选取问题上目前尚缺乏针对性研究,部分选取方法带有主观性.本文提出了一种基于时间滞后互相关分析的周边站点优化选取方法,并以上海十五厂空气质量监测站（国控站）为例,构建了预测该站未来1~24 h PM_2.5浓度的随机森林回归模型集,比较分析了预测模型中各输入因子的重要性.研究发现,预测站点当前PM_2.5浓度值对未来1~16 h的预测最为重要,而气象要素中的风向则对于未来17~24 h的预测重要性最高;周边站点PM_2.5信息随着预测时间的延长,其重要程度排名有明显提升,且不同站点对不同时间预测的影响具有显著差异,在建模时应区别对待,优化选取.比较结果表明,使用本文方法选取周边站点建立的预测模型不仅在RMSE等精度指标上具有一定优势（12 h和24 h预报RMSE分别降低11.8%和13.3%）,还在有实用价值的污染事件空报率上有明显降低（12 h和24 h预报空报率分别降低16.1%和25.6%）,具有业务应用潜力.     

以问题为驱动的环境综合实验教学方式探讨

针对《环境综合实验》课程中学生机械学习、不会思考、缺乏分析和实际运用能力的现象,结合以问题为驱动教学方式的理论基础,探讨了以问题为驱动教学方式在环境综合实验教学中的实际运用,包括预习阶段的学习、实践过程中的学习和实践后的复习,并对以问题为驱动教学方式的运用策略进行了评价和总结。实践表明:以问题为驱动的教学方式能够有效地提高环境综合实验的教学效果,提升学生自主学习和实际动手操作的能力。     

基于MODIS数据与多机器学习法的日PM2.5模拟研究

为了深入了解地面PM_2.5的空间分布,以山东省为研究区,利用2019年的PM_2.5站点实测数据,结合中分辨率成像光谱仪(MODIS)的L3级别的MCD19A2气溶胶光学厚度产品,充分考虑人口、地形、气象等因素,使用RF、SVR、BPNN、DNN等4种机器学习算法对山东省2019年逐日PM_2.5进行了模拟.结果表明:随机森林模型(RF)的RMSE和MAE的值分别为12.67和6.62,优于BPNN、SVR和DNN模型.随机森林模型(RF)最适合山东省的日PM_2.5模拟.     

Wrestling with the complexity of evaluation for organizations at the boundary of science,policy, and practice

Boundary organizations are situated between science, policy, and practice and have a goal of supporting communication and collaboration among these sectors. They have been promoted as a way to improve the effectiveness of conservation efforts by building stronger relationships between scientists, policy makers, industry, and practitioners (Cook et al. 2013). Although their promise has been discussed in theory, the work of and expectations for boundary organizations are less defined in practice. Biodiversity conservation is characterized by complexity, uncertainty, dissent, and tight budgets, so boundary organizations face the challenging task of demonstrating their value to diverse stakeholders. We examined the challenges boundary organizations face when seeking to evaluate their work and thus aimed to encourage more productive conversations about evaluation of boundary organizations and their projects. Although no off‐the‐shelf solution is available for a given boundary organization, we identified 4 principles that will support effective evaluation for boundary organizations: engage diverse stakeholders, support learning and reflection, assess contribution to change, and align evaluation with assumption and values.     

基于“3A”理论的高职高专院校学生英语学习的研究

结合英语教学实践,依据Leo van Lier的意识原则(Awareness)和自主性原则(Autonomy),对高职高专院校英语专业学生进行了问卷调查。调查结果显示,高职高专院校英语专业学生的英语学习意识薄弱,动机不强。3A理论的启示是:发挥学生的主体作用,树立学生的语言学习意识,调动学生学习英语的主观能动性。     

A framework for learning from incidents in the workplace

Learning from Incidents (LFI) in the workplace has been gaining increasing importance in the Health, Safety and Environment context. Although organisations adopt a variety of LFI initiatives, it is often unclear what learning approaches are the most appropriate and the most effective for different types of incidents across a range of contexts. The aim of the paper is to surface factors that are important for effective Learning from Incidents (LFI). The paper builds on a conceptual framework for learning from incidents, developed through an earlier study. This conceptual framework was validated through empirical data collected at two multinational corporations in the energy sector. From this data a refined framework for learning from incidents was devised with five factors important for LFI: participants of learning, type of incidents, learning process, type of knowledge and learning context. This framework can be used as an evaluation tool and as a guidance tool to develop holistic, organisational learning approaches.     

Black swans,cognition, and the power of learning from failure

Failure carries undeniable stigma and is difficult to confront for individuals, teams, and organizations. Disciplines such as commercial and military aviation, medicine, and business have long histories of grappling with it, beginning with the recognition that failure is inevitable in every human endeavor. Although conservation may arguably be more complex, conservation professionals can draw on the research and experience of these other disciplines to institutionalize activities and attitudes that foster learning from failure, whether they are minor setbacks or major disasters. Understanding the role of individual cognitive biases, team psychological safety, and organizational willingness to support critical self‐examination all contribute to creating a cultural shift in conservation to one that is open to the learning opportunity that failure provides. This new approach to managing failure is a necessary next step in the evolution of conservation effectiveness.