The non-hydrostatic atmospheric Model for Prediction Across Scales (MPAS-A), a global variable-resolution modeling framework, is applied at a range of resolutions from hydrostatic (60, 30, 16 km) to non-hydrostatic (4 km) scales using regional refinement over East Asia to simulate an extreme precipitation event. The event is triggered by a typical wind shear in the lower layer of the Meiyu front in East China on 25–27 June 2012 during the East Asian summer monsoon season. The simulations are evaluated using ground observations and reanalysis data. The simulated distribution and intensity of precipitation are analyzed to investigate the sensitivity to model configuration, resolution, and physics parameterizations. In general, simulations using global uniform-resolution and variable-resolution meshes share similar characteristics of precipitation and wind in the refined region with comparable horizontal resolution. Further experiments at multiple resolutions reveal the significant impacts of horizontal resolution on simulating the distribution and intensity of precipitation and updrafts. More specifically, simulations at coarser resolutions shift the zonal distribution of the rain belt and produce weaker heavy precipitation centers that are misplaced relative to the observed locations. In comparison, simulations employing 4 km cell spacing produce more realistic features of precipitation and wind. The difference among experiments in modeling rain belt features is mainly due to the difference in simulated wind shear formation and evolution during this event. Sensitivity experiments show that cloud microphysics have significant effects on modeling precipitation at non-hydrostatic scales, but their impacts are relatively small compared to that of convective parameterizations for simulations at hydrostatic scales. This study provides the first evidence supporting the use of convection-permitting global variable-resolution simulations for studying and improving forecasting of extreme precipitation over East China and motivates the need for a more systematic study of heavy precipitation events and the impacts of physics parameterizations and topography in the future.
The key points are as follows.
Model for Prediction Across Scales (MPAS) simulations at global uniform and variable resolutions share similar characteristics of precipitation and wind in the refined region. Numerical experiments reveal significant impacts of resolution on
simulating the distribution and intensity of precipitation and updrafts. This study provides evidence supporting the use of convection-permitting
global variable-resolution simulation to study extreme precipitation.
Extreme precipitation receives great attention because of its potential for generating floods, landslides, and other hazardous conditions. East China, occupied by more than 70 % of the total population of China, is one of the areas with the most frequent, intense, and extreme precipitation around the world (Zhai et al., 2005; Z. Li et al., 2016). The socioeconomic development in regions such as the Yangtze River Delta (YRD) in East China is remarkably vulnerable to extreme precipitation, making accurate forecasts of extreme precipitation of great importance. The spatiotemporal variations of extreme precipitation over East China and their possible causes and underlying mechanisms have been investigated in many previous studies using observations and models (e.g., Ding et al., 2008; Zhang and Zhai, 2011; Li et al., 2013; W. Li et al., 2016; Q. Zhang et al., 2011, 2017; Hui et al., 2015; Liu et al., 2015; Lin and Wang, 2016; Zhao et al., 2016; Zheng et al., 2016). Zhang et al. (2017) established a relationship between the western North Pacific subtropical high (WNPSH) and precipitation over East China and explored the underlying processes. Liu et al. (2015) analyzed data from meteorological stations in East China and found significant increases in heavy precipitation at both rural and urban stations during 1955–2011. This enhanced precipitation intensity in East China has been partly attributed to localized daytime precipitation events (Guo et al., 2017). Recently, a regional climate model was used to simulate the regional climate extremes of China and noted large sensitivity of the simulated summer heavy precipitation over East China to the choice of cumulus parameterizations (Hui et al., 2015).
Numerical modeling is an important tool for understanding the underlying mechanisms of extreme precipitation and predicting the precipitation characteristics that contribute to environmental impacts. Although precipitation modeling has improved in the last decades, accurate prediction of extreme precipitation remains challenging because of the multiscale nonlinear interactions of processes that generate heavy rainfall (Fritsch and Carbone, 2004; L. Zhang et al., 2011; Sukovich et al., 2014). Although not a panacea for weather and climate modeling (NRC, 2012), previous studies suggested that increasing grid resolution could significantly improve the modeling of extreme precipitation because the impacts of topography, land use, land–atmosphere interaction, and other important processes are better resolved (e.g., Giorgi and Mearns, 1991; Giorgi and Marinucci, 1996; Leung and Qian, 2003; Bacmeister et al., 2014; ECMWF, 2016). With advances in computing and numerical modeling, convection-permitting modeling offers even more hope for reducing biases in simulating precipitation as convection and the strong vertical motion that are key to generating extreme precipitation are more explicitly resolved (Pedersen and Winther, 2005; Déqué et al., 2007; Gao et al., 2017; Yang et al., 2017; Prein et al., 2015, 2017). Previous studies suggested that convection-permitting modeling is needed for more accurate predictions of the timing, distribution, and intensity of extreme precipitation events over China (e.g., Zhang et al., 2013; Xu et al., 2015).
Most studies of convection-permitting simulations have adopted non-hydrostatic regional models developed for weather forecasting or regional climate modeling (Prein et al., 2015). Global models capable of simulating non-hydrostatic dynamics are not as common as regional models, but they offer some advantages including the ability to provide global forecasts or simulations while avoiding numerical issues associated with lateral boundary conditions that are major sources of uncertainty in regional modeling and also limit regional feedback to large-scale circulation (e.g., Giorgi and Mearns, 1991; Wang et al., 2004; Laprise, 2008; Leung et al., 2013; Prein et al., 2015). Non-hydrostatic global variable-resolution models, in particular, are useful as they allow convection-permitting simulations to be performed using regional refinement that significantly reduces computational cost compared to global convection-permitting modeling. Although global hydrostatic variable-resolution climate models, such as the variable-resolution version of Community Earth System Model, have been used in various applications in the last few years (e.g., Rauscher et al., 2013; Zarzycki et al., 2014, 2015; Rhoades et al., 2016; Huang et al., 2016; Wu et al., 2017; Gettelman, et al., 2018; Wang and Ullrich, 2018; Burakowski et al., 2019), so far few studies have used global non-hydrostatic variable-resolution models to investigate weather or climate simulations, particularly at convection-permitting scales (e.g., Prein et al., 2015). This study explores the use of a non-hydrostatic global variable-resolution model, the Model for Prediction Across Scales (MPAS), to model an extreme precipitation event in East China.
MPAS is a new multiscale modeling approach developed to take advantage of
advances in mesh generation by employing the spherical centroidal Voronoi
tessellations (SCVTs) (Du et al., 1999; Ringler et al., 2008). The SCVTs in
MPAS enable local mesh refinement through the mesh generation process whereby
a specified scalar density function determines higher- and lower-resolution
regions in the mesh (see, e.g., Ju et al., 2011). Meshes can be configured
with multiple high-resolution regions, and high resolution in one region
does not need to be balanced by coarser resolution elsewhere. The underlying
theory of SCVTs is robust concerning mesh properties and mesh generation.
The atmospheric solver in MPAS (Skamarock et al., 2012) integrates the
non-hydrostatic equations, and as such it is suitable for both weather and
climate simulation, i.e., for both non-hydrostatic and hydrostatic flow
simulation. MPAS has been evaluated and used in previous studies to
investigate the resolution impact on modeling clouds and precipitation
(O'Brien et al., 2013; Zhao et al., 2016), the structure of the
intertropical convergence zone (ITCZ) (Landu et al., 2014), precipitation
extremes (Yang et al., 2014), atmospheric river frequency (Hagos et al.,
2015), the position and strength of the eddy-driven jet (Lu et al., 2015),
global atmospheric predictability at convection-permitting scales (Judt,
2018), and regional climate modeling (Sakaguchi et al., 2015, 2016). Except
for Zhao et al. (2016) and Judt (2018), the aforementioned studies used a
hydrostatic version of MPAS applied at resolutions ranging from
To date, few studies have examined the MPAS performance in modeling extreme
precipitation events, particularly at grid scales of
The rest of the paper is organized as follows. Section 2 describes briefly the MPAS, the physics parameterizations, and the model configuration for this study, followed by a description of data for evaluation. The series of global uniform- and variable-resolution experiments are analyzed in Sect. 3. The findings are then summarized in Sect. 4.
This study uses a fully compressible non-hydrostatic model (MPASv5.2) developed for weather prediction and climate applications. The non-hydrostatic dynamical core of MPAS is described in Skamarock et al. (2012). MPAS uses C-grid staggering of the prognostic variables and centroidal Voronoi meshes to discretize the sphere. The unstructured spherical centroidal Voronoi tessellation (SCVT) generation algorithms can provide global quasi-uniform-resolution meshes as well as variable-resolution meshes through the use of a single scalar density function, hence opening opportunities for regional downscaling and upscaling between mesoscales and non-hydrostatic scales to hydrostatic scales within a global framework. The vertical discretization uses the height-based hybrid terrain-following coordinate (Klemp, 2011), in which coordinate surfaces are progressively smoothed with height to remove the impact of small-scale terrain structures. The dynamical solver applies the split-explicit technique (Klemp et al., 2007) to integrate the flux-form compressible equations. The basic temporal discretization uses the third-order Runge–Kutta scheme and explicit time-splitting technique (Wicker and Skamarock, 2002), similar to that used in the Weather Research and Forecasting (WRF) model (Skamarock and Klemp, 2008). The scalar transport scheme used by MPAS on its Voronoi mesh is described in Skamarock and Gassmann (2011), and the monotonic option is used for all moist species. The extensive tests of MPAS using idealized and realistic cases verify that smooth transitions between the fine- and coarse-resolution regions of the mesh lead to no significant distortions of the atmospheric flow (e.g., Skamarock et al., 2012; Park et al., 2013).
In the current version (v5.2) of MPAS, there are a few physics schemes
available. Three convective parameterizations can be used. The Kain–Fritsch
(KF; Kain, 2004) and the new Tiedtke (NTD; Bechtold et al., 2004, 2008,
2014) schemes represent both deep and shallow convection using a mass flux
approach with a convective available potential energy (CAPE) removal timescale (Kain, 2004). The third one, the GF scheme (Grell and Freitas, 2014),
is based on the Grell–Dévényi ensemble scheme (Grell and Dévényi, 2002)
using the multi-closure, multi-parameter, ensemble method but with
improvements to smooth the transition to cloud-resolving scales following
Arakawa et al. (2011). This scale awareness is critical for global variable-resolution simulation across hydrostatic (e.g., tens of kilometers) and
non-hydrostatic scales (e.g., 4 km). Fowler et al. (2016) implemented the GF
convective parameterization in MPAS and examined the impacts of horizontal
resolution on the partitioning between convective-parameterized and
grid-resolved precipitation using a variable-resolution mesh in which the
horizontal resolution varies between hydrostatic scales (
In this study, the height coordinate of MPAS is configured with 55 layers, and the model top is at 30 km. Multiple experiments are conducted with MPAS using quasi-uniform-resolution meshes and variable-resolution meshes. Two quasi-uniform-resolution meshes and three variable-resolution meshes are configured, similar to those shown in Fig. 1a and b that are coarsened to display the structure of the individual mesh cells. The quasi-uniform mesh has essentially the same mesh spacing globally, while the variable-resolution mesh has finer mesh spacing in the refined region with a transition zone between the fine- and coarse-resolution meshes. More details about the mesh generation can be found in Ringler et al. (2011). The two quasi-uniform meshes have grid spacing that approximately equals 15 km (U15km) and 60 km (U60km). The three variable-resolution meshes feature a circular refined high-resolution region centered over East China as shown in Fig. 1c. Figure 1c shows the exact mesh size distribution of the 4–60 km variable-resolution mesh (V4km) that has a refined region with grid spacing of approximately 4 km, and the mesh spacing gradually increases through a transition zone to approximately 60 km for the rest of the globe. The other two variable-resolution meshes (V16km and V30km) have a similar mesh structure as the V4km mesh but with a mesh spacing of 16 and 30 km, respectively, over the refined region that gradually increases to 128 and 120 km, respectively, elsewhere.
Experiments U15km and V16km are compared to examine the difference between
global uniform- and variable-resolution simulations in capturing the
precipitation in the refined region in order to explore the potential of
regional refinement for regional weather and climate simulation. It is
noteworthy here that the U15km mesh comprises
The U60km, V30km, V16km, and V4km experiments are conducted to quantify the
impacts of horizontal resolution on simulating precipitation
characteristics. The numbers of grid cells in the U60km, V30km, V16km, and
V4km meshes are
As discussed above, GF is the only convective parameterization that has been tested with scale-aware capability for use across hydrostatic (e.g., tens of kilometers) and non-hydrostatic scales (e.g., 4 km). Therefore, in order to investigate the difference among the experiments with the four meshes (U60km, V30km, V16km, and V4km), they are all conducted with the GF convective parameterization. Since the cloud microphysics has a significant impact on the V4km simulations (discussed latter), the experiments of V4km with both the WSM6 (V4km.WSM6) and Thompson (V4km.Thompson) cloud microphysics schemes are analyzed in this study. When examining the difference between the global uniform- and variable-resolution simulations and investigating the impact of mesh spacing, the same physics schemes and parameter values are used in multiple experiments if not specified explicitly. All the numerical experiments discussed above are summarized in Table 1.
Numerical experiments conducted and analyzed in this study.
(1) U and V represent quasi-uniform and variable-resolution meshes, respectively, as described in Sect. 2.1.2. (2) “WSM6” and “Thompson” represent two cloud microphysics schemes as described in Sect. 2.1.1. NTD and GF represent two cumulus parameterizations as described in Sect. 2.1.1.
Due to the large computing cost and data storage of the experiments conducted, particularly for the U15km and V4km experiments, this study does not perform ensemble simulations. Instead, bootstrapping statistical analysis is used to test the statistical significance of the difference among multiple experiments investigated in this study. The bootstrap method uses a resampling technique to extract certain samples, called bootstrap samples, within the range of the original data. Statistical metrics, such as averages, variances, and correlation coefficients, can be calculated for each bootstrap sample. For a given confidence level (e.g., 95 %), bootstrap confidence intervals for specific statistical metrics can be estimated (e.g., Efron, 1992; Efron and Tibshirani, 1994).
To simulate the heavy precipitation event that occurred during 25–27 June 2012 over the YRD in East China, all the MPAS experiments were initialized
at 00:00 UTC on 23 June 2012 to allow for appropriate spin-up time, and the
modeling results for 25–27 June 2012 are analyzed. The simulations were
initialized using the analysis data at 1
Several datasets are used to evaluate the MPAS simulations. The hourly
precipitation dataset from the National Meteorological Information Center of
CMA is used to evaluate the simulated precipitation characteristics. In
this dataset, rainfall was measured by tipping buckets,
self-recording siphon rain gauges, or automatic rain gauges. The data
were subject to strict three-step quality control by station, provincial,
and national departments. The methods of quality control mainly include
checking the climate threshold value, extreme value, spatial and temporal
consistency, and human–computer interaction. All the
data used in this study are quality controlled. The distribution of stations
over the study domain is shown as the color-filled circles in Fig. 2. Over
the YRD region of East China (25–36
Spatial distributions of precipitation and wind fields at 850 hPa
averaged during the event (25 June 00:00 to 27 June 12:00 UTC) from the
simulations with global uniform (15 km) and variable (16 km over the
refined region as shown in Fig. 1c) resolutions. The observed mean
precipitation from the CMA stations and the wind fields from the ERA5
reanalysis are shown. The black contour lines represent precipitation
larger than 20 mm d
Figure 2 shows the spatial distributions of precipitation and wind at 850 hPa averaged during the event (25 June 00:00 UTC to 27 June 12:00 UTC) from
the simulations with global uniform (15 km) and variable (16 km over East
China) resolutions (U15km.NTD and V16km.NTD). The mean precipitation from
the CMA stations and the winds from the ERA5 reanalysis are also shown. The
CMA observations show an average precipitation rate exceeding 50 mm d
As mentioned above, the precipitation during this event is concentrated in a
narrow west–east belt. For a more quantitative comparison, Fig. 3 shows
the zonal averaged precipitation during the event over the YRD region of
East China (25–36
Zonal distributions of precipitation averaged during the event (25 June 00:00 UTC to 27 June 12:00 UTC) averaged over East China (denoted as the black box in Fig. 2) from the CMA station observations and the simulations with global uniform (15 km, solid lines) and variable (16 km over the refined region as shown in Fig. 1c, dashed lines) resolutions with two convective parameterizations (GF, red lines; NTD, green lines). The modeling results are sampled at the CMA station.
Figure 4 shows the meridional precipitation propagation over East China
(denoted as the black box in Fig. 2) during the event. The CMA observations
indicate that the rain belt propagates from 26
Time–latitude cross section of precipitation during the event averaged over East China (denoted as the black box in Fig. 2) from the CMA station observations and the simulations with global uniform and variable resolutions with two convective parameterizations. The modeling results are sampled at the CMA stations.
Overall, for the selected event, the MPAS simulations at global uniform and variable resolutions produce consistent results over the refined region with comparable horizontal resolution in terms of the spatial patterns of precipitation and wind fields as well as the precipitation propagation. This finding is in general agreement with findings from previous studies for MPAS with idealized experiments (e.g., Zhao et al., 2016) and real-world experiments (e.g., Sakaguchi et al., 2015). These findings provide the basis for using global variable-resolution configurations of MPAS to model extreme precipitation over East China. In the following, the impacts of resolution on modeling extreme precipitation during this event are investigated with multiple global variable-resolution experiments.
Multiple experiments using MPAS at various resolutions are conducted as
stated in the “Data and methodology” section. The resolution crosses the scales from 60, 30, and 16 to 4 km. For global variable-resolution configurations, a
scale-aware convective parameterization is needed, especially for the
configuration that crosses hydrostatic (convective parameterization is
required) and non-hydrostatic scales (convection permitting). Therefore, the
experiments analyzed below are all conducted with the GF scheme that is
developed for simulations down to
Spatial distribution of averaged parameterized and resolved precipitation during the event over East China from the simulations with resolutions of 60, 16, and 4 km.
Figure 6 shows the observed and simulated spatial distributions of
precipitation and wind fields at 850 hPa averaged during the event. For
comparison, the GFS forecast results at resolutions of 1.0 and
0.5
Spatial distributions of precipitation and wind fields at 850 hPa
averaged during the event from the MPAS simulations at resolutions of 60, 30, 16, and 4 km. The observed mean precipitation from the CMA
stations and the wind fields from the ERA5 reanalysis are shown as well. The
black contour lines represent precipitation larger than 20 mm d
The mean bias (MB) and root mean square root (RMSE) of the simulated results shown in Figs. 6–8 and 10 against CMA observations.
In order to test the statistical significance of the difference in spatial
distributions among the experiments, the 95 % confidence intervals of
spatial correlation are estimated based on the bootstrap analysis. Although
the correlation coefficients estimated above have an uncertain range, at the
95 % confidence level the results still indicate that the V16km simulation
produces a better spatial pattern of precipitation than other
hydrostatic-scale simulations. In addition, the simulation at the
convection-permitting scale is comparable to, if not better than, the V16km
simulation. The results are summarized in Table 3. It is noteworthy that,
although the difference in precipitation over East China is significant
among the GFS forecasts at 0.5 and 1.0
The zonal distributions of precipitation can better demonstrate the
difference among the simulations. Figure 7 shows the observed and simulated
zonal distributions of precipitation averaged during the event over the YRD
region of East China. For comparison, the GFS forecasts at 1 and
0.5
The correlation coefficients and the corresponding 95 % confidence intervals based on the bootstrap analysis for the results shown in Figs. 6–10.
The values inside the parentheses indicate the lower and higher bounds of the 95 % confidence intervals; the values outside are estimated directly based on the results shown in Figs. 6–10.
Zonal distributions of precipitation averaged during the event
averaged over East China (denoted as the black box in Fig. 6) from the CMA
station observations and the simulations with resolutions of 60, 30, 16, and 4 km. For comparison, the GFS forecasts at 1 and 0.5
Figure 8 compares the observed and simulated precipitation propagation
during the event over East China. The modeling results are sampled at the
CMA stations. The GFS forecasts at 0.5 and 1.0
Time–latitude cross section of precipitation during the event
averaged over East China (denoted as the black box in Fig. 6) from the CMA
station observations, GFS forecasts at 0.5 and 1.0
The MPAS simulations are highly dependent on the resolutions. All simulations roughly produce the two peaks of precipitation as observed during the event. However, the experiment at 60 km simulates the first precipitation peak southward and the second peak northward of the observations, while the experiment at 30 km simulates the second peak further south and a few hours earlier. The time and location shifts correspond well to biases in simulated wind shear (Fig. S10). The spatial correlation coefficients of precipitation are 0.30 and 0.32 between the observations and the simulations at 60 and 30 km, respectively. The experiments at 16 and 4 km with the WSM6 cloud microphysics scheme can better capture the timing and latitude of the observed precipitation event than U60km and V30km (Fig. S11); however, both V16km and V4km overestimate the first peak precipitation and underestimate the second peak. The experiment at 4 km with the Thompson scheme overestimates the precipitation amount of both peaks. Overall, all the simulations overestimate the observed precipitation during the event (Table 2). The correlation coefficients are 0.41 and 0.42 (0.38) for 16 and 4 km with the WSM6 (Thompson) cloud microphysics schemes, respectively. At the 95 % confidence level (Table 3), the experiments at 16 and 4 km are comparable in terms of simulating the propagation of this rain belt and better than the experiments at other resolutions. It is interesting to note that MPAS and GFS forecasts share the same initial condition and simulate different large-scale circulation, particularly the wind shear structure, with the system evolution (Fig. S10). The model capability in successfully capturing the wind shear structure during this event determines the performance in generating the rain belt evolution. The formation and evolution of wind shear during the Meiyu front over East China have been found to interact with multiscale processes and systems, including terrain and convective latent heat (Yao et al., 2017). A different representation of the terrain over East China in various resolutions may impact the simulated wind shear structure. Previous studies also found that convective latent heat may vary with resolutions and physics (Hagos et al., 2015; Zhao et al., 2016), which can further affect the simulation of wind shear structure. Therefore, the difference in resolution and physics between MPAS and GFS may result in their difference in simulating the formation and evolution of wind shear structure during the event. A more detailed exploration of the differences between the MPAS and GFS simulations is beyond the scope of this study.
The spatial distribution of the rain belt can also be reflected by the
vertical wind distributions. Figure 9 compares the height–latitude cross
section of the winds averaged over the region (shown as in Fig. 6) during
the event from the ERA5 reanalysis, the GFS forecasts, and the MPAS
simulations. In the ERA5 reanalysis wind fields, vertical motion is located
primarily around 31
Height–latitude cross section of wind fields averaged over the
region (the entire domain as shown in Fig. 6) during the event from the ERA5
reanalysis, the GFS forecasts at 0.5 and 1.0
Besides predicting the spatial and temporal variations of the rain belt, it
is also critical to capture the location and intensity of extreme
precipitation within the heavy rain belt. Since the GFS forecasts shift the
entire rain belt northward, only the MPAS simulations are analyzed here.
Figure 10 shows the spatial distributions of precipitation averaged during
the event over the heavy rain region (27–32
Spatial distributions of precipitation averaged during the event
over the heavy precipitation region (27–32
Figure 11 shows the probability density functions (PDFs) of hourly
precipitation at all the CMA stations over East China during the event. The
simulations are sampled at the CMA stations. Precipitation above
Probability density functions (PDFs) of hourly precipitation at all the CMA stations during the event over East China (denoted as the black box in Fig. 6) from the CMA observations and the MPAS simulations at resolutions of 60, 30, 16, and 4 km. The simulations are sampled at the CMA stations.
Previous studies found that the distribution of extreme precipitation
correlates well with that of the lower-tropospheric upward vertical velocity
(e.g., Zhao et al., 2016). Figure 12 shows the PDFs of hourly upward
vertical velocity averaged below 700 hPa at all the CMA stations during the
event from the MPAS simulations. In general, the comparison of lower-level
upward vertical velocity among the experiments is consistent with that of
precipitation (Fig. 11) in simulations at hydrostatic scales (i.e., 60, 30, and 16 km in this study) that produce higher frequencies of updrafts
Figure 13 shows the PDFs of the upward moisture flux and the relationship
between hourly precipitation versus upward moisture flux at 850 hPa during
the event from the MPAS simulations at 60, 30, 16, and 4 km. It is
evident that the simulations at higher resolutions produce more frequent
intense upward moisture fluxes at 850 hPa, consistent with Rauscher et al. (2016) and O'Brien et al. (2016). Rauscher et al. (2016) found a linear
relationship between precipitation and upward moisture fluxes at the lower
level. The relationship lines from this study as shown in Fig. 13 parallel
the
Probability density functions (PDFs) of hourly upward vertical velocity averaged below 700 hPa at all the CMA stations during the event over East China (denoted as the black box in Fig. 6) from the MPAS simulations at resolutions of 60, 30, 16, and 4 km.
In this study, a series of MPAS simulations of a heavy precipitation event over East China, triggered by a typical southwest vortex in the middle and high troposphere and wind shear in the lower layer of the Meiyu front during the East Asian summer monsoon, are compared. The simulations are performed at various resolutions from hydrostatic (60, 30, 16 km) to non-hydrostatic (4 km) scales. Consistency between the MPAS simulations at global uniform and variable resolutions is also investigated. Besides the impacts of resolution on simulating heavy precipitation, the impacts of convective and cloud microphysics schemes are also examined. All the MPAS simulations are evaluated using the CMA station observations of precipitation and the ERA5 reanalysis of winds; they are compared against the NCEP GFS forecasts that share the same initial condition as the MPAS simulations.
In general, the MPAS simulations at global uniform (U15km) and variable (V16km) resolutions produce similar results in terms of the spatial and temporal distributions of precipitation and winds inside the refined region over East China. Both experiments can capture the observed precipitation characteristics. This suggests that the global variable-resolution configuration of MPAS may be appropriate to simulate heavy precipitation over East China, which is also consistent with findings from previous studies using variable-resolution MPAS with regional refinement over other parts of the globe (e.g., Sakaguchi et al., 2015; Zhao et al., 2016). The simulations with two different convective parameterizations show that the MPAS-simulated distributions of precipitation are affected by the convective schemes at hydrostatic scales, while the impacts from the cloud microphysics schemes are small.
Hourly precipitation versus upward moisture flux at 850 hPa during the event over East China (denoted as the black box in Fig. 6) from the MPAS simulations at resolutions of 60, 30, 16, and 4 km (solid line, left axis), with the PDFs of the upward moisture flux (dashed line, right axis).
The variable-resolution simulations spanning hydrostatic and non-hydrostatic
scales reveal that the scale-aware GF convective parameterization produces
less convective-parameterized precipitation as the horizontal resolution
increases. Comparison against the station observations indicates that the
MPAS simulations at 16 and 4 km can generally better capture the observed
temporal and zonal distribution of the rain belt in the simulated event than
the simulations at coarser resolutions. The experiments at 4 km can better
capture the areas with heavy precipitation (
The performance of MPAS at the convection-permitting scale is quite sensitive to the cloud microphysics scheme in terms of the distribution and intensity of extreme precipitation. This is consistent with Feng et al. (2018), who found that cloud microphysics parameterizations in convection-permitting regional simulations have important effects on macroscale properties such as the lifetime, precipitation amount, and stratiform versus convective rain volumes of mesoscale convective systems in the US. They attributed the impacts to the representation of ice-phase hydrometeor species that influence mesoscale convective systems through their influence on the diabatic heating profiles that provide dynamic feedback to the circulation (Yang et al., 2017). Hence, more efforts may be needed to improve cloud microphysics processes for modeling extreme precipitation at the convection-permitting scale in the future. In the meantime, aerosols have been found to play a critical role in simulating some heavy precipitation events over China through their impacts on cloud microphysics and/or radiation (e.g., Zhong et al., 2015, 2017; Fan et al., 2015). The current version of MPAS does not represent aerosol–radiation and aerosol–cloud interactions, which may also contribute to the biases of extreme precipitation at convection-permitting scales. Lastly, it is also noteworthy that the resolution of 4 km may still be insufficient to resolve some convective cells, which may also contribute to the modeling biases (Bryan and Morrison, 2012).
This study provides the first evidence supporting the use of the global variable-resolution configuration of MPAS to simulate extreme precipitation events over East China. In particular, the MPAS variable-resolution experiment at the convection-permitting scale (4 km) improves the simulated distribution and intensity of precipitation over the area of interest, which is consistent with previous studies using regional convection-permitting models (e.g., Zhang et al., 2013; Prein et al., 2015; Yang et al., 2017; Gao et al., 2017; Feng et al., 2018). The higher-resolution MPAS experiments better simulate the spatial distribution of heavy precipitation over the complex topographic region of East China, which suggests that topography may play a critical role and deserves further investigation in the future. Our results show that cloud microphysics parameterizations have important effects on convection-permitting simulations, but modeling other physical processes such as boundary layer turbulence, radiation, and aerosols may also affect the skill of convection-permitting simulations. The GFS forecasts analyzed in this study show significant biases in precipitation distribution. The zonal shift of the rain belt by the MPAS simulations at coarser resolutions compared to simulations at finer resolutions suggests that resolution may have contributed to the GFS forecast biases. A more detailed exploration of the differences between the MPAS and GFS simulations is beyond the scope of this study.
Previous studies (Xue et al., 2007; Clark et al., 2016) noted the importance of ensemble simulations in predicting heavy precipitation. Due to the computational limitation, only one set of experiments with different physics and resolutions is evaluated in this study. The MPAS simulations of heavy precipitation with different initial conditions and refinement sizes deserve more evaluations. Finally, some studies noted that convection-permitting modeling does not always add value in simulating heavy precipitation compared to hydrostatic-scale modeling (e.g., Kain et al., 2008; Rhoades et al., 2018; Xu et al., 2018). Rhoades et al. (2018) found that improvement through increasing resolution may also depend on the cloud microphysics parameterization. Increasing horizontal resolution alone can sometimes even lead to worse model performance. The impacts of increasing horizontal resolution on the overall model performance in simulating extreme precipitation may also be affected by the model structure and coupling among model components and processes (Jeevanjee et al., 2017; O'Brien et al., 2016; Herrington and Reed, 2017, 2018; Gross et al., 2018). This study also found some sensitivity of modeling extreme precipitation to cloud microphysics, particularly at the convection-permitting scale. More events involving heavy precipitation over East China should be investigated in the future to more systematically evaluate the MPAS variable-resolution modeling framework and the impacts of resolution and physical parameterizations.
The MPAS release v5.2 can be obtained at mpas-dev.github.io. Global meshes generated for the experiments used in this study are available upon request by contacting the corresponding author Yu Wang (wangyu09@ustc.edu.cn) or Chun Zhao (chunzhao@ustc.edu.cn).
The supplement related to this article is available online at:
CZ and YW designed research. MX performed the simulations. CZ, MX, MZ, and ZH analyzed the simulations. JG collected and analyzed the observations. CZ, MX, and YW wrote the paper. LRL, MD, and WS guided the experiment design and edited the paper.
The authors declare that they have no conflict of interest.
This research was supported by the Ministry of Science and Technology of China under grant 2017YFC1501401 and the Fundamental Research Funds for the Central Universities. The study used computing resources from the High-Performance Computing Center of the University of Science and Technology of China (USTC) and the TH-2 of the National Supercomputer Center in Guangzhou (NSCC-GZ). L. Ruby Leung was supported by the U.S. Department of Energy Office of Science Biological and Environmental Research as part of the Regional and Global Modeling and Analysis program. PNNL is operated for the Department of Energy under contract DE-AC05-76RL01830.
This research has been supported by the Ministry of Science and Technology of China (grant no. 2017YFC1501401) and the Fundamental Research Funds for the Central Universities (grant no. NA).
This paper was edited by Jason Williams and reviewed by three anonymous referees.