Research On Technology And Application Of Dynamic Fusion Of Multi-dimensional Meteorological Observation Data

With the rapid development of detection science and technology and the increasi ng modernization level of meteorological field,meteorological data presents the chara cteristics of diversified sources and rapid increase of data volume.The rapid storage a nd fusion analysis of meteorological data has become an important research direction in the field of computer.At present,there are two urgent problems to be solved in the storage and analysis of meteorological data.First,meteorological data from various s ources,such as ground observation,radar inversion and satellite observation,need to be integrated and stored.Second,due to the number of observation equipment,precisi on and relief and so on factors,lack of meteorological data from different sources exi st,abnormal,different precision and data storage type is not unified,to different sourc es of data fusion processing,the formation of precipitation and other meteorological f orecast demand of high precision data set.Qilian mountain terrain is complex,the im portant factors that affect the ecological environment of surrounding area,how will th e qilian mountain area ground observation data,satellite data,radar data storage and f usion for data sets,a high accuracy for solving the above two questions have certain r esearch significance,but also provide high quality data for local meteorological envir onment prediction.In view of the above questions,combined with the characteristics of the original meteorological data,this paper focuses on two directions of research work.First,desi gn meteorological data storage from multiple sources;Secondly,according to the char acteristics of multisource precipitation data,the existing fusion models and methods a re improved and optimized to achieve higher data accuracy.The detailed research con tents and results are as follows:(1)The storage methods of data in different formats of ground observation,radar inversion and satellite observation are designed.Ground observations have unfixed data formats,which are unstructured;radar inversion and satellite observation formats are relatively regular and can abstract the relationship characteristics.According to the characteristics of the above data from different sources,this paper designs a unified format conversion interface to facilitate storage.(2)According to the characteristics of large amount and complex meteorological data,the data storage of different formats is realized.The data storage system includes three subsystems:data entry subsystem,automatic data upload subsystem and central station upload subsystem.Through the design of logical model and physical model of the data storage system,the effective storage and efficient storage of data can be realized.In the design stage of logical model,several fact tables and dimension tables with different meanings are designed,and the relationship between fact tables and dimension tables is established.In the physical model design stage,the entity table is established,and the ETL technology is also used to analyze the meteorological data from different sources,so as to realize the unified management of the original meteorological data.At the same time,through a series of visual interfaces to display the functions of different subsystems,realize the real-time meteorological data and historical meteorological data query.(3)According to the characteristics of multi-source precipitation data,a precipitation fusion model based on deep learning is proposed.By analyzing the characteristics of multi-source precipitation data,adaptive multi-split regression algorithm and random forest algorithm are introduced to fuse ground observation data and satellite data,and Gaussian process regression algorithm is used to fuse ground observation data,satellite data and elevation data.The satellite precipitation data and the ground precipitation data are fused by the adaptive multi-stripe regression algorithm to improve the accuracy,and the random forest algorithm is used to fuse the satellite precipitation data and the ground observation data to eliminate the blocky traces.The Gaussian process regression algorithm of the surface fusion method and the station bias correction method fuses the three kinds of precipitation data.Through experimental comparison and analysis,it can be concluded that the effect of the Gaussian process regression algorithm based on the point-surface fusion method is higher than that of the site deviation correction method,and the fusion result of the regression algorithm increases the relative error of accuracy from 0.43 to 0.79,and the spatial resolution to 1km.At the same time,significant spatial distribution can be seen,which provides a more accurate data set for precipitation prediction.