An Ensemble Forecast Experiment Of Initial Disturbance Of Torrential Rain In The Warm Region Of The Middle And Lower Reaches Of The Yangtze River And Its Predictability

Warm-sector rainstorm events under weak synoptic-scale background are difficult to predict accurately due to complex initiation and development mechanism,thus convection-allowing ensemble forecasts(Ca EF)considering uncertain factors have become one of the crucial options.Through the Ca EF experiments considering initial perturbations sensitivity,this paper explores the forecast error sources,evolution characteristics and the limitation of practical and intrinsic predictability for warm-sector rainstorm over the middle-lower reaches of the Yangtze River(MLRYZ),analyzes and discusses the response of ensemble forecasts performance to the scale and amplitude of initial perturbations,in order to provide technical accumulation for the Ca EF operation.The results show that:(1)The growth of ensemble spread and the effect of probability forecast in different stages are closely related to the scale characteristics of the initial perturbations.Four Ca EF experiments with different scale information including dynamic downscaling(DOWN),breeding of the growing mode(BGM),Local BGM(LBGM),and BLEND,were carried out for a typical warm-sector rainstorm over MLRYZ in May 4-5,2018.The results find that ensemble spread of LBGM are higher than DOWN and BGM in first 0-6 h,which indicates the effective growth of small and medium scale perturbations during this stage.Meanwhile,the precipitation probability forecast generated by LBGM reduces the overestimation of small probability events effectively.However,after 12 hours of integration,the performance of DOWN is better than that of BGM and LBGM instead,which indicates that large-scale perturbations begin to play a major role after the rapid increase of initial error for a period of time.The unconditional deviation is also avoided due to the better identification ability of DOWN for high probability events during this stage.(2)BLEND contains advantages of both LBGM at early stage and DOWN after 12 h integration,and has overall optimal forecast performance.BLEND with multi-scale perturbation information maintains high spread almost during the whole short-range forecast(0-24 h)and can identify events with different probabilities.This result shows that BLEND can combine small-scale perturbations from LBGM and large-scale perturbations from DOWN effectively,helping to capture the multi-scale uncertainty in warm-sector rainstorm.(3)Bifurcation characteristics are apparent between ensemble forecasts for warm-sector rainstorm over MLRYZ,which are highly related to the boundary layer process.According to above-mentioned scale sensitivity,a multi-scale Ca EF system based on ensemble data assimilation which assimilates simulated radiosonde and radar data was constructed for a warm-sector rainstorm on July 21,2013 to further investigate its practical and intrinsic predictability.The results of Ca EF show that there are two bifurcation stages with significant differences on precipitation time and location among ensemble members.The first bifurcation is related to interaction between perturbed boundary layer wind and meso-scale topography in convection initiation stage,while the second in convection development stage is due to the difference of cold pool intensity caused by initial perturbations and strong(weak)outflow from cold pool could lead to a south(north)precipitation location.The results of ensemble sensitivity analysis also confirm that the error of boundary layer wind is a vital influence factor for the accuracy of the model prediction in the bifurcation stages.(4)Further quantitative analysis of predictability suggests a serious limitation in both practical and intrinsic predictability for warm-sector rainstorm during the bifurcation stages.The accurate initial condition could effectively improve the forecast when the practical predictability of weather process is limited by initial perturbations(errors).However,nonlinear characteristic between difference total energy(DTE)or measure of forecast challenge(MFC)time series curves in sensitivity experiments reducing the amplitude of initial perturbations reveals an additional intrinsic predictability limitation during bifurcation stages caused by chaotic nonlinear dynamics.In this situation,the failure of single deterministic forecast may still occur when improving initial condition,thus the need for ensemble(probability)forecast is especially highlighted.This result is helpful to adopt different ways to optimize the forecast performance according to the difference of predictability limitation caused by errors from different sources.