Evaluation of Convective-Scale Ensemble Forecast for a Severe Precipitation Event in the Plateau Region

Using FNL （Final Reanalysis Data）， ERA5 （ECMWF Reanalysis V5） reanalysis data， and GPM （Global Precipitation Measurement） global half-hourly precipitation data， a strong precipitation event in the southwestern plateau of China was selected to study the forecasting ability of two initial perturbation methods， Breeding Growth Mode （BGM） and Local Breeding Growth Mode （LBGM）， in convective-scale ensemble forecasting of complex terrain rainfall.The MODE （Method for Object-Based Diagnostic Evaluation） method based on object diagnostics was used to evaluate the model's ability to predict the location， structure， and intensity of precipitation objects， and compared with scoring methods such as Threat Score （TS） to comprehensively assess the model's forecasting performance.The results show that： （1） The ensemble forecast systems BGM-EPS and LBGM-EPS， generated using BGM and LBGM methods to produce initial perturbations， have better ensemble mean forecast scores for precipitation of all magnitude levels at 24 hours compared to the control forecast， and LBGM-EPS has a higher TS score for heavy rainfall compared to BGM-EPS， this indicates that the LBGM method has a certain improvement effect on ensemble forecasts for heavy precipitation.However， the underlying mechanisms behind the different initial perturbation methods are worthy of further investigation； （2） Overall， the WRF model can capture precipitation objects well， especially for rainfall forecasts in complex terrain of the plateau mountains， with a better overall similarity in precipitation targets for LBGM-EPS compared to BGM-EPS， highlighting the advantage of LBGM method in representing convective-scale ensemble forecasting of intense convection.The initial perturbation total energy of BGM and LBGM shows a developing trend with forecast time.In the same forecast time， LBGM has a larger perturbation total energy than BGM， which better represents the growth of forecast error.This can partially explain why the LBGM method outperforms the BGM method in terms of precipitation object matching in the MODE evaluation； （3） Compared with traditional TS scoring and other verification methods， the MODE method can better reflect the spatial position information of precipitation forecasts， and under the same convolution radius and precipitation threshold， the ensemble mean forecast based on LBGM method performs better in identifying precipitation objects.By flexibly setting the convolution radius and determining the precipitation threshold， the WRF model can capture precipitation objects in complex terrain areas during heavy precipitation events.However， the matching degree of precipitation targets in high-altitude areas is lower than that in low-lying areas.The LBGM-EPS method outperforms the BGM-EPS method in terms of the shape of precipitation objects and the matching of precipitation areas， resulting in better identification of precipitation objects.The quality of precipitation object matching using the MODE method is related to parameter settings such as precipitation threshold and convolution radius， rather than the complex terrain background related to terrain gradients.