1. Introduction
The Tibetan Plateau (TP) and its surrounding mountain ranges, including the Hindu Kush, Himalayas, and Karakoram, together represent one of the world’s largest mountain regions. Due to the enormous freshwater storage in the cryosphere of the TP, this region is crucial for mountain ecosystems and downstream societies (Immerzeel et al. 2020). While extensive hydroclimatic changes have been observed during the past decades (Bibi et al. 2018), many of the underlying processes and the extent of future climate change impacts remain largely unexplored. A major uncertainty in understanding the regional water cycle is related to convective precipitation, which is difficult to observe due to the sparse distribution of gauge stations and topography-related challenges in remote sensing. Since convective precipitation occurs on various temporal and spatial scales, it is linked to multiple processes and shows remarkable regional differences across the TP region (Maussion et al. 2014; Kukulies et al. 2020).
A specific type of weather system that often evolves in the lee of high mountains is a mesoscale convective system (MCS). MCSs are complexes of individual storm cells that grow by generating their own circulation or by interacting with the large-scale flow (Houze 2004). Consequently, they are characterized by strong convective precipitation and large areas of stratiform precipitation (Houze 2004). MCSs often occur downstream of mountain regions because the dynamic (e.g., wave–orography interactions; Buzzi et al. 1987; Schultz and Doswell 2000) and thermodynamic (e.g., capping inversions; Carlson et al. 1983; Rasmussen et al. 2011) modifications of the atmosphere by the mountain barrier together with the increased moisture supply at lower elevations can create favorable conditions for storm initiation (Rasmussen and Houze 2016). The importance of MCSs in a regional climate change context is twofold. On one hand, such storm systems significantly contribute to the total summer rainfall in the south and east of the TP (Kukulies et al. 2021) as well as over large parts of eastern China (Li et al. 2020), which means that a change in their climatology could affect water availability in these regions. On the other hand, MCSs pose a severe risk for people’s lives and livelihoods, as they can produce substantial amounts of rainfall within a short time. Since extreme precipitation is expected to intensify in a warmer climate (e.g., Donat et al. 2016; Prein et al. 2017), while the buffering capacity of snow in high mountain regions is expected to decrease (Hock et al. 2019), this could imply that MCSs around mountain regions will more often develop into natural hazards and lead to destructive flooding.
A key challenge with state-of-the-art climate models is to predict future changes in convective storms and orographic precipitation. Such predictions are hindered by systematic model biases related to their coarse grid spacing (Prein et al. 2015). Kilometer-scale climate simulations offer a promising solution to this challenge as they have can better represent mesoscale processes and explicitly resolve deep convection at horizontal grid spacings < 4 km. In particular, processes related to extreme precipitation and convection at shorter time scales have been shown to be more realistically represented in kilometer-scale simulations (Prein et al. 2013; Ban et al. 2014; Lind et al. 2020; Ban et al. 2021). Previous studies have shown that simulations with grid spacings around 4 km can capture MCS characteristics reasonably well (Prein et al. 2020), even if individual thunderstorms may still not be resolved (Prein et al. 2021). While a few kilometer-scale simulations in the TP region have shown added value in simulating precipitation (Ou et al. 2020; Gao et al. 2020; Zhou et al. 2021; Li et al. 2021), there are no studies that have focused on MCS characteristics or the representation of organized convection in such simulations. Besides the improvement gained by simulating convective processes more realistically, kilometer-scale models better represent topography and small-scale surface features that are not resolved in coarser resolution models (Rasmussen et al. 2011; Ikeda et al. 2021). This can be particularly important in high-altitude regions, where convective processes may be influenced by complex land–atmosphere interactions (Sugimoto and Ueno 2010) and diurnal mountain wind systems (Liu et al. 2009; Johnson 2011).
In an ongoing Coordinated Regional Climate Downscaling Experiment (CORDEX) flagship pilot study named Convection-Permitting Third Pole (CPTP), we explore the benefits and limitations of kilometer-scale simulations in the TP region (Prein et al. 2023). The overall purpose of the project is to acquire new knowledge about mesoscale processes that control the regional water cycle and hydrological regimes, with a focus on organized convection and precipitation. While the long-term goal is to perform multiyear simulations that cover both past and future periods, the modeling strategy is to first evaluate specific cases and simulations on shorter time scales and to perform a sensitivity analysis to test different model options (Prein et al. 2023). The rationale is that these cases can provide insights into model sensitivities and skills to represent certain processes, which, in turn, will guide the setup and evaluation of long-term climate simulations in the TP region.
After an overarching model evaluation of three meteorological cases presented in Prein et al. (2023), we focus here on the in-depth analysis of one of the three cases, which was an MCS case in July 2008 that led to flooding in the Sichuan basin. This heavy precipitation event occurred in conjunction with a so-called Tibetan Plateau vortex (TPV), a frequently occurring low pressure system that forms over the TP and occasionally moves off into the eastern downstream regions (Curio et al. 2019). Previous studies have highlighted the role of diabatic heating in TPV genesis and development (e.g., Dell’Osso and Chen 1986; Zhang et al. 2019), because surface or latent heating processes can restructure potential vorticity (Wu et al. 2020, 2022). In addition, there is also some evidence that large-scale factors such as upper-level baroclinicity and jet latitude are correlated with the interannual variability of TPV occurrences (Hunt et al. 2018), but there is no consensus about what controls the longevity of TPVs, especially not on climatological time scales. Curio et al. (2019) suggest that the position and intensity of the westerly jet control how far TPVs can travel, so the background flow could be an important factor that influences the impact of TPVs on downstream weather. While it is widely recognized that TPVs can trigger heavy precipitation events when they enter the moister and lower elevation basins (e.g., Curio et al. 2019), the details of their interaction with organized convection and water vapor transport and the ability of kilometer-scale models to simulate these complex interactions are largely unknown. It should be noted that the definition of a TPV is clearly distinct from the definition of MCSs because they occur at larger scales (typically between 400 and 1000 km) and are not, per definition, linked to convection and precipitation (Feng et al. 2017). They may, however, act as a mesoscale weather system that generates or enhances downstream convection. In that way, TPVs and MCSs do not necessarily describe two different mechanisms but rather two diagnostic quantities that may be linked to the same weather phenomenon, as we will discuss in more detail in our analysis.
This study complements the overarching evaluation of meteorological cases and aims to provide a more detailed view of how different model configurations and models simulate physical processes that influence the quality of the simulated downstream MCS and associated precipitation in a multimodel and multiphysics ensemble. The goal is also to link insights gained from the sensitivity study of a specific MCS case with an MCS climatology to discuss potential implications for long-term climate simulations.
The more specific research questions are as follows: 1) What were the key processes for MCS formation and associated precipitation in this specific case? 2) How do different kilometer-scale regional modeling systems represent these processes and what determines the performance of capturing this MCS? 3) what can we learn from this case study for future simulations of MCSs in the TP region?
To address these questions, we evaluate an ensemble of 36 simulations against different observational datasets and compare simulated mesoscale and large-scale processes with reanalysis data. More specifically, we apply an MCS tracking algorithm and analyze differences in water vapor transport, vortex evolution, upper-level zonal wind, and vertical velocity to better understand the driving mechanisms of the MCS formation. The remainder of the paper is structured as follows: section 2 provides an overview of the simulation ensemble and describes the evaluation methods we used. Section 3a gives a more detailed description of the observed characteristics of the MCS case, and section 3b presents an evaluation of the simulated precipitation. The key processes for MCS formation are further explored in sections 3c–3g. Finally, the generalizability of the case is discussed based on results from a 1-yr-long simulation in section 4, and the main conclusions are presented in section 5.
2. Data and methods
a. Simulations
The simulation ensemble used in this study includes 36 members from five models: the Weather Research and Forecasting (WRF) Model (Skamarock and Klemp 2008), Model for Prediction Across Scales (MPAS; Skamarock et al. 2012), Consortium for Small-Scale Modeling–Climate Limited Area (COSMO-CLM; Rockel et al. 2008), Icosahedral Nonhydrostatic (ICON; Zängl et al. 2015), and Regional Climate Model (RegCM) system (Coppola et al. 2021). In addition to different modeling systems, the ensemble members cover differences in domain size, horizontal grid spacing, physics options, and initialization time. Physics sensitivities were tested with the WRF model by replacing one parameterization scheme at a time with respect to a WRF reference run (WRFD1). We used different physics schemes for microphysics (namely, WRFWDM6, WRFWDM7, WRFYlin, and WRFMorrison), PBL (WRFMYJ, WRFMYNN2.5, WRFMYNN3, WRFMYNN3-WDM7, WRFYSU, WRFBougeault, WRFQNSE, WRFShinHong, WRFGBM, WRFCAMUW, WRFCAM, WRFMYNN2.5-MYNN, WRFMYNN3-MYNN, WRFMYNN2.5, and WRFMYJ-Monin-Obukov), and land surface processes (WRFNoah). In addition, we performed one simulation with cumulus parameterization turned on (WRFMSKF), while all other simulations are run with no parameterization of convection or only with parameterization of shallow convection (ICON, MPAS, MPASLA). Table 1 provides an overview of all ensemble members, their respective model, horizontal grid spacing, domain, and the main difference to WRFD1. Note that the simulations WRFD1-Cardiff and WRFD1-NJU were conducted with the same settings as WRFD1 but by different research groups, which allows us to check the effect of the computer system used for the simulation.
Overview of simulation ensemble. LBC is lateral boundary condition; this column refers to the frequency at which the lateral boundary conditions are updated from ERA5.
The reference run was conducted using a one-way nesting approach in which D1 (Fig. 1) represents the inner domain with a horizontal grid spacing of 4 km, driven by D2 (Fig. 1) as the outer domain with a horizontal grid spacing of 12 km. The parent domain was forced by 6-hourly ERA5 data (Hersbach et al. 2020) and started 18 months before the simulation period of the MCS case (0000 UTC 14 July 2008–2300 UTC 24 July 2008). Land surface variables that usually need a longer spinup time (i.e., snow and soil moisture) were initialized with the output from the 12-km simulation. The simulation WRFD1-direct was directly driven by ERA5 to test the effect of internal variability developing in the 12-km model domain. The reference run uses Thompson microphysics (Thompson et al. 2008), the Yonsei University (YSU) (Hong et al. 2006; Hong 2010) planetary boundary layer scheme, the RRTMG longwave and shortwave radiation scheme (Ruiz-Arias et al. 2013), and the Noah land surface scheme with multiparameterization options (Noah-MP; Niu et al. 2011). More details about the parameterization schemes we used and the general model strategy can be found in Prein et al. (2023).
Topographic map of the 4-km model terrain height (m MSL) showing the five domains used for the kilometer-scale simulations. Note that all WRF, RegCM, and ICON experiments are run on D1, whereas CCLM is run on D4 and MPAS is run on D5. The zone between the two circles of D5 (magenta lines) indicates the MPAS transition zone between the 4-km domain and the coarser resolution. The larger domains D2 and D3 are only used for additional WRF experiments to test the domain size sensitivity. The black solid line within the domains indicates the 1000-m contour, which outlines the Sichuan basin east of the TP.
Citation: Journal of Climate 36, 17; 10.1175/JCLI-D-22-0240.1
The model domains for the different simulations are shown in Fig. 1. The smallest domain, D1, is the standard domain that is used for most of the simulations, except for CCLM that is run on D4, MPAS that is run on D5, and a few additional WRF experiments that are run on D2 and D3 to test the domain-size sensitivity (Table 1). The size and location of D1 are motivated by a compromise between including as much of the TP topography as possible while keeping the domain size (and hence the needed computational resources) low so that more research groups were able to participate in this model experiment.
Five additional WRF simulations were performed specifically for the MCS case in this study that are not included in Prein et al. (2023): an additional run (WRFD3), starting the simulation one day for the event (WRFtimelag), using spectral nudging (WRFnudging), WRF8km, and WRF8km-vertical. The purpose of these simulations was to improve the simulated MCS location because most of the ensemble members show a displaced precipitation center compared to observations (see section 3b). Since Prein et al. (2023) found systematically improved precipitation statistics when enlarging the 4-km WRF domain from D1 to D2, we tested the additional run (WRFD3) with a further enlargement of ∼5° in each direction from D2 to D3 (Fig. 1). The reasoning behind WRFtimelag and the use of WRFnudging was to constrain the simulated large-scale circulation. In theory, this can be achieved by limiting the time difference between initialization and the simulated event (by avoiding a model drift over time) or through spectral nudging, which constrains the large-scale atmospheric state in the computational domain to be consistent with the ERA5 forcing data. This is done by imposing large-scale atmospheric variables at specific wavelengths on the regional model (von Storch et al. 2000). Specifically, we applied a 3-hourly nudging of horizontal wind speeds, geopotential height, and temperature above the boundary layer, assuming that boundary layer processes are more realistically represented by the regional model. The chosen wavenumber truncations above which the nudging is applied were two and three along the x and y directions. These correspond to the recommended synoptic scale of about 1000–2000 km for our specific domain (Gómez and Miguez-Macho 2017). The simulations WRF8km and WRF8km-vertical were used to test the effect of the vertical resolution on the simulation because the troposphere is considerably shallower over the TP and because the vertical grid spacing east of the TP is coarser compared to the vertical grid spacing at the same height over the TP. This could affect the simulation because turbulence is generated at the height of the TP and transported downstream at this height. In WRF8km-vertical, we used 100 model levels (instead of 50 model levels for all other WRF simulations; Table 1) to assess if an increased vertical resolution improves the simulation. Note that WRF8km-vertical is run on D2 and with a grid spacing of 8 km to assure a stable simulation, since a doubling of the vertical resolution could result in instabilities caused by the steep topography. Therefore, we also conducted WRF8km as a reference run, which looks similar to WRFD2 and can be directly compared to WRF8km-vertical. An 8-km horizontal grid spacing should be sufficient to capture the large-scale and nonconvective processes in the formation and intensification of the TPV. Note that the five additional simulations—WRFD3, WRFtimelag, WRFnudging, WRF8km, and WRF8km-vertical—were directly driven by ERA5 instead of the 12-km simulation.
Finally, we use a WRF simulation for the water year 2020 (September 2019–September 2020) to evaluate the model performance in simulating MCSs in the Sichuan basin. Here, the focus is on the general characteristics of MCSs and downstream precipitation to assess representatives of the analyzed MCS case in a longer time context. The 1-yr WRF simulation is configured the same way as WRFD2, which means 1) a direct downscaling of ERA5 on D2 and 2) a larger domain size compared to the reference run, because Prein et al. (2023) and the present study reveal a significant improvement in model performance when the domain is increased from D1 to D2. The first month of the simulation was used as spinup, and the initial conditions are taken from a 12-km simulation that was started on 1 October 2016. We performed the simulation with and without spectral nudging to assess the effect of spectral nudging on the 1-yr MCS climatology.
b. Model evaluation
The two primary observational datasets used to evaluate the simulated cloud shield and precipitation are hourly precipitation estimates from GPM IMERG, version 06 (Huffman et al. 2015), and hourly NCEP CPC brightness temperatures from merged geostationary satellite observations (Janowiak et al. 2017). Daily precipitation accumulations from gauge stations in the Sichuan basin are used as an additional data source to verify precipitation timing and intensity of the event. The ERA5 reanalysis is used as a reference dataset to identify synoptic processes that control the MCS formation, since the precipitation center is represented sufficiently well in ERA5 (Fig. 3e). Additional fields that are used in the presented analysis are relative vorticity at 500 hPa, vertically integrated water vapor transport, geopotential height at 500 hPa, and horizontal wind speed at 200 hPa.
To evaluate the model performance in simulating precipitation, we use the fraction skill score (FSS) as a spatial verification metric that takes different horizontal scales and different precipitation thresholds into account (Roberts and Lean 2008). Since we are interested in both the timing and intensity of the simulated event, we assess the model skill for hourly precipitation rather than the mean over the entire simulation period. The FSS is able to avoid the so-called double-penalty problem that occurs in a point-to-point verification where the forecast is penalized twice when precipitation is simulated at a different location or time (once for a miss and once for a false alarm) (Prein et al. 2013). The FSS avoids this problem because it compares the fractional number of grid points in model and observations that contain precipitation above a given threshold. Moreover, the comparison of binary fields (above versus below the specified precipitation threshold) has the advantage of revealing how well the model represents spatial patterns of heavy versus more moderate precipitation. While the FSS is a monotonic function that naturally increases for larger spatial scales, it helps identify at which spatial scales the model performance is acceptable. This is done by screening the precipitation fields with squares of different neighborhood sizes that move across the domain (Roberts and Lean 2008). The skill difference between the simulations is typically larger for smaller neighborhood sizes because localized precipitation depends to a larger extent on the model and model configuration used than on large-scale precipitation that is mainly controlled by the same initial and boundary conditions.
c. Scale separation
To investigate simulated vortices over the TP, we first apply a spectral filter to relative vorticity fields, because this facilitates the identification of mesoscale vortices and removes submesoscale noise. For this, we use the discrete cosine transformation (DCT), which transforms atmospheric fields into the spectral space by representing the data points as a sum of cosine functions with different frequencies (Denis et al. 2002). Since kilometer-scale simulations are characterized by high frequencies in the vorticity field, the removal of specific wavelengths results in a smoother field and facilitates the identification of TPVs. We filter out wavelengths between 400 and 1000 km since the horizontal extent of TPVs typically lies within that range (Curio et al. 2019). After transforming hourly fields of relative vorticity to spectral coefficients in the wavenumber space using the python package Scipy (Virtanen et al. 2020; https://docs.scipy.org/doc/scipy/reference/generated/scipy.fft.dctn.html), a two-dimensional transfer function with the same shape as the input data is constructed to multiply the spectral coefficients with values between 0 and 1. This multiplication acts as a bandpass filter and thus removes wavelengths beyond the range of interest before the field is transferred back to the original domain by applying an inverse DCT (https://docs.scipy.org/doc/scipy/reference/generated/scipy.fft.idctn.html). The reader is referred to Denis et al. (2002) for a more detailed description of the DCT and the construction of the two-dimensional transfer function. Note that we also used the filtered vorticity field to automatically track TPVs in ERA5, using the criteria presented in Curio et al. (2019). However, in results, we present Hovmöller diagrams to compare the TPV evolution between simulations, because most of the simulations do not exhibit continuous vortex tracks persisting for more than 24 h, as required by the tracking presented in Curio et al. (2019).
d. MCS tracking
An efficient way to evaluate the simulated MCS life cycle is to track contiguous clouds and precipitation and then compare track characteristics with those in observations (Feng et al. 2021a). We use an objective tracking algorithm that was previously used in Prein et al. (2021) and Poujol et al. (2020) to follow the detailed evolution of the cloud shield and its associated precipitation during the simulated MCS case as well as in the 1-yr simulation.
MCSs are defined as long-lasting and extensive storm systems with cloud shields exceeding 104 km2 and areas of contiguous precipitation > 100 km (Houze 2004). Note that the spatial scale of MCS-associated precipitation is, hence, much smaller than the spatial scale of the triggering vortex (see section 2c). To compare simulated cloud features with observed brightness temperatures (Tb), we first convert outgoing longwave radiation (OLR) from the model output to Tb using its empirical relationship to flux equivalent brightness temperature (for details, see Yang and Slingo 2001). We then match hourly brightness temperatures with hourly precipitation data in both models and observations and classify MCSs by applying the criteria presented in Feng et al. (2021b). This means a cloud must contain Tb values < 225 K, a cloud shield of <241 K over a minimum area of 104 km2 and a contiguous area with rain rates > 3 mm h−1 and a major axis of >100 km, whereby all three criteria have to persist continuously for more than 4 h. The MCS lifetime is defined by the time the cloud area < 241 K over 104 km2 persists. Note that the model simulations and the NCEP CPC brightness temperatures are first regridded to the native grid of GPM IMERG (0.1° × 0.1°) to facilitate collocations and to be consistent in the comparison between models and observations. The linking of convective features over time is done by connecting overlapping areas in space and/or time (Prein et al. 2021). This means that multiple small cloud features that later connect to one large cloud or a large cloud that decays into multiple small clouds will be included in the same track. To identify the target track among all tracks in the case simulation, we choose the tracked cloud with the largest mean area that is closest to the observed time and location of our event (see section 3a).
3. Results and discussion
a. Case description
The MCS case that is subject to this study occurred during 20 and 21 July 2008 in the Sichuan basin, close to the eastern slopes of the TP. The highest gauge-recorded rainfall was 280 mm within 48 h, which corresponds to more than a quarter of the annual mean precipitation in the basin (i.e., 960 mm for 1960–2010; not shown). Figure 2a shows the cloud shield from geostationary satellite infrared brightness temperature observations and accumulated precipitation from GPM IMERG during the 12 h with maximum rainfall (1600 UTC 20 July 2008–0400 UTC 21 July 2008). The observed MCS exhibits a T-shaped cloud shield with a deep convective core in the south and a northward extended stratiform region that covers the entire area along the eastern slope of the TP. The spatial extent of precipitation approximately followed the basin boundaries and the maximum precipitation occurred along the steep slopes of the TP, which suggests that the slopes played a crucial role in precipitation formation, for example, through orographic lifting or turbulence generation as a consequence of stratification.
Summary of observed MCS characteristics. (a) Satellite retrieved mean brightness temperatures (K) and (b) accumulated precipitation (mm) from GPM IMERG during the 12 h with maximum rainfall and cloud shield extent (1600 UTC 20 Jul 2008–0400 UTC 21 Jul 2008). The orange lines indicate the tracks of the cloud shield as tracked by Prein et al. (2021) in (a) and the vortex as tracked by Curio et al. (2019) in (b). The time evolution is shown by (c) the daily accumulated precipitation averaged over all gauge stations of the Sichuan basin compared to the daily accumulations of corresponding grid cells in GPM IMERG and (d) the hourly cloud shield area and its associated precipitation volume. The time evolution of the identified maximum in relative vorticity is also shown for comparison [black line in (c)].
Citation: Journal of Climate 36, 17; 10.1175/JCLI-D-22-0240.1
The downstream MCS formation occurred in conjunction with an eastward-moving TPV. The orange lines in Fig. 2 mark the trajectories of the cloud shield tracked according to the criteria described in section 2d (Fig. 2a) and the vortex that was tracked using the tracking algorithm presented in Curio et al. (2019) but based on ERA5 instead of ERA-Interim (Fig. 2b). While the vortex prevailed for more than 72 h (from 18–21 July 2008), the MCS had a shorter lifetime (in total, 42 h) and was first detected when the TPV moved off the TP (Figs. 2c,d). The TPV intensified when moving off the TP, and the identified maximum in relative vorticity of the TPV (black line in Fig. 2c) coincided with the observed maximum in daily precipitation.
The temporal evolution of daily and hourly precipitation during the event is shown in Figs. 2c and 2d. The peak of daily accumulated precipitation averaged over all gauges coincides well with the peak of daily accumulated precipitation averaged over GPM grid cells closest to the station locations (Fig. 2c). The same peak is also reflected in the identified maximum relative vorticity value associated with the TPV (Fig. 2c), with the maximum extent of the MCS cloud shield and maximum total precipitation volume under the tracked cloud shield (Fig. 2d).
b. Simulated precipitation
The spatial pattern of accumulated precipitation in the hours before and after the precipitation peak on 20 July looks similar in different observational datasets. Figures 3a–e show that GPM IMERG, CMORPH (Joyce et al. 2004), Multi-Source Weighted-Ensemble Precipitation (MWSEP; Beck et al. 2017), Climate Hazards Group Infrared Precipitation with Station Data (CHIRPS; Funk et al. 2015), and even ERA5 agree well on the precipitation center of the MCS in the Sichuan basin (within the black 1000-m contour line, between 27° and 35°N).
Accumulated precipitation during 19–21 Jul 2008 for observations and simulations. (a)–(f) Observations from gauge stations (with accumulations > 50 mm), GPM IMERG, CMORPH, CHIRPS, MWSEP, and ERA5, as well as the 12-km nest for WRF. (g)–(x) Simulated precipitation for experiments with varying model systems, domain sizes, initialization times, and nudging. (y)–(S) WRF ensemble members with varying physics options. The white line marks the 3000-m contour of the TP, and the black line marks the 1000-m contour.
Citation: Journal of Climate 36, 17; 10.1175/JCLI-D-22-0240.1
In contrast, most of the simulations do not capture the observed location of the precipitation center and show, instead, a displacement of precipitation toward the south, especially in the simulations of the WRF ensemble (Fig. 3). The simulations that are most similar to the observations are ICON2.6.1 (Fig. 3r), ICON2.6.3 (Fig. 3s), WRFnudging (Fig. 3k), and MPASLA (Fig. 3u) simulations. The differences between WRF with varying physics options are relatively small and all show the erroneous location of heavy precipitation southeast of the Sichuan basin (Figs. 3y–S). Although these results indicate that the simulated precipitation is not significantly affected by the changing physics, the fact that none of the physics members shows the MCS case at the correct location limits our capability to draw any conclusions about the impact of different physics options on MCS characteristics. The WRF simulation that was directly downscaled from ERA5 on D1 (WRFD1-direct) does not show any area with enhanced precipitation, but the WRF simulation on the relatively larger domain (WRFD2) that was directly downscaled from ERA5 performs better than WRFD1 (Figs. 3g,h). The additional simulation with WRFD3 (Fig. 3i) is similar to WRFD2, and WRFtimelag does not show any obvious improvements compared to WRFD2 or the WRF simulations on D1. We also notice that increasing the vertical grid resolution (WRF8km-vertical) does not improve WRF8km (Figs. 3l,m).
We computed FSSs based on hourly rain rates on 20–21 July for thresholds of 1, 3, 5, and 8 mm h−1. Figures 4a–d show the FSSs of different WRF experiments (WRFnudging, WRFD1, WRFD2, WRFD3, WRFtimelag, WRF8km, and WRF8km-vertical), and Figs. 4e–h show the FSS for the different models (and WRFnudging for comparison). The model spread of the WRF multiphysics ensemble is visualized by the gray shading in all panels in Fig. 4.
Median FSSs of hourly precipitation fields during the event 20 and 21 Jul 2008 compared with the reference observation field GPM IMERG for (a)–(d) different WRF experiments compared to (e)–(h) different regional climate models. The gray shading indicates the spread of the WRF multiphysics ensemble.
Citation: Journal of Climate 36, 17; 10.1175/JCLI-D-22-0240.1
In general, the FSS is strongly dependent on the precipitation threshold and the model used. A notable aspect that could not clearly be seen in the spatial pattern of accumulated precipitation (Fig. 3) is that the FSS increases with increasing domain sizes (from D1 to D3), whereby the difference between D1 and D3 is almost as large as the spread of the entire WRF multiphysics ensemble for higher rain rates. This means that WRFD3 yields an improved performance in simulating moderate to heavy precipitation at various spatial scales during the time period of the MCS event. However, WRFD3 did not show any improvement in simulating the timing and extent of MCS accumulated precipitation in Fig. 3i, and the improved model skill is not necessarily valid for precipitation on climatological time scales. Sensitivities to the domain size and the used modeling system are higher than the sensitivities to different physics options and initialization times. However, the sensitivities of physics options for other meteorological fields than hourly precipitation can be higher and similar to the sensitivities in domain size, as shown and discussed in more detail in Prein et al. (2023). The member with the highest skill from the original WRF ensemble is WRFD2, which, therefore, will be included in the following in-depth analysis as a reference run to which the additional WRF experiments are compared. WRFtimelag slightly improves WRFD2, but only for lower rain rates (Fig. 4a), which is surprising and hints at spinup issues or internal variability that hinders the simulation from being effectively constrained. WRF8km shows a better performance than WRF8km-vertical despite the increased vertical grid resolution in WRF8km-vertical (Figs. 4a–d).
The simulation with the highest skill is ICON2.6.3, followed by WRFnudging and ICON2.6.1. WRFnudging outperforms all other WRF simulations for all precipitation thresholds and is only exceeded by ICON2.6.3 for all thresholds and by ICON2.6.1 for 8 mm h−1 (Fig. 4d). The spread between different models is relatively low for the 1 mm h−1 threshold (Fig. 4e) but increases for larger rain rate thresholds (Figs. 4f–h). The sensitivity to WRF physics shows, in contrast, a large spread for the 1 mm h−1 threshold but a much smaller spread for higher precipitation thresholds. Note also that all model systems except for RegCM perform better than the WRF ensemble run on D1.
As the FSS increases for larger scales by definition, one important aspect to look at is the minimum scale at which the model provides an acceptable skill. This is reflected by the intersection of the FSS with FSSuniform, which Roberts and Lean (2008) defined as 0.5 + FSSrandom/2, with FSSrandom referring to the fraction of values above the threshold in the reference dataset (Fig. 4). For WRFnudging and ICON 2.6.3, the minimum spatial scale for an acceptable simulation lies at approximately 200 km for rain rates above 3 and 5 mm h−1, which corresponds to the spatial extent of typical MCS-associated precipitation (Houze 2004). That the minimum scales for WRFD3 and WRFnudging are more than twice as low as those for the rest of the WRF ensemble confirms the obvious improvement of the simulation through spectral nudging and increasing domain size. At rain rates greater than 8 mm h−1, the differences between the models become largest. Here, CCLM and both ICON simulations outperform WRFnudging for scales > 500 km, indicating that large-scale precipitation with higher rain rates is better captured in these model systems compared to WRF, even when the large-scale circulation in WRF is constraint.
c. Cloud shield characteristics
To gain a more complete picture of the simulated MCS case besides precipitation characteristics, we compare the simulated cloud tracks to observations. A mesoscale cloud shield formed in some simulations, for example, in WRFD2 (Fig. 5a). However, the cloud shield in WRFD2 develops to the south of the Sichuan basin, whereas the simulated cloud in WRFnudging looks more similar to the observed cloud shield (Fig. 2a) and better corresponds to the location of observed precipitation (Fig. 2b).
(a)–(f) Simulated cloud shields in different modeling systems at 2200 UTC 20 Jul 2008 (when observed cloud shield reaches its maximum extent). (g) The boxplots show distributions of the tracked cloud extent grouped by the multiphysics WRF ensemble, the multimodel ensemble, and different domain sizes to highlight the respective sensitivities. (h) The line plots show the PDFs of simulated Tb in the Sichuan basin during the simulation period compared to observations.
Citation: Journal of Climate 36, 17; 10.1175/JCLI-D-22-0240.1
The simulations with RegCM and CCLM stand out owing to their very large, contiguous cloud features < 241 K (Figs. 5c,d), while other simulations do not show any larger cloud features that indicate the presence of an MCS (Figs. 5e,f). This explains the greater variation in simulated cloud shield extent within the multimodel ensemble, as compared to, for example, the WRF multiphysics ensemble, which, again, exhibits a narrow spread between the simulations (Fig. 5g). Compared to the MCS track in satellite observations, the simulated cloud shield extents are generally smaller (Fig. 5g). Looking at the general cloudiness produced by the different models in the Sichuan Basin during the simulation period (Fig. 5h), it can be seen that both RegCM and CCLM exhibit a different Tb distribution compared to most other simulations that show a higher peak in Tb values > 280 K and lower frequencies for Tb values < 240 K. The spread of the WRF ensemble is relatively small and the WRF simulations have generally higher brightness temperatures, meaning fewer or lower clouds than in the observations (Fig. 5h). Part of the differences in the brightness temperature PDF between observations and simulations in Fig. 5h can be explained by the higher effective resolution in the simulations that expectedly lead to higher frequencies of clear-sky brightness temperature values (as in MPAS, CCLM, and RegCM) and higher extreme values (as in MPAS and all WRF simulations).
It is worth pointing out that none of the WRF simulations except for WRFnudging fulfills the MCS criteria from Feng et al. (2021b) to be classified as an MCS, because the tracked clouds do not produce enough precipitation, while the observed cloud shield clearly does (data not shown).
The difference in the cloud shield location between WRFD2 and WRFnudging leads not only to the displacement of precipitation but also to a difference in the total precipitation amount, which is substantially smaller in WRFD2 compared to WRFnudging, even though the more southerly located precipitation is formed in a region where water vapor is much more abundant. The role of the general moisture supply in the Sichuan basin and differences in moisture advection between the two simulations are, therefore, examined in more detail in the next section.
d. Water vapor transport
Since precipitation is clearly underestimated even if clouds form in many of the ensemble members, the question arises of if there is enough local moisture supply in the simulations to feed a convective system. Figures 6a and 6b show that moisture in ERA5 is mainly transported from the southwest and that there is a clear peak in the northward water vapor flux toward the basin during the event (20–21 July). The Hovmöller diagrams of the vertically integrated northward water vapor flux averaged over longitudes 100°–108°E show the same pattern for ERA5 and WRFnudging (Figs. 6a,c) but look different for WRFD2 and WRFD3 (Figs. 6e–g). Even if there is a small peak in northward water vapor flux around 20 July, little moisture is transported beyond the approximate boundary of the basin at ∼28°N (indicated by the dashed line) during a shorter time period compared to ERA5 and WRFnudging. Consequently, the total water vapor availability that can potentially converge in the Sichuan basin is much more limited during the time of the event (Figs. 6f,h). The moisture from the south (usually from the Bay of Bengal) is transported farther toward the east, where the WRFD2 and WRFD3 exhibit a maximum in accumulated precipitation that cannot be seen in the observations (Figs. 3h,i).
Hovmöller (latitude–time) diagrams of (a),(c),(e),(g) northward vertically integrated water vapor flux (kg m−1 s−1) and (b),(d),(f),(h) total mean water vapor transport (IVT) during the event (kg m−1 s−1), where
Citation: Journal of Climate 36, 17; 10.1175/JCLI-D-22-0240.1
Another feature that is more pronounced in WRFnudging and ERA5 than in the other two simulations is the velocity couplet between 28° and 34°N during the days of the event (Figs. 6a,c). This couplet is linked to the rotation of vortex and indicates the passage of the TPV through the basin over time. Hence, two important aspects about the MCS formation can be learned from Fig. 6: 1) WRFnudging gives the best performance in simulating the observed precipitation structure, because it best represents the water vapor transport from the south into the basin; and 2) the couplet indicates rotation with a strong northward water vapor transport into the basin as the vortex enters the basin and southward water vapor transport as the vortex moves farther eastward (left side of the vortex). The presence of the rotating vortex explains why the enhanced water vapor supply leads to precipitation in this case, because horizontal water vapor flux convergence can be caused by rotation around the low pressure center (inward turning wind vectors) and because the flux component in one direction (here, the northward water vapor flux) is stronger than the other.
Under the hypothesis that the TPV is needed to transport water vapor into the basin from the south and to trigger water vapor flux convergence in the downstream region, we further investigate how well the vortex evolution over the TP is represented in the different modeling systems.
e. Vortex evolution
The observed and simulated vortex evolutions are visualized in the form of Hovmöller diagrams of the spectrally filtered relative vorticity fields in Fig. 7. The fields are averaged over latitudes 33°–36°N because the observed TPV was tracked within this latitude segment in ERA5 (Fig. 2) and because TPVs typically occur in this region (Curio et al. 2019). A few vortices are clearly distinguishable as coherent streaks of positive vorticity whose movement can be followed over time. For example, a vortex formed at around 85°E on 14 July and moved toward the eastern edge of the TP (Figs. 7a–f). Interestingly, this vortex is well represented in all of the shown simulations. While the formation of the TPV on 18 July is likewise present in the simulations, the vortex deflects toward the south (not shown) and, therefore, disappears in the shown Hovmöller diagrams by the end of 19 July in WRFD1, WRFD2, and WRFD3 (Figs. 7b–d). The sensitivity of the simulation to domain size becomes apparent again, since the vortex simulation in WRFD2 and WRFD3 is much closer to ERA5 as compared to WRFD1.
Hovmöller (time–longitude) diagrams of 3-hourly, spectrally filtered relative vorticity (×10−5 s−1) averaged over latitudes 33°–36°N for (a) ERA5 compared to (b)–(f) model simulations. The dashed lines mark the western and eastern boundaries of the TP.
Citation: Journal of Climate 36, 17; 10.1175/JCLI-D-22-0240.1
Furthermore, Fig. 7e shows the advantage of WRFnudging being more similar to ERA5 through the assimilated large-scale forcing. The tracks in WRFnudging, ICON2.6.1, and ICON2.6.3 are most similar to the observed TPV track and, most importantly, they persist until the TPV moves off the TP at around 102°N. This, together with the fact that these three simulations significantly showed the best performance for the simulation of the precipitation center, confirms the hypothesis that the vortex evolution is crucial for water vapor inflow and formation of heavy precipitation in the Sichuan basin during this case.
In general, the differences in relative vorticity among the modeling systems are quite large. While RegCM and ICON 2.6.3 show strong coherent vortices with fewer fine-scale structures, MPAS and ICON2.6.1 exhibit weaker signals in the vorticity field (Figs. 7g,i). The evolution of the vortices that formed 14 and 18 July are well captured in MPAS and the two ICON simulations. However, the CCLM simulation does not show any distinguishable vortex track; instead, it shows a more stationary vortex that does not move off the TP during the simulation period (Fig. 7h).
f. The role of the westerly jet stream
Figure 7 shows that the vortex also forms in simulations that do not accurately represent its further evolution and linkage to moisture advection into the Sichuan basin. Therefore, the main question is what factors control the dissipation or deflection of the vortex in the different simulations, rather than the genesis itself.
Since the vortex is a mesoscale weather system, it is only indirectly assimilated through the nudging process that focuses on wavelengths between 1000 and 2000 km (section 2a). In the preceding section, we showed that WRFnudging improves the simulated vortex, in addition to the simulated water vapor advection and downstream precipitation. This improvement suggests that the large-scale forcing plays a crucial role for the vortex evolution. The fact that the simulation of this specific MCS case is more sensitive to domain size than to the physics options also supports this hypothesis because the domain size and the position of the lateral boundaries are more directly linked to the atmospheric circulation than the different parameterization schemes are. Furthermore, it is plausible to assume that the prevailing westerly background flow controls, at least to some extent, if TPVs move off and impact the downstream weather, because Curio et al. (2019) found that the total traveled distance of TPVs is dependent on the intensity and position of the westerly jet (Curio et al. 2019). In this section, we investigate, therefore, the influence of the upper-level wind circulation on the vortex evolution over the TP.
Figure 8 shows the upper-level zonal wind circulation for the 12 h during which the vortex disappears in WRFD2 and WRFD3. The position of the westerly jet stream is indicated by the white line, which corresponds to the 20 m s−1 contour in the 200-hPa zonal wind field (∼80th percentile of all zonal wind speeds at 200 hPa; e.g., Hunt et al. 2018). The position of the low pressure center associated with the vortex is indicated by the 500-hPa geopotential anomaly (blue contours) for the same 12 h with respect to the entire simulation period. For context, the different domain sizes are shown in relation to the mean westerly jet stream position during the simulation period in Fig. 8a. The figure reveals that the location of D1 is not optimal, as its northern boundary intersects the maximum of the jet. Several studies have shown that the position of lateral boundaries in limited-area modeling can strongly affect the simulated weather systems, for example, through structural distortions in the large-scale flow near the boundaries (Hashimoto et al. 2016; Imberger et al. 2020). It is, therefore, preferable to locate the boundaries far away from any systems that represent a strong external forcing to the simulated weather, especially when the update frequency of boundary conditions is relatively low (Gustafsson 1990). The domains D2 and D3 of this study include the extent of the westerly jet, meaning that boundary effects are less likely to have occurred in simulations with these domains.
Mean zonal wind at 200 hPa (m s−1) during 1200 UTC 19 Jul 2008–2300 UTC 19 Jul 2008 for (b) ERA5 compared to (c)–(h) simulations. The white lines represent the 20 m s−1 contour for the zonal wind at 200 hPa and indicate the approximate jet stream position. The blue contours refer to the 500-hPa geopotential anomaly with respect to the entire simulation period (m). The domain sizes are shown in relation to the mean jet stream position in ERA5 for context in (a).
Citation: Journal of Climate 36, 17; 10.1175/JCLI-D-22-0240.1
While RegCM, WRFnudging, and the two ICON simulations show the same jet stream position and the same perturbation around 95°E as ERA5, the other simulations show a different spatial pattern (Figs. 8c–i). The same is true for the intensity and extent of the 500-hPa geopotential anomaly. CCLM and MPAS stand out as two simulations with only little structure and small-scale variation in the jet stream (Figs. 8h,i). Again, the overall differences between the models are striking and seem consistent with their performance in simulating the MCS precipitation center (see section 3b). It should be noted that the differences in upper-level zonal wind circulation between the regional models could potentially stem from different damping options at the top of the model domain. Because orographic flow can induce vertically propagating waves (i.e., gravity waves) in a nonhydrostatic model configuration, one can use a damping layer at the top of the model domain that reduces spurious reflections at the upper boundary in simulations that do not include large parts of the stratosphere (e.g., Klemp et al. 2008). The solution of velocity fields at the upper levels of the model is, therefore, influenced by the applied damping method, and the vertical extent of the damping layer is important for adjustment of nested model pressure to that of initial boundary conditions within the model domain. In the particular case of the TP, Rayleigh damping may, for instance, limit the vertical extent of TPVs (see Fig. 10) and influence the structure of the jet stream.
We notice, however, that differences in the jet structure also exist between different WRF simulations that use the same damping option (Rayleigh damping with a damping depth of 5000 m from the model top at 25 hPa). The WRF simulation with the poorest performance for the vortex track and associated precipitation (WRFD1) shows the most different spatial pattern of the westerly jet stream, as evidenced by an early break off in the jet stream upstream of the geopotential anomaly (Fig. 8c). WRFD2 and WRFD3 also exhibit differences in the simulated jet, whereby the vortex seems to move faster in WRFD2 and dissipate earlier in WRFD3. This is a manifestation of the domain-size sensitivity and shows that the domain size does not only affect the simulated precipitation (Figs. 3, 4), cloud (Fig. 5), and vorticity fields (Fig. 7) but also the evolution of the large-scale circulation. This finding also suggests that it is favorable to include the extent of the jet stream in this simulation instead of cutting through it to assure that mesoscale processes remain coupled to the large-scale forcing at the lateral boundaries. The differences between WRFD2 and WRFD3 show that a too-large domain (here, WRFD3) can decrease the model skill in representing the large-scale circulation again, in particular when nudging is not applied. This was also found by Jones et al. (1995) and Prein et al. (2019), who highlight that successful regional climate model simulations require a domain that is large enough to allow mesoscale features to evolve freely but small enough to avoid the large-scale circulation of the fine-scale model departing from the driving model.
Following the observed vortex evolution over time (Figs. 2b,c, and 7), we see that the vortex enters the Sichuan basin on 20 July. Figure 9 shows the location of the westerly jet stream averaged over the 12 h when the vortex moves off the TP. The westerly jet stream not only affects the vortex intensity but also influences where and when the vortex enters the downstream regions. While the large-scale pattern in RegCM was very similar to ERA5 in the beginning of the vortex evolution (Fig. 8c), the geopotential anomaly in RegCM grew much larger and stronger and covered a large area downstream that extended too far north (Fig. 9f). Hence, RegCM shows the opposite effect of what we see in the WRF ensemble: the simulated vortex is stronger compared to ERA5 (Fig. 7f) and deflects northward instead of southward (Fig. 9f). This explains this model’s dislocation of the precipitation center to the north (Fig. 3n).
As in Fig. 8, but for 1200 UTC 20 Jul 2008–2300 UTC 20 Jul 2008, when the vortex moves off the TP.
Citation: Journal of Climate 36, 17; 10.1175/JCLI-D-22-0240.1
In contrast, the jet stream exhibits a southward deflection in WRFD1, which also transports the low pressure center more toward the south of the Sichuan basin (Fig. 9e). The same southward component can be seen in WRFD2 and WRFD3 (Figs. 9f,g), where the location and intensity of the geopotential anomaly fit well with the location of the simulated precipitation (Fig. 3).
As was the case for the water vapor transport, WRFnudging clearly results in the most accurate representation of the jet stream and vortex evolution among the WRF simulations (Fig. 9e). While MPAS simulates the low pressure center at the right location, the jet stream structure looks substantially different from the jet stream structure in ERA5 (Fig. 9g). The two ICON simulations perform comparatively well and show the right location of the low pressure center as well as the same break off in the jet stream as ERA5, when the vortex enters the Sichuan Basin (Figs. 9i,j). Figures 9f and 9j reveal that ICON2.6.3 captures the right extent and location of the geopotential anomaly while the low pressure center in RegCM is too far north and too strong. Although the vortex evolution of RegCM and ICON2.6.3 looked similar in Fig. 7, we can see here that the jet structure and the intensity and location where the vortex moves off the TP are significantly different, which explains the different performance in simulating the MCS precipitation center (Fig. 3). One reason for the good performance of ICON2.6.3 could be the considerably higher update frequency of the lateral boundary conditions compared to most other simulations (hourly instead of 6-hourly; Table 1). Frequently updating the lateral boundary conditions is another way of constraining the large-scale forcing in a stricter manner, even though the large-scale circulation within the model domain can still freely evolve.
The representation of the westerly jet stream as large-scale forcing is thus an important factor for the vortex evolution and the consequent water vapor inflow and precipitation formation downstream. These findings highlight the importance of correctly capturing the large-scale forcing in this specific case. Similar results have also been found by Coppola et al. (2020), who used a comparable ensemble-based approach to simulate convective phenomena at short time scales and found a large spread between different regional models. The model spread decreased when the large-scale forcing was stronger, and convective events that were mainly large-scale driven could even be reproduced in simulations that started one month before the event (Coppola et al. 2020). Our results and the findings of Coppola et al. (2020) imply that there is value in understanding to which degree convective events in a specific region are driven by large-scale factors, since these might significantly affect the uncertainties in modeling representatives of such events in climate simulations.
g. Vertical cross sections
A view into the vertical structure of the observed and simulated event helps us better understand the relationship between the TPV and the jet stream, because TPVs usually occur around 500 hPa, while the jet stream reaches its maximum at around 200 hPa. Figures 10 and 11 show vertical longitudinal cross sections of the zonal and meridional wind components as well as vertical velocities at the latitudes of the MCS during the day of TPV formation over the plateau (Fig. 10); during the first day, the MCS was observable in the Sichuan basin (Fig. 11).
Vertical, longitudinal cross sections of wind components and vertical velocity averaged over the latitudes of the MCS (32°–36°N) and averaged for time period 1200 UTC 19 Jul 2008–2300 UTC 19 Jul 2008.
Citation: Journal of Climate 36, 17; 10.1175/JCLI-D-22-0240.1
As in Fig. 10, but for 1200 UTC 20 Jul 2008–2300 UTC 20 Jul 2008, when the vortex moves off the TP.
Citation: Journal of Climate 36, 17; 10.1175/JCLI-D-22-0240.1
It is discernable that ERA5 and WRFnudging show a similar vertical structure of horizontal and vertical velocities, while WRFD2, WRF8km, and, in particular, WRF8km-vertical look different from ERA5. First, the upper-level zonal wind and jet stream show a more wavy pattern in ERA5 and WRFnudging (Figs. 10a–e). There are easterly winds in the lower boundary layer in WRFD2, WRF8km, and WRF8km-vertical that are strongest in WRF8km-vertical and almost absent in ERA5 and WRFnudging. Second, the velocity couplet of meridional wind located over the central TP between 500 and 300 hPa indicates that the vortex in ERA5 and WRFnudging has similar horizontal and vertical dimensions. In the other WRF simulations, the rotation associated with the vortex is less evident and the vertical structure of meridional winds looks different from ERA5 (Figs. 10a–e).
While brightness temperatures <220 K and rain rates >100 mm day−1 (Figs. 2a,b) clearly hint at the convective nature of the event, the cloud and precipitation fields only provide a view from the top. Figures 11k and 11l show the vertical extent of updrafts east of the TP reached up to 200 hPa during the hours when the MCS was observed in the Sichuan Basin. The updrafts show similar characteristics as the updrafts over the central TP at the time of TPV formation (Figs. 10k–o). In fact, the updrafts exhibit a typical pattern for cyclonic vortices with moderate vertical velocities in the center of the vortex stretching over a horizontal area that is significantly smaller than the horizontal area of the vortex itself (e.g., Figs. 11k,l). This indicates that the MCS observed in the Sichuan basin is linked to the same mechanism as the TPV and can be interpreted as a further development of the latter. The updrafts over the TP reach up to almost 100 hPa (∼18 km MSL), which is typical during summer when thermal mixing and turbulence diffusion causes the heat source over the central TP to extend into the upper troposphere (Zhang et al. 2022). However, these updrafts are not associated with extensive clouds and heavy precipitation until the vortex moves off the TP. When the TPV enters the Sichuan basin, the updrafts intensify significantly and extend down to the ground in ERA5 and WRFnudging (Figs. 11k,l). This is likely because of column stretching and vortex twisting (due to the conservation of potential vorticity), as well as convective instability associated with the cold low pressure center. At the lower elevations, warm and moist air can be transported upward so that clouds and heavy precipitation are generated. The vertical and horizontal wind fields (Figs. 10, 11), the coincidence of maximum vorticity and precipitation (Fig. 2c), and the moisture fluxes into the Sichuan basin (Fig. 6) evidence that this particular MCS case is a modification of the TPV entering a moisture-abundant, warmer region.
At the time and location of the observed MCS, the updrafts are absent in WRFD2, WRF8km, and WRF8km-vertical, although WRFD2 and WRF8km simulated updrafts over the TP during the precedent day when the vortex was formed (Figs. 11m–o). Interestingly, the vertical structure of zonal wind and vertical velocity is most different in WRF8km-vertical when compared to ERA5. This suggests that the increased vertical resolution enhances small-scale processes in the boundary layer over the TP that potentially feed back to the upper-level, large-scale circulation in a way that the vortex evolution and subsequent precipitation are suppressed in this case. However, while the vertical cross sections reveal significant differences in zonal, meridional, and vertical velocity between WRFnudging and the other WRF simulations, it is not clear how these differences are linked to the formation and transport of the TPV. A detailed investigation of the causes for the reversed zonal wind around 500 hPa in the WRF simulations (e.g., Fig. 10c) remains for future studies. In particular, it will be relevant to determine if biases in the boundary layer winds lead to systematic errors in kilometer-scale climate simulations over the TP.
4. Are there systematic biases in the simulation of MCSs in the Sichuan basin?
The presented case evaluation raises the question of how generalizable this specific MCS case is and if there are systematic biases in simulating MCSs in the Sichuan basin. To address this question, we complement our analysis with an evaluation of a 1-yr WRF simulation with 4-km grid spacing for the water year 2020 (see section 2a). Torrential rainfall led to several devastating floods in the Sichuan and Yangtze River basins east of the TP during the 2020 summer season (e.g., Xia et al. 2021), which lets us assume that there was a high MCS activity in this region. We apply an MCS tracking algorithm with the criteria described in section 2d for the 1-yr WRF simulation output regridded to the GPM IMERG grid and compare the simulated MCSs to those tracked in satellite observations (based on GPM IMERG and NCEP/CPC brightness temperatures; see section 2d).
Figure 12 shows the total number of MCSs in each month for the entire domain D2 (Fig. 12a) and for the Sichuan basin (Fig. 12b), here defined as tracks that traverse the region 28°–33°N, 102°–105°N (blue box including SI1 and SI2 in Figs. 12c–e) during their lifetime. The seasonal cycle of simulated MCSs agrees well with the seasonal cycle of observed MCSs, both in terms of total numbers and phase (Figs. 12a,b). However, WRFD2 slightly overestimates the total MCS number, while WRFnudging slightly underestimates the total MCS number (Fig. 12a). The peak season of MCSs in the entire domain and in the Sichuan basin is between June and August, with a sharp increase in MCSs from May to June. In April, May, and September, the total numbers of simulated MCSs in the Sichuan basin show larger differences compared to observations. However, these differences are not surprising given that the total number of MCSs is small in this limited region for one season and might be attributed to model internal variability and the tracking algorithm, because the latter can be sensitive to the chosen thresholds, especially for smaller systems that have horizontal extents close to the minimum required area.
Observed vs simulated (WRFnudging, WRFD2) MCS characteristics for the water year 2020 (October 2019–September 2020). Seasonal cycles are shown for (a) all MCSs tracked in domain D2 and (b) MCSs that pass the Sichuan basin [blue box in (c) and (d)] during their lifetime. (c),(d) The observed and simulated MCS frequency between June and August. The MCS frequency refers to the number of times an MCS occurs in a grid cell, meaning that both the lifetime and the area of the MCS affect this number. The black contour line marks the 3000-m-elevation contour, and the boxes depict the subregions used to derive the spatial averages in Table 3.
Citation: Journal of Climate 36, 17; 10.1175/JCLI-D-22-0240.1
Figures 12c–e depict the MCS frequency calculated as the total number of times an MCS is identified in each grid cell. These numbers are thus affected by both the lifetime and area of the MCSs and reflect how often MCSs occur at a given location. The smaller MCS numbers of the 1-yr simulation with WRFnudging for the whole the domain D2 compared to observations and WRFD2 (Fig. 12a) is most likely linked to the significantly lower MCS frequencies in WRFnudging south of the Himalayas (Figs. 12c,d; Table 2). Both WRF simulations exhibit an underestimation of MCSs over the TP and an overestimation of MCSs south of the Sichuan Basin, but these biases are less pronounced in WRFnudging compared to WRFD2 (Figs. 12c–e; Table 2). This is consistent with the case study and suggests that the biases are linked to different mechanisms for MCS formation in the lee of the TP. More specifically, the underestimation of MCSs over the TP, which is slightly improved through nudging, could be related to an underestimation of TPV-associated MCSs. In contrast, the overestimation of MCSs to the south of the Sichuan basin could indicate a systematic overestimation of MCSs that originate from other processes than TPVs, for example, from local convection. Alternatively, these could also indicate biases in the simulated TPV trajectories, that is, when TPVs move farther into the southeastern regions of the TP (see case study discussed in section 3). It will be relevant for future climate simulations to investigate the underlying mechanisms of MCSs in different subregions, because both the climate change response of MCSs and the ability of climate models to capture this response will depend on the underlying mechanisms.
Spatial means of observed vs simulated MCS frequencies for June–August 2020 over and around the TP. The spatial extents of the respective subregions are shown in Fig. 12.
A realistic representation of the seasonal and spatial patterns of MCSs is crucial for simulating the hydrological cycle and its extremes over the TP and its surrounding areas. As shown in Figs. 13a–c, the spatial pattern of summer rainfall in 2020 is reasonably well simulated by WRF compared to GPM IMERG, and most of the regions with high rainfall (i.e., in the southern and eastern catchments of the TP) are characterized by large amounts of MCS-associated precipitation (Figs. 13d–f). The more detailed structure of mean precipitation in both WRF simulations (Figs. 12b,c), especially over and along the 3000-m contour can be explained by the higher spatial resolution of topography in WRF and the comparatively low effective resolution of GPM IMERG in mountainous regions (Guilloteau et al. 2017).
(a)–(c) Observed vs simulated (WRFnudging, WRFD2) mean summer precipitation and (d)–(f) the contribution of MCS-related precipitation to total summer precipitation for June–August 2020. The total precipitation amount under the tracked 240-K cloud shield is considered to be produced by the MCS. (g),(h) The differences between observed and simulated MCS precipitation are shown. Blue values indicate higher values in WRF and red values indicate higher values for GPM. (i) The difference in the simulated MCS precipitation between WRFnudging and WRFD2. The black line marks the 3000-m-elevation contour.
Citation: Journal of Climate 36, 17; 10.1175/JCLI-D-22-0240.1
The regions in which downstream MCSs produce most summer rainfall (>5 mm day−1 on average) are generally well captured in the simulations (Figs. 13d–f) and match also with the main regions of MCS rainfall on decadal time scales (e.g., Kukulies et al. 2021; Feng et al. 2021b). One difference that can be noticed is the high MCS-associated rainfall over eastern China that is visible in the observations and much less pronounced in the WRF simulations (Fig. 13c). This needs some further investigation but could be related to the location of the eastern domain boundary that was not designed to optimize the representation of the East Asian summer monsoon component. In fact, the largest differences in total and MCS associated summer precipitation can be found along the lateral boundaries of our domain (e.g., in western India or over the Bay of Bengal). This is even true for WRFnudging and highlights, again, the importance of the domain setup.
Table 3 shows the spatial averages of total and MCS-associated accumulated summer precipitation averaged over the domain and different subregions. Although WRFnudging reduced the biases in MCS frequency over the TP and south of the Sichuan basin (Table 2), WRFnudging shows higher biases in MCS-associated precipitation than WRFD2 in most subregions (Fig. 13; Table 3). In addition, the domain-averaged total and MCS-associated precipitation in WRFD2 is closer to the observations than WRFnudging (Table 3). We conclude that there are systematic biases in MCS frequencies over, as well as east and south of, the TP. Some of these biases are reduced when using spectral nudging, which is consistent with the results of the case study. It is, however, difficult to determine whether WRFnudging improves the 1-yr simulation, because the simulation shows also higher biases in total and MCS-associated precipitation in other subregions. This reflects a more general challenge in regional climate modeling and highlights that it is important to define the added value of a given simulation based on multiple parameters.
Spatial mean values of observed vs simulated accumulated precipitation for June–August 2020 over and around the TP. The spatial extents of the respective subregions are shown in Fig. 12.
Finally, we show the simulated water vapor transport into the Sichuan basin as water vapor fluxes from the south were found to be associated with the MCS in the lee of the TP (Fig. 6, section 3d). Since the water vapor transport from the south was not well captured in the WRF simulations that failed to generate the MCS, we investigate how this water vapor transport is captured in the 1-yr simulation. Figure 14 depicts the observed and simulated MCS locations weighted by the extent of their precipitating area on top of northward water vapor transport from ERA5 (upper panels) and the two 1-yr simulations with WRF (lower panels). Most observed and simulated MCSs in the Sichuan basin seem to occur during periods with enhanced water vapor transport from the south into the basin. Since this water vapor transport looks similar in ERA5, WRFD2, and WRFnudging, we conclude that WRFnudging does not show the same obvious improvement for the MCS climatology as it showed for the case, and that periods with enhanced water vapor transport into the Sichuan basin are not systematically missed by, for example, the potentially missing out of off-moving TPVs.
Hovmöller (time–latitude) diagrams of vertically integrated, northward water vapor transport at longitudes 100°–108° for summer months in 2020 in (a)–(c) ERA5, (d)–(f) WRFD2, and (g)–(i) WRFnudging. The white line indicates the southern boundary of the Sichuan basin. The white markers indicate the latitudinal locations of MCSs detected in the same longitude range at the given time points. The marker size is weighted by the size of the MCS-associated precipitating area. Note that the MCSs locations in (a)–(c) refer to the observed MCSs as tracked with GPM IMERG and NCEP/CPC, and those in (d)–(i) refer to the simulated MCSs.
Citation: Journal of Climate 36, 17; 10.1175/JCLI-D-22-0240.1
5. Conclusions
In this study, we evaluated a multimodel and multiphysics ensemble of kilometer-scale simulations over the TP by identifying and analyzing the key processes in the simulations that led to the presence or absence of MCS formation in the downstream-located Sichuan basin during July 2008. We demonstrate how a simulation ensemble with different models and model configurations can enhance our understanding of processes involved in MCS development in the Sichuan basin. The identified model sensitivities can help inform the design of longer-term simulations based on the gained physical understanding of how this example MCS case evolved.
The main conclusions from this case evaluation can be summarized as follows:
-
The analysis of vertical and horizontal wind fields as well as the coincidence of maximum of vorticity and precipitation indicate that the observed MCS case can be seen as a further development of a vortex moving off the TP. The preceding vortex evolution over the TP and water vapor fluxes into the Sichuan basin were identified as key processes that need to be correctly simulated to reproduce the MCS precipitation center at the right location. The interaction between off-moving vortices and enhanced water vapor fluxes into the Sichuan basin thus represents a mechanism to focus on when analyzing future simulations of downstream extreme precipitation.
-
Simulations with deviating vortex movements exhibited remarkable differences in the intensity and position of the westerly jet stream. This implies that a realistic representation of the westerly jet stream is necessary to correctly simulate the vortex and subsequent downstream precipitation.
-
The most notable impacts of model configurations on the simulation quality of precipitation and related processes were the following:
-
Large-scale forcing: While most of the model ensemble members showed either a dislocation or absence of precipitation in the Sichuan basin, the best performance was achieved by constraining the large-scale flow through spectral nudging using WRF and by ICON2.6.1 and ICON2.6.3, in which the used update frequency of lateral boundary conditions was higher compared to the other simulations.
-
Physics options, internal variability, and vertical resolution: The effect of the different physics options in the WRF ensemble on the simulated MCS case could not be investigated, because none of the members captured the MCS case. The results indicate, however, that changing the physics did not significantly change the precipitation field of the simulation, and the physics ensemble spread of FSSs for high precipitation thresholds was significantly smaller than the spread between different models. The different initialization times in WRF and MPAS had a small impact on the simulation quality. Increasing the vertical grid spacing did not improve the simulated MCS center and even deteriorated the FSS, most likely due to enhanced small-scale processes that led to a different vertical structure of zonal winds over the TP compared to ERA5 and the other simulations.
-
Model system: The jet stream structure and vortex evolution exhibited notable differences between the models we used, and these differences also were reflected in the large model spread of precipitation performances. Most WRF simulations showed, for example, southerly displaced vortices that weakened as they moved away from the jet maximum, as opposed to the RegCM simulation that showed a northerly displaced vortex resulting in an intensified low pressure system and an MCS too far in the north that was much stronger than the observed MCS. In all simulations, except for MPAS, it is a consistent finding that the simulation quality of the TPV and MCS is better when the jet stream structure is more similar to ERA5.
-
Domain size and location: Increasing the domain size of the WRF simulation improved the FSS of simulated precipitation and the spatial patterns of the MCS-associated cloud shield, vortex, water vapor transport, jet stream, and vertical structure of winds over the TP. Increasing the domain to an even larger size led to a higher FSS but did not improve the simulated MCS location and the aforementioned processes. The effect of the WRF domain size on the simulation quality is most likely linked to the location of the northern boundary rather than the total domain size. This is because the northern boundary of the smaller domain D1 cuts through the jet, whereas the larger domains allow a more realistic simulation of the jet within the WRF domain.
-
-
Results from a 1-yr WRF simulation with a 4-km grid spacing show that the shape and magnitude of the seasonal cycle of MCS numbers and the spatial pattern of MCS-associated precipitation are generally captured when compared with satellite observations. The simulation also captures periods of enhanced northward water vapor transport into the Sichuan basin, which suggests that the identified biases in water vapor transport, vortex evolution, and jet stream from the case study do not significantly affect the simulated precipitation and associated water vapor transport in the Sichuan basin in the 1-yr WRF simulation. However, there is a systematic overestimation of MCSs south of the Sichuan basin and an underestimation of MCSs over the TP. These biases are consistent with the case study and point to different MCS mechanisms being differently well captured by the model. This needs to be considered in future simulations because the representation of specific MCS types might affect the climate change response of MCSs downstream of the TP. The simulation with spectral nudging reduced the biases in MCS frequencies over the TP and south of the Sichuan basin, but it also showed higher biases in domain-averaged precipitation and a strong underestimation of MCSs south of the Himalayas. Thus, the improvements through WRFnudging suggested by the case study are not as obvious and consistent in the 1-yr simulation.
Future studies will focus on a more detailed analysis of the connection between TPVs and downstream MCS occurrence using the presented 1-yr-long WRF simulation as well as other simulations that will cover the same period and are currently in preparation. Decadal-long current and future climate kilometer-scale simulations will be performed in the upcoming years within the CPTP project, which will allow us to study large-scale forcing biases in kilometer-scale simulations on longer time scales as well as the interannual variability and climate change impacts on TPVs and MCSs over the TP region.
Acknowledgments.
Julia Kukulies was sponsored, in part, by NCAR’s Graduate Student Advanced Study Program. We acknowledge high-performance computing support from Cheyenne (https://doi.org/10.5065/D6RX99HX) provided by NCAR’s Computational and Information Systems Laboratory. NCAR is sponsored by the National Science Foundation under Cooperative Agreement 1852977. This work was also supported by the Swedish Research Council (VR; 2019-03954) and Swedish National Space Agency 608 (SNSA; 188/18 4). It is Contribution Number 9 to CORDEX-FPS-CPTP and Swedish MERGE. We acknowledge the modeling working group of the CORDEX-FPS-CPTP for providing and postprocessing a large part of the simulations included in the presented analysis. We also thank Andreas Will and the two anonymous reviewers for their comments and suggestions, which greatly improved this manuscript.
Data availability statement.
NCEP/CPC brightness temperatures can be downloaded from https://disc.gsfc.nasa.gov/datasets/GPM_MERGIR_V1/summary and GPM IMERG v06 can be downloaded from https://disc.gsfc.nasa.gov/datasets/GPM\_3IMERGHH\_06/summary or from https://gpm.nasa.gov/data/directory. The latest version of the hourly ERA5 reanalysis at pressure levels can be accessed through the Copernicus Climate Data Store (https://cds.climate.copernicus.eu/cdsapp#!/dataset/reanalysis-era5-pressure-levels?tab=overview). The code for the data analysis and MCS tracking can be found at https://github.com/JuliaKukulies/cptp.
REFERENCES
Bae, S. Y., S.-Y. Hong, and W.-K. Tao, 2019: Development of a single-moment cloud microphysics scheme with prognostic hail for the weather research and forecasting (WRF) model. Asia-Pac. J. Atmos. Sci., 55, 233–245, https://doi.org/10.1007/s13143-018-0066-3.
Ban, N., J. Schmidli, and C. Schär, 2014: Evaluation of the convection-resolving regional climate modeling approach in decade-long simulations. J. Geophys. Res. Atmos., 119, 7889–7907, https://doi.org/10.1002/2014JD021478.
Ban, N., and Coauthors, 2021: The first multi-model ensemble of regional climate simulations at kilometer-scale resolution, part I: Evaluation of precipitation. Climate Dyn., 57, 275–302, https://doi.org/10.1007/s00382-021-05708-w.
Beck, H. E., A. I. Van Dijk, V. Levizzani, J. Schellekens, D. G. Miralles, B. Martens, and A. De Roo, 2017: MSWEP: 3-hourly 0.25° global gridded precipitation (1979–2015) by merging gauge, satellite, and reanalysis data. Hydrol. Earth Syst. Sci., 21, 589–615, https://doi.org/10.5194/hess-21-589-2017.
Bibi, S., L. Wang, X. Li, J. Zhou, D. Chen, and T. Yao, 2018: Climatic and associated cryospheric, biospheric, and hydrological changes on the Tibetan Plateau: A review. Int. J. Climatol., 38, e1–e17, https://doi.org/10.1002/joc.5411.
Bougeault, P., and P. Lacarrere, 1989: Parameterization of orography-induced turbulence in a mesobeta–scale model. Mon. Wea. Rev., 117, 1872–1890, https://doi.org/10.1175/1520-0493(1989)117<1872:POOITI>2.0.CO;2.
Buzzi, A., A. Speranza, S. Tibaldi, and E. Tosi, 1987: A unified theory of orographic influences upon cyclogenesis. Meteor. Atmos. Phys., 36, 91–107, https://doi.org/10.1007/BF01045143.
Carlson, T., S. Benjamin, G. Forbes, and Y. Li, 1983: Elevated mixed layers in the regional severe storm environment: Conceptual model and case studies. Mon. Wea. Rev., 111, 1453–1474, https://doi.org/10.1175/1520-0493(1983)111<1453:EMLITR>2.0.CO;2.
Coppola, E., and Coauthors, 2020: A first-of-its-kind multi-model convection permitting ensemble for investigating convective phenomena over Europe and the Mediterranean. Climate Dyn., 55, 3–34, https://doi.org/10.1007/s00382-018-4521-8.
Coppola, E., and Coauthors, 2021: Non-hydrostatic REGCM4 (REGCM4-NH): Model description and case studies over multiple domains. Geosci. Model Dev., 14, 7705–7723, https://doi.org/10.5194/gmd-14-7705-2021.
Curio, J., R. Schiemannm, K. I. Hodges, and A. G. Turner, 2019: Climatology of Tibetan Plateau vortices in reanalysis data and a high-resolution global climate model. J. Climate, 32, 1933–1950, https://doi.org/10.1175/JCLI-D-18-0021.1.
Dell’Osso, L., and S.-J. Chen, 1986: Numerical experiments on the genesis of vortices over the Qinghai-Tibet Plateau. Tellus, 38A, 236–250, https://doi.org/10.3402/tellusa.v38i3.11715.
Denis, B., J. Cote, and R. Laprise, 2002: Spectral decomposition of two-dimensional atmospheric fields on limited-area domains using the discrete cosine transform (DCT). Mon. Wea. Rev., 130, 1812–1829, https://doi.org/10.1175/1520-0493(2002)130<1812:SDOTDA>2.0.CO;2.
Donat, M. G., A. L. Lowry, L. V. Alexander, P. A. O’Gorman, and N. Maher, 2016: More extreme precipitation in the world’s dry and wet regions. Nat. Climate Change, 6, 508–513, https://doi.org/10.1038/nclimate2941.
Eaton, B., 2011: User’s guide to the Community Atmosphere Model CAM-5.1. NCAR, https://www2.cesm.ucar.edu/models/cesm1.0/cam/docs/ug5_1/ug.html.
Ek, M., K. Mitchell, Y. Lin, E. Rogers, P. Grunmann, V. Koren, G. Gayno, and J. Tarpley, 2003: Implementation of NOAH land surface model advances in the National Centers for Environmental Prediction operational mesoscale Eta model. J. Geophys. Res., 108, 8851, https://doi.org/10.1029/2002JD003296.
Feng, X., C. Liu, G. Fan, and J. Zhang, 2017: Analysis of the structure of different Tibetan Plateau vortex types. J. Meteor. Res., 31, 514–529, https://doi.org/10.1007/s13351-017-6123-5.
Feng, Z., F. Song, K. Sakaguchi, and L. R. Leung, 2021a: Evaluation of mesoscale convective systems in climate simulations: Methodological development and results from MPAS-CAM over the United States. J. Climate, 34, 2611–2633, https://doi.org/10.1175/JCLI-D-20-0136.1.
Feng, Z., and Coauthors, 2021b: A global high-resolution mesoscale convective system database using satellite-derived cloud tops, surface precipitation, and tracking. J. Geophys. Res. Atmos., 126, e2020JD034202, https://doi.org/10.1029/2020JD034202.
Funk, C., and Coauthors, 2015: The climate hazards infrared precipitation with stations—A new environmental record for monitoring extremes. Sci. Data, 2, 150066, https://doi.org/10.1038/sdata.2015.66.
Gao, Y., F. Chen, and Y. Jiang, 2020: Evaluation of a convection-permitting modeling of precipitation over the Tibetan Plateau and its influences on the simulation of snow-cover fraction. J. Hydrometeor., 21, 1531–1548, https://doi.org/10.1175/JHM-D-19-0277.1.
Gómez, B., and G. Miguez-Macho, 2017: The impact of wave number selection and spin-up time in spectral nudging. Quart. J. Roy. Meteor. Soc., 143, 1772–1786, https://doi.org/10.1002/qj.3032.
Grenier, H., and C. S. Bretherton, 2001: A moist PBL parameterization for large-scale models and its application to subtropical cloud-topped marine boundary layers. Mon. Wea. Rev., 129, 357–377, https://doi.org/10.1175/1520-0493(2001)129<0357:AMPPFL>2.0.CO;2.
Guilloteau, C., E. Foufoula-Georgiou, and C. D. Kummerow, 2017: Global multiscale evaluation of satellite passive microwave retrieval of precipitation during the TRMM and GPM eras: Effective resolution and regional diagnostics for future algorithm development. J. Hydrometeor., 18, 3051–3070, https://doi.org/10.1175/JHM-D-17-0087.1.
Gustafsson, N., 1990: Sensitivity of limited area model data assimilation to lateral boundary condition fields. Tellus, 42A, 109–115, https://doi.org/10.3402/tellusa.v42i1.11864.
Hashimoto, A., J. M. Done, L. D. Fowler, and C. L. Bruyère, 2016: Tropical cyclone activity in nested regional and global grid-refined simulations. Climate Dyn., 47, 497–508, https://doi.org/10.1007/s00382-015-2852-2.
Hersbach, H., and Coauthors, 2020: The ERA5 global reanalysis. Quart. J. Roy. Meteor. Soc., 146, 1999–2049, https://doi.org/10.1002/qj.3803.
Hock, R., and Coauthors, 2019: High mountain areas. IPCC Special Report on the Ocean and Cryosphere in a Changing Climate, H.-O. Pörtner et al., Eds., Cambridge University Press, 131–202.
Hong, S.-Y., 2010: A new stable boundary-layer mixing scheme and its impact on the simulated East Asian summer monsoon. Quart. J. Roy. Meteor. Soc., 136, 1481–1496, https://doi.org/10.1002/qj.665.
Hong, S.-Y., Y. Noh, and J. Dudhia, 2006: A new vertical diffusion package with an explicit treatment of entrainment processes. Mon. Wea. Rev., 134, 2318–2341, https://doi.org/10.1175/MWR3199.1.
Houze, R. A., 2004: Mesoscale convective systems. Rev. Geophys., 42, RG4003, https://doi.org/10.1029/2004RG000150.
Huffman, G. J., D. T. Bolvin, D. Braithwaite, K. Hsu, R. Joyce, P. Xie, and S.-H. Yoo, 2015: NASA Global Precipitation Measurement Integrated Multi-satellitE Retrievals for GPM (IMERG). NASA Algorithm Theoretical Basis Doc., version 4.5, 26 pp., https://gpm.nasa.gov/sites/default/files/document_files/IMERG_ATBD_V4.5.pdf.
Hunt, K., J. Curio, A. Turner, and R. Schiemann, 2018: Subtropical westerly jet influence on occurrence of western disturbances and Tibetan Plateau vortices. Geophys. Res. Lett., 45, 8629–8636, https://doi.org/10.1029/2018GL077734.
Ikeda, K., and Coauthors, 2021: Snowfall and snowpack in the western US as captured by convection permitting climate simulations: Current climate and pseudo global warming future climate. Climate Dyn., 57, 2191–2215, https://doi.org/10.1007/s00382-021-05805-w.
Imberger, M., X. Guo Larsen, N. Davis, and J. Du, 2020: Approaches toward improving the modelling of midlatitude cyclones entering at the lateral boundary corner in the limited area WRF model. Quart. J. Roy. Meteor. Soc., 146, 3225–3244, https://doi.org/10.1002/qj.3843.
Immerzeel, W. W., and Coauthors, 2020: Importance and vulnerability of the world’s water towers. Nature, 577, 364–369, https://doi.org/10.1038/s41586-019-1822-y.
Janowiak, J., B. Joyce, and P. Xie, 2017: Merged IR V1, edited by Andrey Savtchenko. Goddard Earth Sciences Data and Information Services Center (GES DISC), accessed 15 March 2022, https://disc.gsfc.nasa.gov/datasets/GPM_MERGIR_1/summary.
Johnson, R. H., 2011: Diurnal cycle of monsoon convection. The Global Monsoon System: Research and Forecast, C.-P. Chang et al., Eds., World Scientific, 257–276.
Jones, R., J. Murphy, and M. Noguer, 1995: Simulation of climate change over Europe using a nested regional-climate model. I: Assessment of control climate, including sensitivity to location of lateral boundaries. Quart. J. Roy. Meteor. Soc., 121, 1413–1449, https://doi.org/10.1002/qj.49712152610.
Joyce, R. J., J. E. Janowiak, P. A. Arkin, and P. Xie, 2004: CMORPH: A method that produces global precipitation estimates from passive microwave and infrared data at high spatial and temporal resolution. J. Hydrometeor., 5, 487–503, https://doi.org/10.1175/1525-7541(2004)005<0487:CAMTPG>2.0.CO;2.
Klemp, J., J. Dudhia, and A. Hassiotis, 2008: An upper gravity-wave absorbing layer for NWP applications. Mon. Wea. Rev., 136, 3987–4004, https://doi.org/10.1175/2008MWR2596.1.
Kukulies, J., D. Chen, and M. Wang, 2020: Temporal and spatial variations of convection, clouds and precipitation over the Tibetan Plateau from recent satellite observations. Part II: Precipitation climatology derived from Global Precipitation Measurement Mission. Int. J. Climatol., 40, 4858–4875, https://doi.org/10.1002/joc.6493.
Kukulies, J., D. Chen, and J. Curio, 2021: The role of mesoscale convective systems in precipitation in the Tibetan Plateau region. J. Geophys. Res. Atmos., 126, e2021JD035279, https://doi.org/10.1029/2021JD035279.
Li, P., C. Moseley, A. F. Prein, H. Chen, J. Li, K. Furtado, and T. Zhou, 2020: Mesoscale convective system precipitation characteristics over East Asia. Part I: Regional differences and seasonal variations. J. Climate, 33, 9271–9286, https://doi.org/10.1175/JCLI-D-20-0072.1.
Li, P., K. Furtado, T. Zhou, H. Chen, and J. Li, 2021: Convection-permitting modelling improves simulated precipitation over the central and eastern Tibetan Plateau. Quart. J. Roy. Meteor. Soc., 147, 341–362, https://doi.org/10.1002/qj.3921.
Lim, K.-S. S., and S.-Y. Hong, 2010: Development of an effective double-moment cloud microphysics scheme with prognostic cloud condensation nuclei (CCN) for weather and climate models. Mon. Wea. Rev., 138, 1587–1612, https://doi.org/10.1175/2009MWR2968.1.
Lind, P., and Coauthors, 2020: Benefits and added value of convection-permitting climate modeling over Fenno-Scandinavia. Climate Dyn., 55, 1893–1912, https://doi.org/10.1007/s00382-020-05359-3.
Liu, X., A. Bai, and C. Liu, 2009: Diurnal variations of summertime precipitation over the Tibetan Plateau in relation to orographically-induced regional circulations. Environ. Res. Lett., 4, 045203, https://doi.org/10.1088/1748-9326/4/4/045203.
Maussion, F., D. Scherer, T. Mölg, E. Collier, J. Curio, and R. Finkelnburg, 2014: Precipitation seasonality and variability over the Tibetan Plateau as resolved by the High Asia Reanalysis. J. Climate, 27, 1910–1927, https://doi.org/10.1175/JCLI-D-13-00282.1.
Mesinger, F., 1993: Forecasting upper tropospheric turbulence within the framework of the Mellor-Yamada 2.5 closure. Research activities in atmospheric and oceanic modeling, WMO/CAS/JSC WGNE Tech. Rep. 18, 2 pp.
Morrison, H., G. Thompson, and V. Tatarskii, 2009: Impact of cloud microphysics on the development of trailing stratiform precipitation in a simulated squall line: Comparison of one- and two-moment schemes. Mon. Wea. Rev., 137, 991–1007, https://doi.org/10.1175/2008MWR2556.1.
Nakanishi, M., and H. Niino, 2009: Development of an improved turbulence closure model for the atmospheric boundary layer. J. Meteor. Soc. Japan, 87, 895–912, https://doi.org/10.2151/jmsj.87.895.
Niu, G.-Y., and Coauthors, 2011: The community Noah land surface model with multiparameterization options (Noah-MP): 1. Model description and evaluation with local-scale measurements. J. Geophys. Res., 116, D12109, https://doi.org/10.1029/2010JD015139.
Ou, T., D. Chen, X. Chen, C. Lin, K. Yang, H.-W. Lai, and F. Zhang, 2020: Simulation of summer precipitation diurnal cycles over the Tibetan Plateau at the gray-zone grid spacing for cumulus parameterization. Climate Dyn., 54, 3525–3539, https://doi.org/10.1007/s00382-020-05181-x.
Poujol, B., A. F. Prein, and A. J. Newman, 2020: Kilometer-scale modeling projects a tripling of Alaskan convective storms in future climate. Climate Dyn., 55, 3543–3564, https://doi.org/10.1007/s00382-020-05466-1.
Prein, A. F., A. Gobiet, M. Suklitsch, H. Truhetz, N. Awan, K. Keuler, and G. Georgievski, 2013: Added value of convection permitting seasonal simulations. Climate Dyn., 41, 2655–2677, https://doi.org/10.1007/s00382-013-1744-6.
Prein, A. F., and Coauthors, 2015: A review on regional convection-permitting climate modeling: Demonstrations, prospects, and challenges. Rev. Geophys., 53, 323–361, https://doi.org/10.1002/2014RG000475.
Prein, A. F., R. M. Rasmussen, K. Ikeda, C. Liu, M. P. Clark, and G. J. Holland, 2017: The future intensification of hourly precipitation extremes. Nat. Climate Change, 7, 48–52, https://doi.org/10.1038/nclimate3168.
Prein, A. F., M. S. Bukovsky, L. O. Mearns, C. L. Bruyère, and J. M. Done, 2019: Simulating North American weather types with regional climate models. Front. Environ. Sci., 7, 36, https://doi.org/10.3389/fenvs.2019.00036.
Prein, A. F., C. Liu, K. Ikeda, R. Bullock, R. M. Rasmussen, G. J. Holland, and M. Clark, 2020: Simulating North American mesoscale convective systems with a convection-permitting climate model. Climate Dyn., 55, 95–110, https://doi.org/10.1007/s00382-017-3993-2.
Prein, A. F., R. Rasmussen, D. Wang, and S. Giangrande, 2021: Sensitivity of organized convective storms to model grid spacing in current and future climates. Philos. Trans. Roy. Soc., 379A, 20190546, https://doi.org/10.1098/rsta.2019.0546.
Prein, A. F., and Coauthors, 2023: Towards ensemble-based kilometer-scale climate simulations over the Third Pole region. Climate Dyn., 60, 4055–4081, https://doi.org/10.1007/s00382-022-06543-3.
Rasmussen, K., and R. Houze Jr., 2016: Convective initiation near the Andes in subtropical South America. Mon. Wea. Rev., 144, 2351–2374, https://doi.org/10.1175/MWR-D-15-0058.1.
Rasmussen, R., and Coauthors, 2011: High-resolution coupled climate runoff simulations of seasonal snowfall over Colorado: A process study of current and warmer climate. J. Climate, 24, 3015–3048, https://doi.org/10.1175/2010JCLI3985.1.
Roberts, N. M., and H. W. Lean, 2008: Scale-selective verification of rainfall accumulations from high-resolution forecasts of convective events. Mon. Wea. Rev., 136, 78–97, https://doi.org/10.1175/2007MWR2123.1.
Rockel, B., A. Will, and A. Hense, 2008: The regional climate model COSMO-CLM (CCLM). Meteor. Z., 17, 347–348, https://doi.org/10.1127/0941-2948/2008/0309.
Ruiz-Arias, J. A., J. Dudhia, F. J. Santos-Alamillos, and D. Pozo-Vázquez, 2013: Surface clear-sky shortwave radiative closure intercomparisons in the Weather Research and Forecasting model. J. Geophys. Res. Atmos., 118, 9901–9913, https://doi.org/10.1002/jgrd.50778.
Schultz, D. M., and C. A. Doswell, 2000: Analyzing and forecasting Rocky Mountain lee cyclogenesis often associated with strong winds. Wea. Forecasting, 15, 152–173, https://doi.org/10.1175/1520-0434(2000)015<0152:AAFRML>2.0.CO;2.
Shin, H. H., and S.-Y. Hong, 2015: Representation of the subgrid-scale turbulent transport in convective boundary layers at gray-zone resolutions. Mon. Wea. Rev., 143, 250–271, https://doi.org/10.1175/MWR-D-14-00116.1.
Skamarock, W. C., and J. B. Klemp, 2008: A time-split nonhydrostatic atmospheric model for weather research and forecasting applications. J. Comput. Phys., 227, 3465–3485, https://doi.org/10.1016/j.jcp.2007.01.037.
Skamarock, W. C., J. B. Klemp, M. G. Duda, L. D. Fowler, S.-H. Park, and T. D. Ringler, 2012: A multiscale nonhydrostatic atmospheric model using centroidal Voronoi tesselations and C-grid staggering. Mon. Wea. Rev., 140, 3090–3105, https://doi.org/10.1175/MWR-D-11-00215.1.
Sugimoto, S., and K. Ueno, 2010: Formation of mesoscale convective systems over the eastern Tibetan Plateau affected by plateau-scale heating contrasts. J. Geophys. Res., 115, D16105, https://doi.org/10.1029/2009JD013609.
Tastula, E.-M., B. Galperin, J. Dudhia, M. A. LeMone, S. Sukoriansky, and T. Vihma, 2015: Methodical assessment of the differences between the QNSE and MYJ PBL schemes for stable conditions. Quart. J. Roy. Meteor. Soc., 141, 2077–2089, https://doi.org/10.1002/qj.2503.
Thompson, G., P. R. Field, R. M. Rasmussen, and W. D. Hall, 2008: Explicit forecasts of winter precipitation using an improved bulk microphysics scheme. Part II: Implementation of a new snow parameterization. Mon. Wea. Rev., 136, 5095–5115, https://doi.org/10.1175/2008MWR2387.1.
Virtanen, P., and Coauthors, 2020: SciPy 1.0: Fundamental algorithms for scientific computing in Python. Nat. Methods, 17, 261–272, https://doi.org/10.1038/s41592-019-0686-2.
von Storch, H., H. Langenberg, and F. Feser, 2000: A spectral nudging technique for dynamical downscaling purposes. Mon. Wea. Rev., 128, 3664–3673, https://doi.org/10.1175/1520-0493(2000)128<3664:ASNTFD>2.0.CO;2.
Wu, G., T. Ma, Y. Liu, and Z. Jiang, 2020: PV-Q perspective of cyclogenesis and vertical velocity development downstream of the Tibetan Plateau. J. Geophys. Res. Atmos., 125, e2019JD030912, https://doi.org/10.1029/2019JD030912.
Wu, G., Y. Tang, B. He, Y. Liu, J. Mao, T. Ma, and T. Ma, 2022: Potential vorticity perspective of the genesis of a Tibetan Plateau vortex in June 2016. Climate Dyn., 58, 3351–3367, https://doi.org/10.1007/s00382-021-06102-2