The Quality Control For Surface Temperature Observation Based On Intelligent Algorithm

The development of Numerical Weather Prediction(NWP)is a high requirement for the assimilation technology of surface observations,and the primary task of assimilation technique is to control the surface observations effectively.Due to the complex terrain and the asymmetry of the distribution of observation stations in our country,only a few of the surface observations enter into the assimilation system,which lead to the conventional Quality Control(QC)method can not meet the needs of meteorological services.In view of this,the paper starting from the reality of distribution of surface observations of china,basing on the time-space distribution characteristics of temperature,considering two aspects of the density of observation station and the difference of observation environment,studies on the corresponding single-station,multi-station and complex QC,and analyzes the practical application.The main work is as follows:Analysis of spatial and temporal distribution characteristics of temperature with Detrended Fluctuation Analysis,Chaos Identification and Moran 'I,which provides the basis for quality control research;A single station quality control method(PSO-PSR-ICA-ELM_QC)for Phase Space Reconstruction(PSR)and Extreme Learning Machine(ELM)is structured based on temporal correlation,and the parameter selection and model over fitting are optimized with Particle Swarm optimization(PSO)and Independent Component Analysis(ICA);Multi-station network quality control method(ID_PCA-MEF_QC)with Multi-quadric Equations Fitting(MEF)interpolation is introduced based on spatial correlation,and the optimization strategy is proposed for feature reference station selection and model generalization with Principal Component Analysis(PCA)and Particle Swarm optimization(PSO);A complex quality control method(ST_OS-ELM_QC),integrated the spatiotemporal information,is proposed based on spatiotemporal neural network of Online Sequential Extreme Learning Machine(OS ELM).Case studies of different methods show,PSO-PSR-ICA-ELM_QC can solve the problem over areas where the station density is low,or for some stations that have no adjacent stations and lack effective internal reference data,such as those newly deployed stations,and have high detecting ratio ability compared to traditional single station quality control method;ID_PCA-MEF_QC performs more stable in the ability of detecting ratio comparing with SRT and IDW in areas of multi-station network observation,but its adaptability under the condition of large difference of observation environment,such as complex terrain and changeable climate,is limited to a certain extent;ST_OS-ELM_QC can handle the question of quality control under the condition of large difference of observation environment,which have a good capacity and strong robustness to distinguish singular values,and can avoid the influence of complex weather on the uncertainty of quality control algorithm to some extent.