Comparative Experiments on the Effect of Radar Data Assimilation and Increasing Horizontal Resolution on Short-term Numerical Weather Prediction

To examine the effect of radar data assimilation and increasing horizontal resolution on the short-term numerical weather prediction, comparative numerical experiments are conducted for a Huabei （North China） torrential rainfall event by using the Advanced Regional Prediction System （ARPS） and ARPS Data Anal- ysis System （ADAS）. The experiments use five different horizontal grid spacings, i.e., 18, 15, 9, 6, and 3 km,respectively, under the two different types of analyses： one with radar data, the other without. Results show that, when radar data are not used in the analysis （i.e., only using the conventional observation data）, increasing horizontal resolution can improve the short-term prediction of 6 h with better representation of the frontal structure and higher scores of the rainfall prediction, particularly for heavy rain situations. When radar data are assimilated, it significantly improves the rainfall prediction for the first 6 h, especially the locality and intensity of precipitation. Moreover, using radar data in the analysis is more effective in improving the short-term prediction than increasing horizontal resolution of the model alone, which is demonstrated by the fact that by using radar data in the analysis and a coarser resolution of the 18-km grid spacing, the predicted results are as good as that by using a higher resolution of the 3-km grid spacing without radar data. Further study of the results under the radar data assimilation with grid spacing of 18-3 km reveals that the rainfall prediction is more sensitive to the grid spacing in heavy rain situations （more than 40 mm） than in ordinary rain situations （less than 40 mm）. When the horizontal grid spacing reduces from 6 to 3 km, there is no obvious improvement to the prediction results. This suggests that there is a limit to how far increasing horizontal resolution can do for the improvement of the prediction. Therefore, an effective approach to improve the short-term numerical prediction is to combine the radar data assimilation with an optimal horizontal resolution.