Research On Quality Control Methods For Ground Observation Data And Its Application In Numerical Models

Surface observation data is an importance source to obtain the accurate description of meso-and micro-scale systems and short-term convective weather phenomenon. The quality of observation data is particularly important for data assimilation in numerical weather predication (NWP) models as it impacts the data assimilation analysis directly. Based on EOF (Empirical Orthogonal Function) QC (quality control), this paper firstly compared it with Fast Fourier Transform quality control (FFT QC). Then the recursive EOF QC is proposed to improve the EOF QC. In order to understand the application effect of EOF QC in numerical models, this method is implanted into two widely used models and its three-dimensional variational data assimilation systems to confirm the potential applications. Lastly, a data thinning program is proposed to reduce the high spatial correlation of observation representativeness error. The main conclusions are as follows:(1) The traditional QC methods can effectively exclude gross errors in surface pressure, temperature and humidity. EOF can extract the temporal and spatial characteristics and FFT can capture the typical oscillations from observations. Both of the EOF QC and FFT QC are applied to surface specific humidity. The result of power spectral analysis shows that the differences of amplitude and phase between observations and background fields result in a non-Gaussian distribution in OMB (observation minus background), which violates the assumption of normalized distribution observation error in data assimilation systems. Comparing the two QC methods, it is found that the EOF QC performs better than the FFT QC, which due to that the FFT is performed in each site sequentially, while the EOF is perfomed in the space-time field, preserving the large-scal fluctuations of the original filed as much as possible. The EOF QC results of surface specific humidity, temperature and pressure display that the observations with large observation errors are excluded effectively and the statistical characteristics of observation error is closer to Gaussian distribution after QC.(2) The Z-score threshold in EOF QC is determined empirically according to the background information, which will affect the business application of EOF QC. In order to confirm the threshold objectively based on the data itself, a recursive EOF quality control (Rec-EOF QC) method is developed in this paper. The results indicate those data rejection rates, the standard deviation of the OMB, skewness and kurtosis have reached a relatively stable state after four recursions. The probability of OMB appears near 0 is imported effectively, making the data more symmetrical, and the probability density function of observed error is closer to a Gaussian distribution when the method is applied to data with different times and different kinds.(3) In order to verity the application of EOF QC in NWP models, the EOF QC program is implanted in WRF and its 3DVAR systems. Two heavy precipitation events during the periods of 28-29 January 2008 and 14-15 July 2008 have been used to investigate the impact of EOF QC. The numerical simulation test is also done for one month of July 2008. The assimilation results indicate that EOF QC can retain weather oscillations in the observations more effectively and reflect the true state of the atmosphere more objective comparing OMB QC which is the quality control method in NWP model. The quality of surface temperature is improved significantly and the prediction error is minimized when assimilating data after EOF QC. The precipitation intensity is improved and closer to the real situation. The one month test also proves that EOF QC has better forecasting capability for precipitation.(4) In order to promote the development of numerical models in our contrary and improve the technology in surface data assimilation, the EOF QC is also introduced to GRAPES and its 3DVAR systems which are exploited independently by researchers in our country. The case in January 2008 is examined in GRAPES-3DVAR. The numerical results indicate that the EOF-QC has significantly improved the quality of surface pressure, surface temperature and the precipitation falling region.(5) A thinning program is proposed to reduce the high spatial correlation of representativeness error. The typical observation within a certain range can be extracted by the thinning program. The thinning program is applied to GRAPES-3DVAR, and the influence of different thinning radius on data assimilation is researched. The numerical simulation shows that the range and level of precipitation are better and ETS scores are higher when the thinning radius is 90km. The above conclusions due to the analytical error of temperature and pressure are minimum and the gradient of the objective function changes gentler and closer to zero with the increasing iterations when comparing other thinning radius, indicating the optimal matching between model resolution and observation resolution.