Downbursts in the Jianghuai region of East China are primarily caused by two storm types: isolated storms (ISs) and linear-embedded storms (LSs). However, systematic differences in their synoptic backgrounds, convective environments, and downburst formation mechanisms remain poorly understood, limiting the accuracy of downburst warnings. This study aims to quantify these differences and identify storm-type-specific triggering mechanisms for associated downbursts. Based on 18 downburst events (9 ISs and 9 LSs) from 2011 to 2024, we performed composite analysis using ERA5 reanalysis data (0.25°, hourly) and examined two representative downburst cases with S-band Doppler radar, automatic weather station (AWS; minute-resolution), and sounding data. Statistical significance of environmental differences was tested with t-tests. The results are as follows. (1) LSs form ahead of upper-level troughs near the 850-hPa front with strong synoptic forcing, whereas ISs develop along the subtropical high edge under weak forcing but with higher convective available potential energy (CAPE: ~3000 vs. ~2000 J kg⁻¹). LSs exhibit significantly stronger low-to-mid-level vertical wind shear (surface to 500 hPa: ~18 vs. ~7 m s⁻¹; 0–1 km: ~21 vs. ~6 m s⁻¹; p < 0.05). (2) For ISs, downburst initiation is dominated by negative buoyancy from hail melting and hydrometeor evaporation, driving a mid-level rear-inflow jet (RIJ) and a rapidly descending weak-echo gap. For LSs, a low-level RIJ, intensified by the surface cold pool and a developing mesocyclone (which generates a downward perturbation pressure gradient force), combines with hydrometeor drag to produce the downburst. (3) Surface observations show that ISs cause a larger pre-downburst temperature drop (mean 5.1 vs. 4.6°C in LSs) and a distinct dewpoint drop (3.7 vs. 1.0°C) within 10 min before the gust. In general, the two storm types develop under fundamentally different environmental controls (instability-dominated vs. shear-dominated) and rely on different primary mechanisms (thermodynamic cooling vs. dynamic forcing from RIJ and mesocyclone). These findings provide specific pre-downburst signatures (e.g., rapid temperature/dewpoint drops within 5–10 min ahead of peak winds) that can support storm-type-specific nowcasting of disastrous downbursts in the Jianghuai region of East China.

A systematic frontal rainfall (FR) event and a localized warm-sector rainfall (WR) event during the warm season in the Yangtze-Huai river basin (YHRB) in East China are selected as research subjects. Seven convective-scale ensemble forecast experiments involving the initial conditions (ICs), lateral boundary conditions (LBCs), and model (MO) perturbations have been conducted evaluate the performance of different ensemble experiments and investigate forecast uncertainties of the FR and WR, respectively. The results indicate that combined perturbation experiments produce greater precipitation dispersion than single perturbation experiments, and the introduction of MO perturbations can effectively modify the deviation of precipitation, especially in the WR scenario. The forecast uncertainties of FR primarily stem from the synoptic low-level jet (LLJ) and the convergence of cold and warm air masses from north and south. The three-dimensional structure of the LLJ such as its intensity, location and height determines the position and strength of the FR. It is important to note that introducing MO perturbations enhances the ability of ensemble simulation to represent the uncertainty in forecasting the convergence location of cold and warm air masses. In contrast, the forecast uncertainties of WR are mainly due to boundary layer dynamics and local wind convergence near the leeward sides of mountains, and the model physics configurations are sensitive to thermal and dynamic fields of boundary layer. Appropriate MO perturbations can more effectively represent the forecast uncertainty associated with localized WR process, and thereby improve the overall performance of ensemble forecasts.

A deep learning-based precipitation nowcast model (PhyDNet-RP) over Jiangsu area is established using a physically-constrained convolutional neural network PhyDNet with both radar reflectivity and ground precipitation observations as inputs. PhyDNet-RP is then compared with INCA (blend TREC with mesoscale numerical model), PhyDNet-P (only precipitation is used as input) and UNet-RP (both precipitation and radar data are used as input for convolutional neural network UNet), especially under convection enhancement scenarios. Results are as follows: (1) The deep learning-based nowcast model performs better than INCA on the forecast of precipitation growth and decay. (2) The comparison of PhyDNet-P and PhyDNet-RP reveals that adding radar reflectivity as a second input factor can improve the forecast of movement and form of strong precipitation. (3) UNet-RP can to some extent reflect the movement and form of precipitation, but it fails to capture the intensity. It is concluded that PhyDNet-RP shows an overall advantage over INCA, PhyDNet-P, and UNet-RP, indicating that precipitation nowcast obviously benefits from employing different factors as input for deep learning model.