In this study, the Advanced Research version of the Weather Research and Forecasting (WRF-ARW) version 4.2 (Skamarock et al., 2021), a non-hydrostatic atmospheric model, was used to conduct the 24-h convective-permitting numerical simulations from 2000 LT 27 to 2000 LT 28 July with two one-way nested domains at horizontal grid spacings of 9 and 3 km (Fig. 2a) and 51 vertical levels.
The Thompson microphysics scheme (Thompson et al., 2008), Yonsei University PBL scheme (Hong et al., 2006), rapid radiative transfer model longwave radiation scheme (Mlawer et al., 1997), Dudhia shortwave radiation scheme (Dudhia, 1989), and Noah-MP land surface scheme (Niu et al., 2011) were used for both two domains. The Kain–Fritsch cumulus scheme (Kain, 2004) was used in the outer domain and the cumulus parameterization scheme was not activated in the inner domain because deep convection can be explicitly simulated at a convection-permitting resolution (Clark et al., 2016). The land use fields used in the model were interpolated from the Moderate Resolution Imaging Spectroradiometer (MODIS) datasets.
Due to the limited deterministic predictability of the CI processes (Zhang et al., 2019) and the uncertainties in the initial conditions (Nystrom and Zhang, 2019), one single numerical simulation using the ERA5 reanalysis as the initial and lateral boundary conditions poorly captured the CI location and timing as well as near-CI environment (omitted) and as a consequence had a weak simulating ability for heavy rainfall (Fig. 7v). We therefore performed a 21-member ensemble forecast initialized with the Global Ensemble Forecast System (GEFS) data (Toth and Kalnay, 1993; Wei et al., 2008; Zhou et al., 2017) with the same model configuration mentioned above to select the best simulation result as the control simulation (CTRL).
Figure 7. Simulated accumulated precipitation (shading; mm) from 1100 to 1400 LT 28 July for the mem00–mem20 runs and ERA5, NOURBAN, and FAKEDRY simulations. Mem07 is indicated by red fonts as the control simulation (CTRL). The gray contours indicate the urban borders.
To investigate the effects and relative roles of the UHI and MFPS in the CI, two additional sensitivity experiments based on the CTRL were conducted. A sensitivity experiment called FAKEDRY with turning off the latent heating and cooling from the microphysics scheme on the north of the SMR after 0200 LT 28 July was conducted to suppress the development of the MFPS. This sensitivity experiment also was called “fake-dry” simulation and was used in Luo et al. (2020) to suppress the upstream convective system. Another sensitivity experiment, NOURBAN, in which urban land use of the SMR was replaced by cropland (Fig. 2c) was conducted to explore the relative role of the UHI in the CI. The method has also been used in many previous studies (Miao et al., 2011; Yu and Liu, 2015) to remove the urban effect. The other model settings in the two sensitivity experiments were identical to those in the CTRL. Table 1 shows detailed description of all numerical experiments.
Experiment Model configuration Aim of the experiment Mem00–mem20 A 21-member ensemble simulation using the GEFS data as the initial and lateral boundary conditions To reproduce the CI process and near-CI environment CTRL The control simulation selected from the mem00–mem20 runs To explore the CI mechanism NOURBAN The urban area of the SMR is removed and replaced by cropland To explore the effects of the UHI FAKEDRY The latent heating and cooling from the microphysics scheme on the north of the SMR are turned off To explore the effects of the MFPS
Table 1. Description of experiment design
Figure 7 shows the simulated 3-h accumulated precipitation from 1100 to 1400 LT 28 July for the 21-member ensemble forecast. Compared with the observed precipitation (Fig. 1a), only the seventh member (mem07) in all ensemble members was able to reproduce the rainfall belt with the maximum rainfall amount of over 50 mm in the SMR (Fig. 7h). Although the simulated maximum rainfall amount of 51 mm in the mem07 was less than the observed maximum rainfall amount of 75 mm, which was partly attributed to the weaker CI intensity (Figs. 3, 8), given that we focused on the physical processes governing the CI rather than the quantitative precipitation forecast. Thus, the mem07 was considered to be the best member of all simulations to capture the intensity and distribution of heavy rainfall in the SMR well, and we then will examine the ability of the mem07 to reproduce the CI process by comparing the simulated composite radar reflectivity with the observations.
Figure 8. Simulated composite radar reflectivity (shading; dBZ) in mem07 at (a) 0800, (b) 1000, (c) 1100, (d) 1130, (e) 1230, and (f) 1300 LT 28 July. The red cross in (a) represents the location of the simulated sounding. The black lines displayed in (d) indicate the cross-sections in Figs. 11, 13.
Figure 8 shows the evolution of the simulated composite radar reflectivity in the mem07. Compared with the observed composite radar reflectivity (Fig. 3), the mem07 successfully captured the general features of the CI and its subsequent development in the SMR, apart from the latter CI over the sea (Fig. 3e). Herein, CI was defined when and where the radar echo first exceed 35 dBZ (Weckwerth, 2000; Du et al., 2020). In the mem07, discrete convective cells with the radar echo of exceeding 35 dBZ occurred in the SMR at 1130 LT and then intensified locally into a northeast–southwest-oriented linear MCS with a contiguous band of over 40 dBZ (Figs. 8e, f). The CI location, timing, and pattern were similar to those in the observations (Figs. 3c, 8d). The process was what we focused on. In addition to the CI in the SMR, the simulation also reproduced the development of the MFPS including the stratiform precipitation that stayed on the southeast coast of Jiangsu at 0800 LT and the convective precipitation produced by the eastward-moving Meiyu front MCS (Figs. 8a–c). But the speed of movement of the Meiyu front MCS was slightly slower than that in the observations. The location of the Meiyu front MCS at 1130 LT in the simulation (Fig. 8d) was closer to the counterpart at 1100 LT in the observations (Fig. 3b).
We further examine the observed and simulated convection environment prior to the CI by comparing the sounding profile around the CI location and the near-surface condition. As shown in Fig. 5, the simulated sounding reproduced the low-level temperature inversion and lapse rate as well as the horizontal winds at 0800 LT (Fig. 5b), which closely resembled the Baoshan sounding (Fig. 5a). Despite slightly underestimating the amount of moisture below 850 hPa, the simulation reasonably captured the general humidity vertical profile where there was relatively dry from approximately 700 to 500 hPa and relatively humid from approximately 850 to 700 hPa.
Figure 9 shows the evolution of the simulated surface wind and temperature fields. The simulation reproduced the convergence zone in the SMR, which formed between the northwesterly outflows from the MFPS and the enhanced northeasterly sea winds at the northeast coast of the SMR (indicated by the purple arrows in Figs. 9c, d) as well as the southerly–southeasterly winds at the south coast of Shanghai. Consistent with the observations, the sea winds at the southeast coast of Jiangsu shifted from the southeasterly to the northwesterly (indicated by the red arrows in Fig. 9) with the intensification of the northwesterly outflows. Besides, the simulation also reproduced the main temperature pattern with warmer air in the SMR. Similar to the observations, the simulated CI occurred within the strong UHI. It occurred on the edge of the high temperature region, not in the peak temperature region because the colder external flows (including the outflows and sea winds) cooled the near-surface atmosphere around the CI location. We have also evaluated the near-CI environment in other members and SMS-WARMSv2, the northwesterly outflows produced by all the other simulation results were too weak to intrude into the SMR (omitted).
Figure 9. Simulated air temperature (shading; °C) and wind barbs (full barb = 4 m s−1) at the surface, and the contours of composite radar reflectivity exceeding 35 dBZ (magenta contours, interval at 10 dBZ) in mem07 at (a) 0800, (b) 1000, (c) 1100, and (d) 1130 LT 28 July. The wind barbs in black represent the easterly component and the wind barbs in white represent the westerly component. The red and purple arrows indicate the general directions of the sea winds.
Overall, the simulation result of the mem07 could capture the CI of interest and the near-CI environment reasonably well. Thus, we considered the mem07 as the CTRL to investigate the processes governing the CI. In the following sections, we will use the model results to address the issues raised in Section 2.
|Experiment||Model configuration||Aim of the experiment|
|Mem00–mem20||A 21-member ensemble simulation using the GEFS data as the initial and lateral boundary conditions||To reproduce the CI process and near-CI environment|
|CTRL||The control simulation selected from the mem00–mem20 runs||To explore the CI mechanism|
|NOURBAN||The urban area of the SMR is removed and replaced by cropland||To explore the effects of the UHI|
|FAKEDRY||The latent heating and cooling from the microphysics scheme on the north of the SMR are turned off||To explore the effects of the MFPS|