基于迁移学习方法的山东冷云冰晶特征分布研究

为分析冷云中冰晶的分布特征,揭示冰晶增长演变机制,根据冰晶形状和尺度特征分为8类并进行标注,同时标注1类隔断栏进行数据质量控制,将9类标签图像整合并建立图像集,利用迁移学习VGGNet16方法进行识别训练,经训练模型分类准确率达98%。将模型应用到秋季冷云冰晶特征研究中,选取3次积层混合云和3次层状云降水过程,分析冰晶形状在不同温度区间的占比及冰晶谱变化特征,结果表明,温度通过影响冰晶基面与棱面的比值来决定冰晶初始形状分布,相同温度区间积层混合云内球状冰晶和线型冰晶占比高于层状云,低于-12℃后各类冰晶占比相对固定;积层混合云内线型冰晶直径集中在300~800 μm,冰晶谱呈多峰分布,聚合体直径大于600 μm,冰晶谱首尾两端浓度相当,球状冰晶直径集中在120～300 μm,冰晶谱呈单调下降趋势。     

基于注意力机制与自适应时序分解的气温预报模型

针对传统方法在捕捉气象序列长期依赖关系及泛化性能上的不足,提出了一种基于稀疏注意力与自适应时序分解的气温预报模型（ATFSAS）。该模型整体采用编码器 解码器架构,结合稀疏注意力机制以有效捕捉气象观测数据间的长期依赖性。为减少编码过程中造成的冗余,提出了一种信息蒸馏方法。通过结合多层解码器与自适应时序分解单元,逐步细化预报信号中的周期性和趋势性分量,实现了较为精准的气温预报。基于德国耶拿气象数据集,进行24 h精细化气温预报,其平均绝对误差为1.7108℃。基于中国地面气候资料日值数据集,进行中短期日平均气温预报和多地区单日平均气温预报,相比传统模型LSTM,ATFSAS模型预报结果的平均绝对误差分别提升了35.56%和23.66%。     

基于机器学习的格点气温预报订正方法

使用2017年9月至2021年3月国家级业务化运行的智能网格实况分析产品和欧洲中期天气预报中心全球模式(EC)产品,根据湖北省的地理分布特征构建6个分区,采用基于LightGBM机器学习算法建立的气温预报方法,生成湖北省0.05°×0.05°格点气温预报产品。利用2021年4—9月的预报产品和格点实况资料进行检验,结果表明:基于机器学习的气温预报方法(MLT)取得了较好的预报效果,其在0～72 h时效内优于中央气象台下发的气温精细化指导预报(SCMOC)和EC产品;MLT在山区的误差较平原大,但山区的订正幅度大于平原,日最高气温的订正幅度大于日最低气温的订正幅度;4—9月MLT、SCMOC、EC产品的平均绝对误差(MAE)日变化都呈现了白天偏高、夜间偏低、午后凸起的单峰特征,MLT的MAE值较SCMOC和EC产品的更低,并且在转折性天气中仍具有优势;站点检验与格点检验结论一致,基于格点建模的气温预报产品对站点预报同样得到了订正。机器学习在格点气温的模式订正方面可以作为一个行之有效的手段。     

Comparing Human and Neural Network Learning of Climate Categories

This study compares how humans and neural networks classify climate types. Human subjects were asked to classify climates from monthly temperature and precipitation patterns. To model their learning process, the same data were used to produce input vectors that trained a pattern associator neural network. Both human subjects and the neural network classified climates accurately after 10 rounds of supervised learning. The neural network successfully modeled the rate of human learning and the ability to learn specific climate categories. Moreover, the neural network weights used to classify climates correspond to distinct visual characteristics in temperature and precipitation. These results suggest that neural networks can model the formation of visual categories.     

Learning about learning: lessons from public engagement and deliberation on urban river restoration

This paper provides a new discussion of how people learn through deliberative processes, drawing upon empirical analysis of a novel public engagement process for urban river restoration. Such critical evaluation is rare and yet will be crucial to both theoretical development and learning about engagement practice, not least in a policy area subject to strong regulatory drivers for public participation. The analysis supports two important learning mechanisms – the use of 'gatekeepers' of knowledge, interests and values, and the privileging of narrative. It provides new evidence of instrumental and communicative learning about shared priorities and criteria for effective river restoration that evolved through the deliberative process and directly informed the restoration scheme. It is important to question whether and how such site or context-specific learning might inform other restoration schemes. Finally, the paper questions the often ignored issue of expert learning, not least the issue of the link between individual and organizational learning.     

Deriving rules from activity diary data: A learning algorithm and results of computer experiments

 Activity-based models consider travel as a derived demand from the activities households need to conduct in space and time. Over the last 15 years, computational or rule-based models of activity scheduling have gained increasing interest in time-geography and transportation research. This paper argues that a lack of techniques for deriving rules from empirical data hinders the further development of rule-based systems in this area. To overcome this problem, this paper develops and tests an algorithm for inductively deriving rules from activity-diary data. The decision table formalism is used to exhaustively represent the theoretically possible decision rules that individuals may use in sequencing a given set of activities. Actual activity patterns of individuals are supplied to the system as examples. In an incremental learning process, the system progressively improves on the selection of rules used for reproducing the examples. Computer experiments based on simulated data are performed to fine-tune rule selection and rule value update functions. The results suggest that the system is effective and fairly robust for parameter settings. It is concluded, therefore, that the proposed approach opens up possibilities to derive empirically tested rule-based models of activity scheduling. Follow-up research will be concerned with testing the system on empirical data. Received: 31 January 2001 / Accepted: 13 September 2001     

利用一种新的神经网络模型识别点状地图符号

着重讨论了用一种新的神经网络模型识别点状地图符号的过程，主要包括网络的结构特点和学习算法以及学习训练过程，并验证了用该网络进行点状地图符号识别的有效性。     

A minimalistic approach for evapotranspiration estimation using the Prophet model

ABSTRACT

This study aimed to evaluate the potential of the recently introduced Prophet model for estimating reference evapotranspiration (ETo). A comparative study was conducted for benchmarking the model results with support vector regression (SVR) and temperature-based empirical models (Thornthwaite and Hargreaves) in southern Japan. The performance of the Prophet, SVR and temperature-based empirical models was evaluated by Nash–Sutcliffe efficiency (NSE) and coefficient of determination (R²). The results indicate that temperature-based Prophet and SVR models have greater accuracy than the empirical models. The Prophet model with sole input of relative humidity, sunshine hours or windspeed showed acceptable accuracy (NSE > 0.80; R² > 0.80), while SVR models with similar inputs showed greater errors. Accuracy improved with increasing number of input parameters, giving excellent performance (NSE > 0.95; R² > 0.95) with all input parameters. Hence, the Prophet model is a new promising approach for modelling ETo with limited input variables.     

星载合成孔径雷达海洋遥感与大数据

星载合成孔径雷达以其全天候、全天时、不受云雨影响的工作特性在空间对海观测中起到了重要作用，又以其高空间分辨率、多极化、多成像模式的特点展示了其在海洋动力要素反演和海洋多尺度动力过程研究中独特的魅力.起步于20世纪70年代末的星载合成孔径雷达技术，迎来了发展的"黄金时期"，大数据和机器学习又赋予了星载合成孔径雷达海洋遥感更强大的生命力.本文首先阐述了星载合成孔径雷达大数据的5"V"特性，进而以高分辨率海面风场反演、海洋内波中尺度动力过程观测两类典型案例，阐述了大数据、机器学习等现代信息科学技术与卫星海洋遥感结合，实现海洋环境参数高精度反演和海洋动力过程科学深层次认知的研究.最后，展望了星载合成孔径雷达海洋遥感与大数据的发展前景.     

Multiobjective analysis of chaotic dynamic systems with sparse learning machines

Sparse learning machines provide a viable framework for modeling chaotic time-series systems. A powerful state-space reconstruction methodology using both support vector machines (SVM) and relevance vector machines (RVM) within a multiobjective optimization framework is presented in this paper. The utility and practicality of the proposed approaches have been demonstrated on the time series of the Great Salt Lake (GSL) biweekly volumes from 1848 to 2004. A comparison of the two methods is made based on their predictive power and robustness. The reconstruction of the dynamics of the Great Salt Lake volume time series is attained using the most relevant feature subset of the training data. In this paper, efforts are also made to assess the uncertainty and robustness of the machines in learning and forecasting as a function of model structure, model parameters, and bootstrapping samples. The resulting model will normally have a structure, including parameterization, that suits the information content of the available data, and can be used to develop time series forecasts for multiple lead times ranging from two weeks to several months.