Research On Quality Control Method For Surface Meteorological Observations

Surface meteorological observation data is the foundation of understanding and predicting weather conditions, providing meteorological services and doing scientific research. The quality of data from AWS is important to meteorological workers when Automatic Weather Stations are widely used in weather forecasting and public weather services. Quality control on meteorological data is the precondition whether data assimilation and numerical weather prediction can be quantified and objectified accurately. A large number of experiments and numerical simulation research indicate that combining and assimilating ground meteorological observation data and other meteorological data can improve the level of numerical weather prediction effectively.Quality control methods on meteorological data are divided into quality control of single-station and quality control of multi-station network. According to the coupling relationship between temperature and relative humidity, a new quality control (QC) method of surface temperature data in hour based on Gene Expression Programming is introduced about quality control of single-station. This method implements the quality control of relative humidity on temperature. In order to assess the suitability of the method, surface temperature observation data of six cities from JiangSu Province is used. Test result indicates that this method can flag suspicious data effectively, as a result, it has the advantages of strong controllability and high accuracy.After completed the research on single station quality control methods, we focus on multi-station network quality control method. For multi-station network control method, the method designed in this paper consists of two steps:error finding and error revision. Firstly, Support vector machine (SVM) prediction model is built and SVM classification algorithm is used to classify correct values and suspicious values. Through experimental analysis, SVM classification algorithm can realize the requirement of looking for error and ensure error revision next step by accurate location.Error is revised by GEP-BP neural network algorithm. In GEP-BP neural network algorithm, the BP neural network is fully encoded in linear chromosomes in a fixed length, and special genetic operators are used in chromosome to make genetic operation. The individual of highest fitness value is retained, and the rest individuals continue to make genetic operation, so we can get the optimal network model finally. At the same time, select GEP-BP algorithm and traditional BP neural network algorithm to revise error of suspicious data. Through experiments, we can conclude that GEP-BP algorithm has advantages of high precision and small error volatility.This paper makes test on the operating results between quality control methods of single-station and multi-station network control methods through a great number of experiments. Experimental results indicate that the methods proposed in this paper are suitable for quality control of surface meteorological observation data, and have the advantages of good stability and high control accuracy.