Quality control of surface station temperature data with non‐Gaussian observation‐minus‐background distributions

Abstract

The comparison of surface air temperatures derived from surface ground station observations and background fields reveals systematic amplitude and phase differences in the diurnal and semidiurnal oscillations, which are dominant features of surface temperature fields. It is shown that these oscillations are responsible for a non‐Gaussian distribution of observation increments, and the removal of the diurnal and semidiurnal cycles leads to a significant reduction of the non‐Gaussian component in “observations‐minus‐background” (OMB) vectors, i.e., the sum of background error and observation error. A quality control (QC) method applicable for temperature observations from surface ground stations is developed that incorporates the removal of the diurnal and semidiurnal components into the standard OMB QC procedure that requires a Gaussian distribution. It is illustrated by application of one month of surface ground station data. Numerical results show that the removal of the diurnal and semidiurnal cycles using an empirical orthogonal function analysis outperforms the spectral method in that the former method avoids rejection of correct observations in situations where the background fields cannot capture small‐scale or initiation of some weather events, while other outliers (about 0.64%) are effectively identified and removed from data assimilation.