The Effect of Various Observations on the Forecast Accuracy of a Numerical Model Based on OSE

A meso--scale precipitation system can cause heavy rain in minutes and give damage to people, agriculture and property. Accurate quantitative precipitation forecast can minimize such damage. However, heavy rain prediction is difficult because systems can change precipitation intensity and its location quickly and have a small radius of 20 to 30 km with short duration of several minutes to hours. To improve the prediction accuracy of heavy rainfall, data assimilation (DA) is essential. In previous Observation System Experience (OSE) studies, mainly synoptic observations were assimilated into a model with a resolution of 100 km or more to analyze the effect of each observation type on the predictive accuracy of the numerical model. However, in order to predict localized torrential rain, it is necessary to analyze how synoptic and asynoptic observation data can affect the accuracy of high-resolution numerical model forecast. In this study, OSE was performed to analyze the effect of automatic weather station (AWS), buoy, radiosonde, radar and wind-profiler data on accurately predicting torrential rainfall for each observation type. A meso--scale precipitation case was simulated by composing an experiment in which data assimilation was not performed (CTRL), an experiment in which all observed types were assimilated (DA All), and experiments in which each observed type was assimilated one by one. Statistical results of Root Mean Square (RMS) error using observed precipitationand that of the model show the highest error in CTRL by 34.304 mm,and the lowest in experiment using radar only (DA radar) by 32.173 mm. The result shows that radar contributed the most in predicting heavy rain. Radar provides an environment to simulate torrential rain by observing the three-dimensional distribution, intensity, and movement of precipitation with high resolution. As the model resolution increases, the assimilation of high-resolution observations in time and space is essential for short-term weather forecasting and natural disaster prevention. Acknowledgments: This work was supported by the National Research Foundation of Korea (NRF) grant funded by the Korea government(MSIT) (No. 2019R1F1A1058620 and 2021R1A4A1032646).