A statistical emulator for multivariate model outputs with missing values

作     者：Finazzi, Francesco Napier, Yoana Scott, Marian Hills, Alan Cameletti, Michela 

作者机构：Univ Bergamo Dept Management Informat & Prod Engn Viale Marconi 5 I-24044 Dalmine BG Italy Univ Glasgow Sch Math & Stat Univ Pl Glasgow G12 8TA Lanark Scotland Scottish Environm Protect Agcy Angus Smith Bldg 6 Parklands Ave Holytown ML1 4WQ North Lanarkshi Scotland Univ Bergamo Dept Management Econ & Quantitat Methods Via Caniana 2 I-24127 Bergamo Italy 

出 版 物：《ATMOSPHERIC ENVIRONMENT》 (大气环境)

年 卷 期：2019年第199卷

页      面：415-422页

核心收录：

学科分类：0830[工学-环境科学与工程（可授工学、理学、农学学位）] 08[工学] 0706[理学-大气科学] 

基　　金：PRIN2015 - MIUR (Italian Ministry of Education, University and Scientific Research) PRIN EphaStat Project funding by the Italian Ministry for Education, University and Research [20154X8K23] NERC [NE/N008367/1] 

主　　题：EM algorithm Sensitivity analysis D-STEM software Nitrogen dioxide 

摘      要：Statistical emulators are used to approximate the output of complex physical models when their computational burden limits any sensitivity and uncertainty analysis of model output to variation in the model inputs. In this paper, we introduce a flexible emulator which is able to handle multivariate model outputs and missing values. The emulator is based on a spatial model and the D-STEM software, which is extended to include emulator fitting using the EM algorithm. The missing values handling capabilities of the emulator are exploited to keep the number of model output realisations as low as possible when the computing burden of each realisation is high. As a case study, we emulate the output of the Atmospheric Dispersion Modelling System (ADMS) used by the Scottish Environment Protection Agency (SEPA) to model the air quality of the city of Aberdeen (UK). With the emulator, we study the city air quality under a discrete set of realisations and identify conditions under which, with a given probability, the 40 mu g m(-3) yearly average concentration limit for NO2 of EU legislation is not exceeded at the locations of the city monitoring stations. The effect of missing values on the emulator estimation and probability of exceedances are studied by means of simulations.