Abstract: Using conventional and encrypted observation data, radar data and NCEP re-analysis data, the mid-scale characteristics and the forecasting ability of different numerical prediction models in the course of an agricultural rainstorm were analyzed in the summer of 2018 in Hunan province.The results showed that the rainstorm process was composed frontal rainfall and warm zone rainfall, and there was no obvious rapids in both forms. More than 40 dBZ of strong precipitation echo band position was stable, forming obvious train effect, more than 50 dBZ of multiple strong echo monomers were constantly evolving, low-mass precipitation echo, middle-low-layer warm upper-cold atmospheric instability layer, higher VIL were favorable conditions for heavy precipitation. The mesoscale filtering experiment could better separate the mesoscale weather system, and different schemes had different effects. For frontal rainfall, both the large-scale model and the mesoscale model had better prediction effect on precipitation falling area, while the mesoscale model was more accurate to simulate precipitation intensity. For local heavy rainfall in warm zone, the large scale model forecast deviation was large and unstable, and the mesoscale model could capture some information, which could be referred to weather forecast.

Key words: heavy rain disaster, mesoscale analysis, radar echo, Barnes filtering, model predictability