Quantifying the analysis uncertainty for nowcasting application

Abstract

This study proposes a method to quantify uncertainty represented by errors in very-high-resolution near-surface analysis, specifically for weather nowcasting applications. Gaussian distributed perturbations are used to perturb the first guess and observation with a variance equal to that of the first-guess error. This error reflects the spatial characteristics of the difference between the first guess and observations and dominates the primary sources of analysis uncertainty. However, mapping perturbations to analyse the grid mesh through interpolation results in underdispersion, particularly in areas without stations. To address this issue, Gaussian perturbations are inflated with an inflation factor to amplify the dispersion. This method was applied to high-resolution analysis and nowcasting for hourly temperature, humidity, and wind components in the Beijing–Tianjin–Hebei region to assess its effectiveness in representing uncertainty. The generated ensemble analysis exhibits reasonable spread and high reliability, indicating accurate quantification of analysis uncertainty. Ensemble nowcasting is extrapolated from ensemble analysis to evaluate the transmission of perturbation during extrapolation. Verification results of ensemble nowcasting reflect the fact that the spread increases effectively during extrapolation up to a lead time of 6 h. However, the increase in the spread is highly dependent on the persistence of numerical weather prediction. The results demonstrate that generating appropriate perturbations based on analysis errors effectively represents the analysis uncertainty and contributes to estimating uncertainty in nowcasting.