Statistical properties of observed residuals from the Mesoscale Analysis and Prediction System (MAPS), a real-time data assimilation system, were investigated. Observed residuals are defined as differences between rawinsonde observations interpolated vertically to the model levels and the predicted values from MAPS interpolated horizontally to the radiosonde locations. One-point statistical moments up to order 4 (including skewness and flatness) were computed to investigate the normality of the probability distribution of observed residuals. The finding of near-zero skewness indicates symmetry in the distribution of observed residuals, but values of flatness significantly different from 3 indicate deviations from a normal (Gaussian) distribution. These results are supported by an effective statistical test. The spatial distributions of these statistical moments show strong local variability, which is ascribed to occasional gross errors in the rawinsonde data. The spatial correlation of observed r... Abstract Statistical properties of observed residuals from the Mesoscale Analysis and Prediction System (MAPS), a real-time data assimilation system, were investigated. Observed residuals are defined as differences between rawinsonde observations interpolated vertically to the model levels and the predicted values from MAPS interpolated horizontally to the radiosonde locations. One-point statistical moments up to order 4 (including skewness and flatness) were computed to investigate the normality of the probability distribution of observed residuals. The finding of near-zero skewness indicates symmetry in the distribution of observed residuals, but values of flatness significantly different from 3 indicate deviations from a normal (Gaussian) distribution. These results are supported by an effective statistical test. The spatial distributions of these statistical moments show strong local variability, which is ascribed to occasional gross errors in the rawinsonde data. The spatial correlation of observed r...