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。      

A Deep Learning Method for Bias Correction of ECMWF 24–240 h Forecasts

Correcting the forecast bias of numerical weather prediction models is important for severe weather warnings. The refined grid forecast requires direct correction on gridded forecast products, as opposed to correcting forecast data only at individual weather stations. In this study, a deep learning method called CU-net is proposed to correct the gridded forecasts of four weather variables from the European Centre for Medium-Range Weather Forecast Integrated Forecasting System global model(ECMWF-IFS): 2-m temperature, 2-m relative humidity, 10-m wind speed, and 10-m wind direction, with a forecast lead time of 24 h to 240 h in North China. First, the forecast correction problem is transformed into an image-toimage translation problem in deep learning under the CU-net architecture, which is based on convolutional neural networks.Second, the ECMWF-IFS forecasts and ECMWF reanalysis data(ERA5) from 2005 to 2018 are used as training,validation, and testing datasets. The predictors and labels(ground truth) of the model are created using the ECMWF-IFS and ERA5, respectively. Finally, the correction performance of CU-net is compared with a conventional method, anomaly numerical correction with observations(ANO). Results show that forecasts from CU-net have lower root mean square error, bias, mean absolute error, and higher correlation coefficient than those from ANO for all forecast lead times from 24 h to 240 h. CU-net improves upon the ECMWF-IFS forecast for all four weather variables in terms of the above evaluation metrics, whereas ANO improves upon ECMWF-IFS performance only for 2-m temperature and relative humidity. For the correction of the 10-m wind direction forecast, which is often difficult to achieve, CU-net also improves the correction performance.     

Linking model design and application for transdisciplinary approaches in social-ecological systems

As global environmental change continues to accelerate and intensify, science and society are turning to transdisciplinary approaches to facilitate transitions to sustainability. Modeling is increasingly used as a technological tool to improve our understanding of social-ecological systems (SES), encourage collaboration and learning, and facilitate decision-making. This study improves our understanding of how SES models are designed and applied to address the rising challenges of global environmental change, using mountains as a representative system. We analyzed 74 peer-reviewed papers describing dynamic models of mountain SES, evaluating them according to characteristics such as the model purpose, data and model type, level of stakeholder involvement, and spatial extent/resolution. Slightly more than half the models in our analysis were participatory, yet only 21.6% of papers demonstrated any direct outreach to decision makers. We found that SES models tend to under-represent social datasets, with ethnographic data rarely incorporated. Modeling efforts in conditions of higher stakeholder diversity tend to have higher rates of decision support compared to situations where stakeholder diversity is absent or not addressed. We discuss our results through the lens of appropriate technology, drawing on the concepts of boundary objects and scalar devices from Science and Technology Studies. We propose four guiding principles to facilitate the development of SES models as appropriate technology for transdisciplinary applications: (1) increase diversity of stakeholders in SES model design and application for improved collaboration; (2) balance power dynamics among stakeholders by incorporating diverse knowledge and data types; (3) promote flexibility in model design; and (4) bridge gaps in decision support, learning, and communication. Creating SES models that are appropriate technology for transdisciplinary applications will require advanced planning, increased funding for and attention to the role of diverse data and knowledge, and stronger partnerships across disciplinary divides. Highly contextualized participatory modeling that embraces diversity in both data and actors appears poised to make strong contributions to the world’s most pressing environmental challenges.     

基于卷积神经网络的京津冀地区高分辨率格点预报偏差订正试验

为了进一步提高RISE系统高分辨率网格化预报产品的准确率,同时考虑到深度学习近年来在地学领域的有效应用,采用2019—2021年高分辨率RISE系统数据,设计出卷积神经网络模型Rise-Unet,实现了未来4～12 h地面2 m温度、2 m相对湿度、10 m-U风速以及10 m-V风速预报结果的订正。订正试验结果表明,采用均方根误差和平均绝对误差作为评分标准,与RISE原始预报结果相比,基于Rise-Unet模型可以有效提高温湿风预报结果的准确率。该基于深度学习的Rise-Unet偏差订正技术可应用于RISE系统的后处理模块,对提升RISE系统百米级分辨率或其他高分辨率模式系统格点预报水平具有重要的科学意义和应用价值。     

Quantitative Precipitation Forecast Experiment Based on Basic NWP Variables Using Deep Learning

The quantitative precipitation forecast (QPF) performance by numerical weather prediction (NWP) methods depends fundamentally on the adopted physical parameterization schemes (PS). However, due to the complexity of the physical mechanisms of precipitation processes, the uncertainties of PSs result in a lower QPF performance than their prediction of the basic meteorological variables such as air temperature, wind, geopotential height, and humidity. This study proposes a deep learning model named QPFNet, which uses basic meteorological variables in the ERA5 dataset by fitting a non-linear mapping relationship between the basic variables and precipitation. Basic variables forecasted by the highest-resolution model (HRES) of the European Centre for Medium-Range Weather Forecasts (ECMWF) were fed into QPFNet to forecast precipitation. Evaluation results show that QPFNet achieved better QPF performance than ECMWF HRES itself. The threat score for 3-h accumulated precipitation with depths of 0.1, 3, 10, and 20 mm increased by 19.7%, 15.2%, 43.2%, and 87.1%, respectively, indicating the proposed performance QPFNet improved with increasing levels of precipitation. The sensitivities of these meteorological variables for QPF in different pressure layers were analyzed based on the output of the QPFNet, and its performance limitations are also discussed. Using DL to extract features from basic meteorological variables can provide an important reference for QPF, and avoid some uncertainties of PSs.     

Turning Contention into Collaboration: Engaging Power,Trust, and Learning in Collaborative Networks

Given the complexity and multiplicity of goals in natural resource governance, it is not surprising that policy debates are often characterized by contention and competition. Yet at times adversaries join together to collaborate to find creative solutions not easily achieved in polarizing forums. We employed qualitative interviews and a quantitative network analysis to investigate a collaborative network that formed to develop a resolution to a challenging natural resource management problem, the conservation of vernal pools. We found that power had become distributed among members, trust had formed across core interests, and social learning had resulted in shared understanding and joint solutions. Furthermore, institutions such as who and when new members joined, norms of inclusion and openness, and the use of small working groups helped create the observed patterns of power, trust, and learning.     

支持全同态密文计算的访问控制加密方案

提出支持全同态密文计算的访问控制加密（FH-ACE）方案,并给出基于带错学习（Learning with Error）困难性问题的具体构造.首先,根据全同态加密（Fully Homomorphic Encryption）概念和访问控制加密（Access Control Encryption）概念,给出支持全同态密文计算的访问控制加密方案的定义以及需要满足的安全模型;其次,提出以满足特定条件的全同态加密方案为基本模块的黑盒构造,并分析基于目前的全同态加密方案,具体构造所面临的困难点以及解决方法;最后,基于带错学习困难性问题,给出支持全同态密文计算的访问控制加密方案的具体构造.     

人工智能技术发展概述

近年来,以深度学习为核心的人工智能技术,取得了一系列重大突破.本文将就人工智能的产业化热潮,主要研究流派及发展历史,以深度学习为核心的成功应用,以及存在的一些问题和今后的可能研究方向做一个介绍.     

An intelligent spatial land use planning support system using socially rational agents

This research presents an intelligent planning support system based on multi-agent systems for spatial urban land use planning. The proposed system consists of two main phases: a pre-negotiation phase and an automated negotiation phase. The pre-negotiation phase involves interaction between human actors and intelligent software agents in order to elicit the actors’ social preferences. The agents employ social value orientation theory, which is rooted in social psychology, in order to model actors’ social preferences. The automated negotiation phase involves negotiation among autonomous software agents, the aim being to achieve consensus about the spatial problem on behalf of the relevant actors and using the information obtained.

This study employs a computationally effective Bayesian learning technique, along with social value orientation theory, to design socially rational intelligent agents who work on behalf of real actors. The proposed system is applied to a real world urban land use planning case study. Human actors participate in a pre-negotiation phase, and their social preferences are elicited by intelligent software agents through a number of interactions. Then, software agents come together to engage in an automated negotiation phase and eventually reach an agreement on the spatial configuration of urban land uses on behalf of the actors. The results of the study show that the proposed system is effective at performing an automated negotiation, plus that the final plan – which is the output of the automated negotiation – produces higher social utility and better spatial land use configurations for the agents.     

基于迁移学习的卫星云图云分类

针对在基于机器学习的云图识别中,由于不存在公认的云分类样本库的现实条件下,带来的训练样本数量不足和不平衡,从而难以获得可靠的分类模型的问题,利用迁移学习中的多源加权Tradaboost算法(内部采用极限学习机作为分类器)来进行卫星云图云的检测。利用多人(多源)标注的大量厚云的样本,构成多源辅助样本集;利用少量标注的薄云样板构成目标样本集。使用迁移学习和辅助样本集,对仅在薄云样本集下的训练获得的极限学习机分类器进行辅助训练,提高其薄云识别率。基于国家卫星气象中心的HJ-1A/B的卫星数据实验结果表明,迁移学习可以充分利用容易获得的大样本厚云辅助样本知识,对同类型有关联的小样本薄云分类器进行识别提高。实验表明,迁移学习算法可以进一步用于更多多源样本和其他云分类的任务。