基于贝叶斯机器学习的生态模型参数优化方法研究

参数优化方法是准确估计生态模型参数、降低其不确定性的有效手段。本文提出一种基于贝叶斯机器学习的No-U-Turn Sampler（NUTS）生态模型参数优化方法。NUTS是一种高效的参数优化方法,每次取样中利用递归算法生成候选参数集（二叉树）推断参数的后验信息,如果满足约束条件“非U型回转”,不断构建子树更新参数;否则,记录本次抽样的“最优”参数集,并开始下一次取样,直到获取足够样本。该算法在每次取样中充分优化参数,避免因随机游走行为产生冗余抽样,提高了参数优化效率。本文以千烟洲亚热带人工针叶林碳通量模拟为例,基于Pymc3框架利用NUTS参数优化方法实现了碳通量（Net Ecosystem Exchange,NEE）模型参数反演,并与Metropolis-Hastings（MH）方法进行对比。结果表明,本文算法的参数值达到稳定波动时的抽样次数减少了85%左右,参数优化效率提升3倍左右。参数优化后,2种NEE模型中7个参数不确定性降低10%~53%。此外,NEE模拟效果明显提升,模拟值与实测值的R²分别提高23%和17%,RMSE分别降低3%和4%。综上所述,本文提出的参数优化方法对生态领域的参数估计或数据同化工作具有一定的借鉴意义。     

网络视角下的国有和民营制造企业对欧盟投资的区位选择

在推进“一带一路”建设和推动形成全面开放格局的进程中,越来越多企业走出国门,在全球构建起愈发紧密的联系网络。国有企业和民营企业的海外投资区位选择和差异成为国内外经济地理研究的前沿领域。有别于传统的对区位资源论的分析,本文尝试将映射在区域尺度的国有企业和民营企业网络资源运用于中资企业对欧盟投资区位选择的研究中,并进一步区分国有企业和民营企业网络资源的异质性影响,以期丰富中资企业对欧盟投资的区位研究。研究结果显示现有国有企业网络资源不仅影响后续国有企业的投资选址,也影响后续民营企业的投资选址,但现有民营企业网络资源仅影响后续民营企业的投资选址。从时间维度看,2008—2012年国有企业网络资源和民营企业网络资源分别对后续国有企业和民营企业投资选址产生影响,2013年以后,国有企业网络资源深刻影响后续民营企业的投资选址。这一结果表明在“一带一路”倡议实施过程中,国有企业对推动民营企业“走出去”发挥重要作用。     

Estimating geographic concentrations of quaternary industries in Europe using Artificial Light-At-Night (ALAN) data

ABSTRACT

Mapping geographic concentrations of quaternary industries (QIs) may help to assess regional performance and formulate informed development policies. However, fine resolution data on QIs concentrations are sparsely reported. Thus, for the year 2010, only 45% of all NUTS3 regions (i.e. regions of the third and most detailed level of the Nomenclature of Units for Territorial Statistics of the EU) provide relevant information. In this study, we investigate a possibility that artificial light-at-night (ALAN), captured by satellite sensors, can help to identify geographic concentrations of QIs. In this study, we use year-2010 NUTS3 Eurostat data, and combine them with data on ALAN intensities, obtained from the U.S. Defense Meteorological Satellite Program (US-DMSP) for the years 2000 and 2010. In both ordinary least squares (OLS) and spatial dependency (SD) models, ALAN emerged as a statistically significant predictor (t?=?8.392–14.608; P?<?.01), helping to explain, along with other predictors, up to 75% of QIs regional variation. The obtained models and regional data presently available enabled estimates of QIs concentrations for European NUTS3 regions with missing data.