A COMPARISON OF THE BLENDING AND CONSTRAINING METHODS TO INTRODUCE LARGE-SCALE INFORMATION INTO GRAPES MESOSCALE ANALYSIS

Abstract:

To solve the problem of mesoscale analysis error accumulation after a period of continuous cycle data assimilation (CCDA), a blending method and a constraining method are compared to introduce global analysis information into the Global/Regional Assimilation and Prediction Enhanced System mesoscale three-dimensional variational data assimilation system (GRAPES-Meso 3Dvar). Based on a spatial filter used to obtain a blended analysis, the blending method is weighted toward the T639 global analysis for scales larger than the cutoff wavelength of 1,200 km and toward the GRAPES mesoscale analysis for wavelengths below that. The constraining method considers the T639 global analysis data as an extra source of information to be added in the 3DVar cost function. The cloud-resolving GRAPES-Meso system (3 km resolution) with a 3 h analysis cycle update is chosen, and forecast experiments on an extreme precipitation event over the eastern part of China are presented. The comparison shows that the inclusion of large-scale information with both methods has a positive impact on the regional model, in which the 3 h background forecasts are slightly closer to the radiosonde observations. The results also show that both methods are effective in improving large-scale analysis while reserving the well-featured mesoscale information, leading to an enhancement in the balance and accuracy of the analysis. Subjective verification reveals that the introduction of large-scale information has a visible beneficial impact on the forecast of precipitation location and intensity. The methodologies and experiences presented in this paper could serve as a reference for ongoing efforts toward the development of multi-scale analysis in GRAPES-Meso.