Resolution Dependence of Initiation and Upscale Growth of Deep Convection in Convection-Allowing Forecasts of the 31 May–1 June 2013 Supercell and MCS

Russ S. Schumacher Department of Atmospheric Science, Colorado State University, Fort Collins, Colorado

Search for other papers by Russ S. Schumacher in
Current site
Google Scholar
PubMed
Close
Full access

Abstract

On 31 May 2013, a supercell thunderstorm initiated in west-central Oklahoma and produced a deadly tornado. This convection then grew upscale, with a nearly stationary line developing early on 1 June that produced very heavy rainfall and caused deadly flash flooding in the Oklahoma City area. Real-time convection-allowing (Δx = 4 km) model forecasts used during the Mesoscale Predictability Experiment (MPEX) provided accurate guidance regarding the timing, location, and evolution of convection in this case. However, attempts to simulate this event at higher resolution degraded the forecast, with the primary supercell failing to initiate and the evolution of the overnight MCS not resembling the observed system. Experiments to test the dependence of forecasts of this event on model resolution show that with grid spacing smaller than 4 km, mixing along the dryline in northwest Texas was more vigorous, causing low-level dry air to move more quickly eastward into Oklahoma. This drying prevented the supercell from initiating near the triple point in the higher-resolution simulations. Then, the lack of supercellular convection and its associated cold pool altered the evolution of subsequent convection. Whereas in observations and the 4-km forecast, a nearly stationary MCS developed parallel to, but displaced from, the supercell’s cold pool, the higher-resolution simulations instead had a faster-moving squall line that produced less rainfall. Although the degradation of convective forecasts at higher resolution is probably unusual and appears sensitive to the choice of boundary layer parameterization, these findings demonstrate that how numerical models treat boundary layer processes at different grid spacings can, in some cases, have profound influences on predictions of high-impact weather.

Corresponding author address: Dr. Russ Schumacher, Department of Atmospheric Science, Colorado State University, 1371 Campus Delivery, Fort Collins, CO 80523. E-mail: russ.schumacher@colostate.edu

Abstract

On 31 May 2013, a supercell thunderstorm initiated in west-central Oklahoma and produced a deadly tornado. This convection then grew upscale, with a nearly stationary line developing early on 1 June that produced very heavy rainfall and caused deadly flash flooding in the Oklahoma City area. Real-time convection-allowing (Δx = 4 km) model forecasts used during the Mesoscale Predictability Experiment (MPEX) provided accurate guidance regarding the timing, location, and evolution of convection in this case. However, attempts to simulate this event at higher resolution degraded the forecast, with the primary supercell failing to initiate and the evolution of the overnight MCS not resembling the observed system. Experiments to test the dependence of forecasts of this event on model resolution show that with grid spacing smaller than 4 km, mixing along the dryline in northwest Texas was more vigorous, causing low-level dry air to move more quickly eastward into Oklahoma. This drying prevented the supercell from initiating near the triple point in the higher-resolution simulations. Then, the lack of supercellular convection and its associated cold pool altered the evolution of subsequent convection. Whereas in observations and the 4-km forecast, a nearly stationary MCS developed parallel to, but displaced from, the supercell’s cold pool, the higher-resolution simulations instead had a faster-moving squall line that produced less rainfall. Although the degradation of convective forecasts at higher resolution is probably unusual and appears sensitive to the choice of boundary layer parameterization, these findings demonstrate that how numerical models treat boundary layer processes at different grid spacings can, in some cases, have profound influences on predictions of high-impact weather.

Corresponding author address: Dr. Russ Schumacher, Department of Atmospheric Science, Colorado State University, 1371 Campus Delivery, Fort Collins, CO 80523. E-mail: russ.schumacher@colostate.edu

1. Introduction

On 31 May 2013, a supercell thunderstorm initiated in west-central Oklahoma and produced a strong and deadly tornado near the town of El Reno, Oklahoma. This deep convection then grew upscale, with a nearly stationary line developing early on 1 June on the western flank of the previous convection. This slow-moving line produced heavy rainfall over much of central Oklahoma and caused flash flooding in the Oklahoma City area. The tornado caused 8 fatalities and the flash flood caused 13 (NWS 2014; Wurman et al. 2014). The combination of the tornado and flash-flood hazards, along with other unique societal circumstances leading up to this event, led to challenges in forecasting and warning communication (NWS 2014). This study will address a particular issue surrounding numerical weather prediction (NWP) model simulations and forecasts of the initiation, upscale growth, and maintenance of the deep convective storms on 31 May–1 June 2013.

Numerical models that explicitly predict convective motions are now a vital component in the suite of experimental and operational guidance to forecasters. With a horizontal grid spacing of 1–4 km, models configured in this way can faithfully represent supercell thunderstorms and organized convective systems (e.g., Weisman et al. 1997), but they do not adequately resolve the properties of individual updrafts and downdrafts (Bryan et al. 2003). Thus, they are typically referred to as “convection allowing” or “convection permitting,” as opposed to truly “convection resolving.” Extensive research has demonstrated that explicitly predicting deep convection provides more realistic representations of storm structure, convective mode, and, in most cases, precipitation (e.g., Done et al. 2004; Clark et al. 2007; Weisman et al. 2008; Kain et al. 2008; Schwartz et al. 2009, among many others) compared to using a parameterization to represent convective motions. Even convection initiation (CI), which remains poorly understood (e.g., Lock and Houston 2014), is generally well represented at 3–4-km grid spacing (Kain et al. 2013; Duda and Gallus 2013). Based on these encouraging results, convection-allowing model configurations have been implemented into operations in the United States and several other countries. Although Bryan et al. (2003) convincingly demonstrate that key processes are poorly resolved at these grid spacings, there has been less evidence to motivate the use of grid spacing smaller than 3–4 km for forecasting applications, in which computing limitations and time constraints are important considerations. For example, Schwartz et al. (2009) and Clark et al. (2012) show that convective structures are more detailed and realistic at 1–2 km compared to 4-km grid spacing, but there is little change to the skill of the precipitation forecasts. Clark et al. (2013) do show one tornado event in which 4-km grid spacing was insufficient, but 1-km grid spacing was sufficient, to predict rotating updrafts in a cluster of supercell storms, but the other cases they studied were very accurately predicted with 4-km grid spacing. Potvin and Flora (2015) show that supercell longevity and low-level rotation are more faithfully represented at smaller grid spacings in idealized simulations of supercells. Burghardt et al. (2014) find that the statistics of CI in their simulations at 429-m grid spacing were similar to those of other studies at 2–4-km grid spacing. Thus, the general conclusion of previous research is that, although resolution higher than that provided by 4-km grid spacing can sometimes be beneficial in the prediction of convective storms, the computational cost outweighs the benefits for real-time applications. However, this conclusion may be modified as computational capacity increases and very short-term predictions of the details of convective storms, such as those envisioned for the Warn-on-Forecast initiative (Stensrud et al. 2009) become commonplace.

In addition to the considerations of computing time and cost, there are also meteorological factors to consider at convection-allowing resolution. Wyngaard (2004) define the “terra incognita” for numerical models of the planetary boundary layer (PBL) as existing in between high-resolution large-eddy simulations (LES) in which turbulent motions are explicitly resolved, and mesoscale models for which parameterizations of the PBL were developed. At these intermediate resolutions, the assumptions used in the mesoscale PBL parameterizations begin to break down, and there is a mixture of explicitly resolved and parameterized turbulence. Several recent studies have highlighted limitations in the simulated properties of important PBL structures (e.g., Moeng et al. 2007; Hacker 2010; Lemone et al. 2010; Coffer et al. 2013; Coniglio et al. 2013; Ching et al. 2014; Nowotarski et al. 2014), and much research is ongoing to improve models of the PBL in the terra incognita.

This study will demonstrate that increased resolution strongly degraded the representation of CI and subsequent growth into a mesoscale convective system (MCS) in numerical forecasts of the 31 May–1 June 2013 event. Furthermore, it will show that this resolution dependence is related to the representation of PBL processes at different grid spacings. In section 2, an overview of the 31 May–1 June case is presented, and section 3 discusses the methods for the numerical model experiments. The results of the experiments are given in section 4; section 5 discusses the implications of this work, and section 6 concludes the manuscript.

2. The 31 May–1 June 2013 supercells and MCS: Overview and control forecast

The large-scale tropospheric pattern prior to the 31 May–1 June 2013 convection was characterized by a broad upper-level trough over the western and central United States and a 55 m s−1 jet streak near the base of the trough at 300 hPa (Fig. 1). A developing surface low pressure center was located over north Texas, and ahead of that low there were south-southwesterly winds and moist, very unstable air [surface-based convective available potential energy (SBCAPE) exceeded 4500 J kg−1 in central Oklahoma; Fig. 1]. Several surface boundaries separated distinct air masses in the Great Plains. A stationary front was oriented from northeast to southwest across southern Kansas, western Oklahoma, and the Texas Panhandle, and two drylines developed through the day over west Texas and moved into western Oklahoma (Figs. 2 and 3). (Here, the leading dryline is analyzed with a dashed curve, and the primary dryline is analyzed with the conventional symbols.) The warm sector was characterized by the high-CAPE air mentioned above, with drier air behind the primary dryline and behind the stationary front. Between the two drylines was a transition zone in which the winds were from the southwest, and the surface dewpoints were higher than behind the primary dryline, but lower than in the warm sector (Fig. 2). In summary, the ingredients for severe convection (e.g., Johns and Doswell 1992) were in place over Oklahoma, with moist, unstable air, several potential lifting mechanisms, and strong vertical wind shear.

Fig. 1.
Fig. 1.

The Rapid Refresh (RAP) model analysis at 1800 UTC 31 May 2013, showing wind speed at 300 hPa (color shading in m s−1), wind barbs at 850 hPa (half barb = 5, full barb = 10, pennant = 50 kt; 1 kt = 0.5144 m s−1), pressure adjusted to sea level (black contours every 4 hPa), and SBCAPE (gray shading in J kg−1).

Citation: Monthly Weather Review 143, 11; 10.1175/MWR-D-15-0179.1

Fig. 2.
Fig. 2.

Surface analysis at 2100 UTC 31 May 2013 over central Oklahoma, including standard and Oklahoma Mesonet surface observations (surface station model is conventional with temperature and dewpoint in °C) and manually analyzed surface boundaries. (Figure courtesy of Mike Coniglio of NSSL.)

Citation: Monthly Weather Review 143, 11; 10.1175/MWR-D-15-0179.1

Fig. 3.
Fig. 3.

GOES-14 visible satellite image at 2145 UTC 31 May 2013 with manually analyzed surface boundaries. (Satellite image provided by Dan Lindsey of NOAA.)

Citation: Monthly Weather Review 143, 11; 10.1175/MWR-D-15-0179.1

Convection initiated between 2100 and 2200 UTC in two locations: along the stationary front in south-central Kansas and north-central Oklahoma, and near the intersection between the stationary front and the leading dryline in west-central Oklahoma (Figs. 3 and 4a,b). Such “triple points,” where three different air masses intersect, are common locations for CI (e.g., Weiss and Bluestein 2002; Wakimoto et al. 2006; Ziegler et al. 2007). The convection near this triple point rapidly intensified into a large supercell that ultimately produced the EF-3 El Reno tornado [see Wurman et al. (2014) for detailed analysis of the supercell and tornado]. This supercell moved toward the southeast over time, and produced a cold pool that became positioned from west to east across central Oklahoma between 0200 and 0400 UTC 1 June; the cold pool is apparent from the approximately 10°C decrease in temperature at stations in central Oklahoma between 0000 and 0200 UTC (Figs. 4b,c) and the southward movement of the 20°C dewpoint contour shown in Fig. 4.1 After the southeastward movement and eventual dissipation of the initial supercell, a new line of convective cells developed to the northwest of the existing cells and displaced from the outflow boundary (Fig. 4d), akin to the “rearward off-boundary development” analyzed by Peters and Schumacher (2015a,b). This convective line became organized into the training line/adjoining stratiform (TL/AS) structure described by Schumacher and Johnson (2005) (Fig. 4e) and remained nearly stationary over central Oklahoma through approximately 0700 UTC before eventually moving southward in the morning hours. Both the initial supercell and the subsequent MCS contributed to large rainfall accumulations over central Oklahoma (Figs. 5a,b), and the evolution of the heavy-rain-producing convection is summarized in the Hovmöller diagram in Fig. 5c. Numerous gauge reports of over 150 mm, with a maximum of 208 mm, were collected by the Community Collaborative Rain, Hail and Snow (CoCoRaHS; Cifelli et al. 2005) network, and the heavy precipitation over the Oklahoma City area led to flash flooding that caused 13 fatalities (NWS 2014).

Fig. 4.
Fig. 4.

Composite radar reflectivity (dBZ) from the Twin Lakes, OK (KTLX), WSR-88D, surface observations (surface station model is conventional with temperature and dewpoint in °C), and the 20°C surface dewpoint contour from the Rapid Refresh (RAP) model analysis at (a) 2200 UTC 31 May, (b) 0000 UTC 1 Jun, (c) 0200 UTC 1 Jun, (d) 0400 UTC 1 Jun, and (e) 0600 UTC 1 Jun 2013. The dewpoint contour is shown as a simple visualization of the location of the dryline and the convectively generated cold pool. The red asterisk in (a) shows the location of the sounding in Fig. 6.

Citation: Monthly Weather Review 143, 11; 10.1175/MWR-D-15-0179.1

Fig. 5.
Fig. 5.

The NCEP stage-IV (Lin and Mitchell 2005) precipitation analysis (mm) for (a) the 15-h period ending at 0300 UTC 1 Jun (when the precipitation was primarily produced by the supercell) and (b) the 9-h period ending at 1200 UTC 1 Jun 2013 (when the precipitation was produced by the MCS). (c) Time–latitude diagram of hourly stage-IV precipitation, averaged over 94.5°–99°W longitude; this region is outlined in red in (b).

Citation: Monthly Weather Review 143, 11; 10.1175/MWR-D-15-0179.1

The 31 May–1 June 2013 supercell and MCS took place during the Mesoscale Predictability Experiment (MPEX; Weisman et al. 2015), which included the collection of mobile radiosonde observations of the preconvective and convectively disturbed environment (Trapp et al. 2015). Although safety concerns associated with collecting data near the tornadic supercell near an urban area prevented the collection of a large number of soundings on this day, one sounding representative of the warm-sector environment prior to CI was obtained at 1920 UTC (Fig. 6; see Fig. 4a for the sounding location). This sounding showed a substantial capping inversion at approximately 770 hPa, but very large values of CAPE (4000 J kg−1, which is an underestimate because the sounding was terminated prior to reaching its level of neutral buoyancy) for parcels that could penetrate this inversion. This observed sounding also revealed low-level and deep-layer shear consistent with other tornadic supercell environments (e.g., Thompson et al. 2012).

Fig. 6.
Fig. 6.

Skew T–logp diagram showing a mobile radiosonde observation collected by Colorado State University (CSU) at 1920 UTC 31 May at the location shown in Fig. 4a (solid red and green lines; hodograph in red) and a profile from the same location in the RT model forecast at 1930 UTC (blue lines and hodograph in blue). The parcel curve (dashed black) is for a parcel with the mean properties of the lowest 50 hPa in the observed sounding. Thermodynamic quantities for both the observed and model soundings are shown at right.

Citation: Monthly Weather Review 143, 11; 10.1175/MWR-D-15-0179.1

In addition to the field data collection, a key component of MPEX was the evaluation of a variety of convection-allowing numerical model forecasts, including both deterministic runs and ensemble forecasts (Weisman et al. 2015). The author’s research group at Colorado State University (CSU) contributed a real-time 4-km ARW (Skamarock et al. 2008) forecast that was used by the MPEX forecasters and mission scientists. (The configuration of the model and other related details will be given in the next section.) The initiation and evolution of deep convection in Oklahoma in the CSU WRF forecast initialized at 1200 UTC 31 May had many close similarities to what was observed (Fig. 7). The initiation of convection was delayed by approximately 30 min in the forecast compared to the observations, but occurred in a very similar location (cf. Figs. 7a,b and 4a,b). This convection developed into a supercell thunderstorm that moved southeastward across central Oklahoma, although the observed supercell intensified more quickly such that it was very intense by 0000 UTC, whereas the modeled supercell intensified more in the 0000–0200 UTC period (cf. Figs. 7b,c and 4b,c). A convective line subsequently developed to the supercell’s northwest and evolved into a TL/AS MCS (Figs. 7c–e). The initiation of this MCS at around 0400 UTC 1 June showed particularly close agreement with the observed convection; the decaying supercell and developing convective line are in almost the exact location in the model as in observations (cf. Figs. 7d and 4d). Although the convective evolution was very similar between the real-time forecast and the observations, the model underpredicted the precipitation in Oklahoma by a considerable amount, especially the precipitation produced by the supercells in the evening (Fig. 8a). This underprediction appears related to the slower intensification and lower intensity (at least in terms of reflectivity) of the supercells in the model. The precipitation in the modeled MCS was also spread out over a somewhat larger spatial area, in contrast to the concentrated band of very heavy precipitation in observations (cf. Figs. 8b,c and 5b,c). Recalling that the rainfall accumulation at a given location is a function of the rain rate and the duration (Doswell et al. 1996), we see that the real-time model forecast was very accurate in its prediction of the timing, location, and duration of heavy rainfall, but underestimated the rainfall rate, and thus the total accumulation was underpredicted. Nonetheless, considering the uncertainty in where heavy rain would occur during this time period as discussed by NWS (2014) (in particular, forecasters highlighted regions to the northeast of where the heaviest rain actually fell), this forecast could have provided beneficial information about the possible evolution of heavy-rain-producing convection on 31 May–1 June 2013.

Fig. 7.
Fig. 7.

Simulated composite radar reflectivity (dBZ), 10-m wind barbs, and the 20°C surface dewpoint contour from the RT forecast at (a) 2200 UTC 31 May, (b) 0000 UTC 1 Jun, (c) 0200 UTC 1 Jun, (d) 0400 UTC 1 Jun, and (e) 0600 UTC 1 Jun 2013. The dewpoint contour is shown as a simple visualization of the location of the dryline and the convectively generated cold pool.

Citation: Monthly Weather Review 143, 11; 10.1175/MWR-D-15-0179.1

Fig. 8.
Fig. 8.

As in Fig. 5, but for the RT forecast. The red rectangle in (a) shows the area-averaging region for the time series shown in Fig. 12. In addition to hourly precipitation, the hourly maximum updraft helicity in the 2–5-km layer (averaged over the same region as precipitation) is contoured at 10 and 30 m2 s−2.

Citation: Monthly Weather Review 143, 11; 10.1175/MWR-D-15-0179.1

Considering the accuracy of the model forecast in capturing the initiation, evolution, and maintenance of convection during this event, and the fact that 4-km horizontal grid spacing only marginally resolves the properties of deep convection (e.g., Bryan et al. 2003), the author sought to conduct a simulation of this event with the same initial and lateral boundary conditions, but with higher spatial resolution. Surprisingly, the output of this simulation showed a very different pattern of convection and precipitation: the primary supercell thunderstorm in central Oklahoma that was well simulated in the real-time forecast was absent from this higher-resolution simulation. This unusual discrepancy prompted additional investigation into its causes, and the purpose of this manuscript is to report on the possible reasons for degraded forecasts at higher spatial resolution in WRF Model simulations of this event.

3. Design and configuration of numerical model experiments

Configuration of real-time forecast

During the MPEX field experiment in 2013, CSU ran real-time ARW forecasts initialized twice per day (0000 and 1200 UTC). These forecasts used version 3.4.1 of ARW, and used forecasts from the NCEP Global Forecast System (GFS) as initial and lateral boundary conditions, with the lateral boundary conditions updated every 3 h. The horizontal grid spacing was 4 km with a single domain that covered much of the western and central United States (Fig. 9), with 51 vertical levels on a stretched grid. The time step was 25 s. The physical parameterizations included the Mellor–Yamada–Janjić (MYJ; Mellor and Yamada 1982; Janjić 2002) PBL parameterization, the Morrison et al. (2009) two-moment cloud microphysics parameterization, the Noah (Chen and Dudhia 2001) land surface model, and the Rapid Radiative Transfer Model for GCMs (Iacono et al. 2008). Convective motions were treated explicitly rather than parameterized. Additional graphics from these runs as they were performed in real time are available in the MPEX field catalog (http://catalog.eol.ucar.edu/mpex).

Fig. 9.
Fig. 9.

Location of the domains used for the numerical simulations. The dashed line shows the domain that was used for the real-time forecast during MPEX and the thick solid line shows the domain used for the simulations presented in this article, which is also commonly used for real-time forecasts at CSU. The thick gray lines show the 1.33-km and 444-m domains for the nested runs. See the text and Table 1 for further details.

Citation: Monthly Weather Review 143, 11; 10.1175/MWR-D-15-0179.1

For the simulations reported in this paper, a few modifications were made to the original real-time configuration to make it easier to conduct meaningful resolution-sensitivity experiments. First, the model version was upgraded to version 3.6.1, the most recent version available at the time of this writing. Second, the model domain was changed to the one typically used at CSU for 4-km runs, which is shifted to the east from the domain used during MPEX (Fig. 9). This domain eliminates problems associated with numerical noise that arose over the complex terrain of the Rocky Mountains far from the region of interest (e.g., in Montana) when running at very high resolutions (sub-2-km grid spacing). The grid size and spacing was unchanged between these, and the differences in the model forecasts over the region of interest were negligible as a result of shifting the domain eastward; the domain maximum precipitation changed by less than 5 mm between the true real-time run and the rerun, with the spatial distribution of precipitation nearly identical (not shown). For simplicity, hereafter this simulation will be referred to as the real-time (RT) simulation, because it reflects the configuration of the actual real-time run.

Then, a series of experiments was conducted in which the horizontal grid spacing is varied, and the number of grid points adjusted so that the model domain is identical to that in the RT simulation. Grid spacings of 5, 3, 2, and 1.33 km were used. Additionally, two simulations using the Yonsei University (YSU; Noh et al. 2003) PBL parameterization (a “nonlocal” scheme, in contrast to the “local” MYJ scheme), and two nested-grid simulations were also conducted. For the nested-grid simulations, the first included a nest at 1.33-km grid spacing inside the RT run with two-way nesting; one-way nesting was also tested and produced essentially the same results for the fields of interest. The second included a nest at 444-m grid spacing inside the 1.33-km run with one-way nesting. Because the 444-m nested run closely resembled its parent simulation through 15 h of integration (0300 UTC 1 June), it was not integrated further owing to computational expense. All simulations used the same 51 vertical levels as RT, and the model time step was adjusted proportionally. All of these experiments, with abbreviations that will be used in the remainder of the manuscript, are summarized in Table 1.

Table 1.

Summary of the ARW experiments, all of which were initialized at 1200 UTC 31 May 2013 with GFS initial and lateral boundary conditions. All simulations were integrated for 24 h, except for 444M-nest, which was only integrated for 15 h owing to its computational expense and its similarity with its parent simulation through that time.

Table 1.

In addition to the experiments reported here, numerous other model configurations were tested at 4-km grid spacing, including initial and lateral boundary conditions from other model analyses and different PBL and microphysics parameterizations. For convective evolution and rainfall over Oklahoma, none of these configurations matched the observations as closely as the RT run, so the details of these simulations (and any possible resolution sensitivities) were not explored further. In general, the heavy precipitation in these simulations was predicted to be farther north and east than it was in RT or in observations. Additionally, a further test of the specific domain location was also conducted by shifting the RT domain to the south by 2.5° latitude; the results were essentially the same as the RT run for the fields of interest. This indicates that the results to follow are not sensitive to the specific choice of the domain location, though of course it was not possible to test all possible domain configurations.

4. Results

a. Preconvective environment

More details regarding the near-storm environment in the model simulations will be presented in the subsections to follow, but a brief comparison between the model output and available observations is given here. In the sounding comparison shown in Fig. 6, the modeled environment near where CI would occur was broadly similar to the observed environment, although the model output showed more moisture in the PBL and its inversion was located slightly lower than that observed. A PBL that is too shallow and moist is a common finding in convection-allowing forecasts using the MYJ PBL parameterization (e.g., Coniglio et al. 2013; Trier et al. 2015). A similar story emerges when comparing the PBL-averaged dewpoint in RT to the RAP analysis, which assimilates available surface observations including those from the Oklahoma Mesonet. There are large errors in low-level dewpoint behind the cold front and dryline in the Texas Panhandle (Fig. 10) with the model-predicted dewpoint much too high. The dewpoint in the transition zone is also predicted to be too high, but the magnitude of the errors is smaller. The orientation of the boundaries in western Oklahoma at 2200 UTC (as roughly estimated by the 20°C dewpoint contour) is similar between RT and the RAP, but the RAP-analyzed dryline and triple point are approximately 50 km farther northeast than those in RT. This is consistent with the initiation of the primary supercell occurring approximately 50 km northeast in observations compared with RT. (In other words, the modeled and observed storms initiated in similar locations relative to their respective triple points.) The PBL-average dewpoint in the warm sector was generally similar between RT and the RAP analysis, with errors generally less than 1 K in magnitude over central and eastern Oklahoma. This brief comparison shows that, as in the case studied by Trier et al. (2015), although there were errors in the PBL conditions between the RT run and observations, the evolution of the relevant boundaries and their relationship to the location of convection initiation were similar.

Fig. 10.
Fig. 10.

Errors in the PBL-average dewpoint (K), calculated as the difference between the RT forecast and the RAP analysis, at 2200 UTC 31 May 2013 (forecast hour 10). The 16° and 20°C contours from each run are also shown, with the RT run in blue and the RAP analysis in green. The blue asterisk indicates the location of CI in radar observations and the black cross shows the location of CI in the model output. The Unified Postprocessor (http://www.dtcenter.org/wrf-nmm/users/overview/upp_overview.php) and wgrib2 software packages were used to regrid all model output to the RT grid, and to calculate the PBL-average dewpoint.

Citation: Monthly Weather Review 143, 11; 10.1175/MWR-D-15-0179.1

b. Convective initiation, evolution, and precipitation

The key difference between the RT simulation and those at higher resolution is that the rainfall associated with the supercell over central Oklahoma is absent in the higher-resolution runs (cf. Figs. 8a and 11b–f). The 5KM simulation, on the other hand, produced a precipitation distribution very similar to that in RT (Fig. 11a). Figure 12 illustrates that, as discussed above, all of the simulations underestimate precipitation in central Oklahoma where the supercell occurred, but the 4KM and 5KM runs correctly begin producing precipitation before 0000 UTC, whereas the higher-resolution forecasts do not predict measurable precipitation over central Oklahoma until after 0300 UTC. This strong decrease in forecast accuracy with higher resolution stands in contrast to most previous findings (discussed in section 1), which show either similar skill or small improvements with higher resolution. The large resolution dependence is limited only to the convection in central Oklahoma, however, as the predicted rainfall locations and amounts are similar along the stationary front in northern Oklahoma in all experiments (Fig. 11).

Fig. 11.
Fig. 11.

As in Fig. 8a, but for (a) 5KM, (b) 3KM, (c) 2KM, (d) 1.33KM, (e) 1.33KM-nest, and (f) 444M-nest.

Citation: Monthly Weather Review 143, 11; 10.1175/MWR-D-15-0179.1

Fig. 12.
Fig. 12.

Time series of accumulated precipitation (mm) averaged over the region shown by the red rectangle in Fig. 11.

Citation: Monthly Weather Review 143, 11; 10.1175/MWR-D-15-0179.1

The time–latitude diagrams in Fig. 13 illustrate the important differences in convective evolution between the coarser- (5KM; Fig. 13a) and finer- (Figs. 13b–f) resolution experiments. In 5KM, as in RT (Fig. 8c) and observations (Fig. 5c), a large supercell initiates in central Oklahoma just before 0000 UTC 1 June, and then moves eastward. In all of the higher-resolution simulations, on the other hand, deep convection only initiates in northern Oklahoma, and then steadily moves southward through the evening. Furthermore, whereas the observed system and those predicted by RT and 5KM remained at approximately the latitude of central Oklahoma for several hours before moving toward the south, the convective line in the higher-resolution experiments moved steadily southward between 0300 and 1200 UTC (Fig. 13). These differences result in the total precipitation in central Oklahoma in all runs with finer resolution than RT being approximately half that of RT and 5KM.

Fig. 13.
Fig. 13.

As in Fig. 8c, but for (a) 5KM, (b) 3KM, (c) 2KM, (d) 1.33KM, and (e) 1.33KM-nest. Updraft helicity is shown in (a)–(c). The magnitude of updraft helicity is scaled so that it matches the RT run (e.g., the magnitude is multiplied by 5/4 for 5KM and multiplied by 3/4 for 3KM). The 444M-nest experiment is omitted because the domain does not cover the entire area used for the calculations for the other experiments.

Citation: Monthly Weather Review 143, 11; 10.1175/MWR-D-15-0179.1

In experiment 3KM, deep convection does attempt to initiate near the triple point in west-central Oklahoma, but it fails to intensify and take on supercell characteristics (Figs. 14a,b and Fig. 15b) and produces minimal rain (Fig. 11b). A line of convection, including embedded supercells, also initiates farther north along the stationary front, consistent with observations and RT. However, because the supercell in central Oklahoma is absent, the convectively generated cold pool is also absent in the 0000–0200 UTC time frame (cf. Figs. 7c and 14c). As a result, instead of the new development of a convective line to the north and west of the outflow boundary that was observed, the high-resolution experiments simply show a squall line moving south through Oklahoma, with convection near the gust front rather than separated from it (Fig. 14). The different convective evolution between RT and 3KM is also illustrated by the presence of a long, coherent swath of updraft helicity in the 2–5-km layer [indicative of an intense rotating updraft (e.g., Kain et al. 2008)] in central Oklahoma in RT (Fig. 15a), but only short and disorganized updraft-helicity paths associated with the embedded supercells ahead of the front in 3KM (Fig. 15b). The evolution in 3KM and shown in Figs. 14 and 15b is very similar to that in the other higher-resolution experiments (not shown), with the primary difference being that the convective cell resulting from the failed supercell initiation was slightly stronger in 1.33KM and 444M-nest. Those cells still produced minimal precipitation, however (Figs. 11d,f). Now, having established that there was indeed a strong resolution dependence of the precipitation forecasts during this event, the next step is to diagnose the underlying reasons for that dependence.

Fig. 14.
Fig. 14.

As in Fig. 7, but for the 3KM simulation.

Citation: Monthly Weather Review 143, 11; 10.1175/MWR-D-15-0179.1

Fig. 15.
Fig. 15.

Maximum 2–5-km updraft helicity (m2 s−2) over the period 2100 UTC 31 May–0400 UTC 1 Jun for (a) RT and (b) 3KM. The magnitude of updraft helicity in 3KM is scaled so that it matches the RT run (the magnitude is multiplied by 3/4 for 3KM). The location of El Reno, OK, is shown for reference.

Citation: Monthly Weather Review 143, 11; 10.1175/MWR-D-15-0179.1

c. Possible causes of resolution dependence

Examination of the environment within the transition zone near the CI location reveals key PBL moisture differences between the coarser and finer-resolution experiments, which in turn are related to differences in the position and motion of the surface boundaries. At 2200 UTC, using the 20°C PBL-averaged dewpoint contour as a simple proxy for the separation between the moister air of the warm sector and the drier air of the transition zone, it is apparent that low-level dry air has advanced much farther to the northeast in the higher-resolution experiments than in RT and 5KM (Fig. 16). In fact, in all but RT and 5KM, the leading dryline had already passed through the location where CI would occur shortly after this time in RT. The position of the leading dryline (Fig. 16a) and the average dewpoint in the vicinity of CI (Fig. 16b) are almost identical in RT and 5KM, with the boundary moving through more quickly (and accordingly reducing the dewpoint) in 3KM, and even more quickly in 2KM. The difference is reduced in 1.33KM, however, with the dryline position and PBL dewpoint falling between RT and 2KM (Fig. 16). These results indicate that the northeastward progression of dry air in simulations of this event was sensitive to the horizontal grid spacing of the model. There were also some differences between the nested-grid runs configurations and their parent runs. The leading dryline in 1.33KM-nest moved the fastest to the northeast of all simulations, and the dewpoint in the CI region accordingly quickly decreased (Fig. 16). The dryline in 444M-nest also progressed farther than that in its parent run (1.33KM), but this difference was much less pronounced than that between RT and 1.33KM, or between 1.33KM and 1.33KM-nest. The possible effects of nesting will be addressed later, as we will first focus on the effects of increased grid resolution.

Fig. 16.
Fig. 16.

(a) The 20°C contour of the PBL-average dewpoint at 2200 UTC 31 May (10-h forecast), with colors as indicated in the legend. The cross shows the location of the supercell initiation in the RT run. (b) Time series of PBL-average dewpoint, averaged over the box shown in (a). The Unified Postprocessor (http://www.dtcenter.org/wrf-nmm/users/overview/upp_overview.php) and wgrib2 software packages were used to regrid all model output to the RT grid, and to calculate the PBL-average dewpoint.

Citation: Monthly Weather Review 143, 11; 10.1175/MWR-D-15-0179.1

The changes owing to increased horizontal resolution will be illustrated with a comparison between RT and 2KM, because they showed the largest differences in terms of low-level moisture. Differences between RT and the other higher-resolution runs showed similar spatial structure and evolution but smaller magnitudes (not shown). Differences in the PBL-average dewpoint exceeding 2 K originate at approximately 1800 UTC over west Texas, with the dewpoint higher in RT than 2KM (Fig. 17a). These differences then expand, increase in magnitude to over 4 K in spots, and move toward the northeast over the next 4 hours (Figs. 17b,c). Differences exceeding 2 K reach southwest Oklahoma by 2200 UTC; both drylines in 2KM have correspondingly moved farther northeastward than those in RT. In total, this comparison shows that dry air advanced northeastward much more quickly in 2KM than it did in RT, leading to large differences in low-level moisture in western Oklahoma by the time that CI would occur. In contrast, there are no systemic differences in PBL-averaged dewpoint elsewhere in the domain.

Fig. 17.
Fig. 17.

Difference (RT minus 2KM) in the PBL-average dewpoint (°C) at (a) 1800, (b) 2000, and (c) 2200 UTC 31 May. The 16° and 20°C contours from each run are also shown, with the RT run in blue and the 2KM run in green. The Unified Postprocessor (http://www.dtcenter.org/wrf-nmm/users/over view/upp_overview.php) and wgrib2 software packages were used to regrid all model output to the RT grid, and to calculate the PBL-average dewpoint.

Citation: Monthly Weather Review 143, 11; 10.1175/MWR-D-15-0179.1

West–east vertical sections through this region show that although the depth of the PBL is similar in both RT and 2KM, the mixing ratio in the PBL is reduced by about 1–2 g kg−1 in all three regions: ahead of the primary dryline, in the transition zone, and farther east in the warm sector (cf. Figs. 18a,b). Viewing these differences on a skew T–logp diagram (Fig. 18c) reveals that they are subtle but important when considering the possibility for CI. Although both thermodynamic profiles have ample CAPE and little CIN, the moister PBL in RT means that the lifted condensation level and level of free convection are lower than those in 2KM. The environment is also much nearer to saturation at the top of the PBL in RT. In total, this indicates that it would be much more difficult for CI to occur in 2KM than in RT, and this is corroborated by the developing convective clouds at this time in RT (Fig. 18a) but the near-absence of clouds in 2KM (Fig. 18b). As in the observations (Fig. 3), CI occurs in RT near the triple point at the intersection of the leading dryline and the stationary front.

Fig. 18.
Fig. 18.

(a),(b) West–east vertical section of water vapor mixing ratio (g kg−1, color shading), virtual potential temperature (K, thin contours), and cloud water (thick contour at 0.05 g kg−1) for (a) RT and (b) 2KM at 2200 UTC 31 May. The values are averaged over the north–south band A–B in Figs. 19c and 19d. (c) Skew T–logp diagrams (RT in red/green, 2KM in blue) at the point shown by the cross in Fig. 17c at 2200 UTC, where CI would occur approximately 30 min after this time in the RT run. The thermodynamic calculations shown are for the RT run.

Citation: Monthly Weather Review 143, 11; 10.1175/MWR-D-15-0179.1

The previous analysis identified that the low-level moisture differences originated in west Texas, but the reasons for those differences have not yet been explored. Torn and Romine (2015) illustrated that low-level moisture, and the evolution of convection and precipitation in an ensemble prediction system, were sensitive to the position and strength of upstream potential vorticity anomalies over Colorado and Kansas in this case. However, little difference was apparent in the representation of these PV anomalies between the simulations at differing resolutions (not shown), which is unsurprising given that identical initial conditions were used across these runs. Thus, the differences do not appear to result from large-scale uncertainties such as these. Ching et al. (2014) showed that convectively induced secondary circulations (CISCs) that develop in the PBL (typically in the form of rolls) can be strongly dependent on the horizontal grid spacing in the model. In particular, PBL growth and magnitude of vertical velocity were found to be much stronger at 1-km horizontal grid spacing than at 3 km in local schemes such as MYJ, which is used in the present study. This, in turn, was attributed to the explicit representation, but poor resolution, of these CISCs in the terra incognita. Ching et al. (2014) found that in schemes with only local mixing (i.e., the turbulent transport depends only on the stability at immediately neighboring vertical levels), the critical Rayleigh number is exceeded at high resolutions where the horizontal wavelength of disturbances can be similar to the depth of the unstable layer, causing the rapid growth of circulations that represent the same physical processes that occur in nature, but without the interactions with (still unresolved) smaller-scale turbulence that are important in producing observed PBL structures. In parameterizations that allow for nonlocal mixing, however, these problems are mitigated (this subject will be explored more fully in section 4d). Similar behavior appears to be at work in the simulations of the 31 May–1 June 2013 event, although the processes in question are not rolls but gravity waves. Figure 19 shows that grid-scale upward/downward velocity pairs develop in the transition zone between the drylines and ahead of the leading dryline in both runs, but they develop much earlier in 2KM than they do in RT.

Fig. 19.
Fig. 19.

Vertical velocity at 1-km MSL (cm s−1, color shading), and the 16° and 20°C dewpoint contours at 0.5-km MSL. (a) RT at 2000 UTC, (b) RT at 2200 UTC, (c) 2KM at 2000 UTC, and (d) 2KM at 2200 UTC. The black line C–D in (a) indicates the location of the vertical sections in Fig. 20, and the black horizontal lines A–B indicate the region over which the vertical sections in Figs. 18a and 18b were calculated. The red rectangle shows the region over which the average sounding was calculated for Fig. 21 and maximum vertical velocity was calculated for Fig. 22. For the 2KM runs, the dewpoint contours have been smoothed with a three-point filter for clarity. The region of enhanced ascent mentioned in the text is shown with the gray lines in (b).

Citation: Monthly Weather Review 143, 11; 10.1175/MWR-D-15-0179.1

The structure and time evolution of these velocity pairs are suggestive of waves, as they propagate coherently toward the northeast with time (not shown) and they generally exhibit an out-of-phase relationship between the vertical velocity and potential temperature (Fig. 20b). Examination of the vertical stability profile in this region reveals a neutral to slightly unstable layer below about 1.5 km MSL, topped by a stable inversion, with an elevated mixed layer above that (Fig. 20a; see also Fig. 6 for an example sounding in the warm sector). The waves develop and are trapped in the stable layer between the near-neutral layer above and the unstable layer below. To assess the possibility for trapped waves in this region, the square of the Scorer parameter {, where N is the Brunt–Väisälä frequency, u is the wind, and c is the phase speed of the waves} was calculated for an average vertical profile in the transition zone and for waves propagating in the direction of line C–D in Fig. 19b. The first term, which relates to the stability, is large within the inversion layer, negative below, and near zero above (Fig. 21). The second term relates to the curvature of the wind profile and opposes the first term through much of the inversion layer. But the total Scorer parameter does show large changes around 2 km MSL, with a layer of negative around 1.5 km MSL, then a sharp increase within the stable layer, and a sharp decrease above that. Regions where changes sharply from positive to negative with height have the potential for gravity wave trapping, as waves can be maintained within the stable layer but are not supported in layers where is negative (e.g., Durran 1990).

Fig. 20.
Fig. 20.

As in Figs. 18a and 18b, but vertical velocity (red contours every 30 cm s−1, with zero contour omitted and negative contours dashed) is shown instead of cloud water for (a) RT and (b) 2KM at 2000 UTC along the vertical section shown by line C–D in Figs. 19a and 19b.

Citation: Monthly Weather Review 143, 11; 10.1175/MWR-D-15-0179.1

Fig. 21.
Fig. 21.

Vertical profile of the Scorer parameter squared (m−2) at 2000 UTC, averaged over the area shown by the red rectangle in Fig. 19. The blue line shows the first term, which relates to stability; the red line shows the second term, which relates to the curvature of the wind profile; and the total is shown by the thick black line. The phase speed of the waves was estimated from animations of the model output to be 13 m s−1, with motion from 205° (i.e., from the south-southwest).

Citation: Monthly Weather Review 143, 11; 10.1175/MWR-D-15-0179.1

Although it is expected that the spatial scale on which the waves occur would change as the grid spacing changes, there are much more substantive differences between the RT and 2KM than just the wavelength. As mentioned above and illustrated in Figs. 19 and 20, the waves develop much earlier in 2KM than in RT, and the vertical velocities in the waves are much stronger as well. Figure 22 demonstrates that the vertical velocities in the transition zone generally become much stronger as horizontal grid spacing decreases. With both the earlier onset and the greater intensity of these waves in the higher-resolution simulations, superimposed on the typical daytime growth of the PBL, mixing of dry air (in both the horizontal and vertical) is more vigorous, and the dryline advances more quickly. It appears that, because the waves occur near the interface between the stable inversion and the neutral-to-unstable layer below, they are responsible for vertical mixing on their own as well as for exciting other motions within the PBL in the higher-resolution simulations but not those at coarser resolution. Because the amplitude of these waves increases at higher resolution, it appears that even though these waves are a different process than PBL rolls that result from shear and/or dynamic instability, the findings by Ching et al. (2014) that short-wavelength PBL circulations are unrealistically amplified at higher resolutions also apply in this situation.

Fig. 22.
Fig. 22.

Time series of the maximum magnitude of vertical velocity (m s−1) in the lowest 15 model levels (approximately the lowest 3 km AMSL) in the area shown by the red rectangle in Figs. 19a and 19b. Simulations that were run with 5-min output frequency are shown with the solid lines, and simulations run with hourly output are shown by the dots.

Citation: Monthly Weather Review 143, 11; 10.1175/MWR-D-15-0179.1

Furthermore, the interaction between the surface boundaries and the waves appears to be more constructive in RT—Fig. 19b shows enhanced ascent where the dryline, front, and waves intersect—than in 2KM, which may also be a factor in whether or not CI occurs in west-central Oklahoma. In other words, even though the magnitude of the upward motion in a given wave at 2-km grid spacing is generally larger (Fig. 22), the upward branches of waves that intersect the dryline become stronger and more likely to initiate deep convection at 4-km grid spacing [akin to the processes at the intersection of drylines and rolls described by Xue and Martin (2006).] In 2KM, a similar region of constructive interaction appears, but much farther south in Texas near the 16°C dewpoint contour in Fig. 19d.

Finally, it remains somewhat unclear as to whether the waves seen in the model simulations shown here are representations of atmospheric processes that did indeed occur on 31 May–1 June 2013, whether they are spurious features arising from the numerical methods and approximations in the model, or some combination of both. Radar radial velocity imagery does show very subtle wavelike signatures in clear air in the transition zone with the same orientation as the modeled waves (not shown), although this is not conclusive evidence that such waves were observed in this case. The most likely scenario is that the environment was indeed favorable for waves trapped in the stable layer, and that the waves did indeed occur, but that the waves in the higher-resolution simulations had much higher amplitude and spatial coverage than any actual waves. Or in other words, although it is not possible to confirm with certainty that the RT simulation is producing the “right result for the right reasons,” the higher-resolution runs are clearly producing results that are less consistent with observations than those at 4- or 5-km grid spacing.

d. Evaluation of a nonlocal PBL parameterization

As noted above, Ching et al. (2014) found large differences in the representation of boundary layer circulations at higher resolutions, particularly in local PBL parameterizations. To examine whether the resolution dependence was also present in forecasts of the 31 May–1 June 2013 case using a nonlocal scheme, experiments analogous to RT and 2KM were conducted using the nonlocal YSU PBL parameterization. Neither of these simulations matched observations of the convective evolution and precipitation distribution as closely as RT did, with supercell storms developing but being displaced to the northeast of where they were observed (not shown). However, the PBL moisture distribution and dryline location was very similar between the YSU simulations with 4- and 2-km grid spacing (Fig. 23). [Consistent with the findings of Clark et al. (2015), the YSU dryline position was displaced farther to the east than that predicted by MYJ; cf. Figs. 17 and 23.] This corroborates the conclusion of Ching et al. (2014) that differences in PBL circulations owing to changes in resolution are most prevalent in local PBL parameterizations. In other words, although nonlocal schemes may be prone to larger overall errors in PBL moisture and dryline position, they appear to be less sensitive to changes in the model grid spacing.

Fig. 23.
Fig. 23.

As in Fig. 17c, but for the difference between simulations using the YSU PBL parameterization at 4- and 2-km grid spacing.

Citation: Monthly Weather Review 143, 11; 10.1175/MWR-D-15-0179.1

e. Influence of nested-grid configuration

Although a thorough examination of the causes of differences between the simulations with and without grid nesting is beyond the scope of this study, there were a few prominent discrepancies that may have led to different results. The spatial pattern of PBL-averaged dewpoint differences between 1.33KM and 1.33KM-nest was remarkably similar to the pattern of differences seen in Fig. 17 for RT minus 2KM (not shown). The differences, with 1.33KM having higher dewpoints than 1.33KM-nest (see also Fig. 16b) originate in west Texas at around 1800 UTC, and then move northeastward into Oklahoma. However, because these two simulations have equivalent grid spacing in this area, the differences cannot be attributed solely to the representation of gravity waves in the PBL and their effects on vertical mixing. Two possible sources of these differences were identified. First, the region of west Texas where these differences originate is near the southern lateral boundary for the inner grid in 1.33KM-nest (Fig. 9). At this boundary, subtle, but possibly important, differences in the mass and wind fields can be seen. For example, the mean sea level pressure field shows differences between the nested and nonnested runs of approximately 0.2–0.3 hPa along this boundary (not shown). Second, convection in Missouri in the first few hours of the forecast appears to affect remote regions differently depending on whether it is captured in the grid with the same resolution (as in 1.33KM) or whether it is on a coarser grid (as in 1.33KM-nest). Mesoscale gravity waves emanating outward from this convection similarly lead to pressure differences over west Texas by 1800 UTC that could conceivably alter the excitation of waves and the subsequent dryline development there. Previous studies have also found that in large-eddy simulations, different scales of PBL circulations may develop depending on whether the forcing comes from a coarser grid or from a grid at the same resolution (e.g., Moeng et al. 2007; Mirocha et al. 2014; Muñoz-Esparza et al. 2014). Identifying the underlying causes of these differences is beyond the scope of this study, but the results shown here do illustrate that such differences can be important for forecasts of deep convection.

5. Discussion

The degradation of convective forecasts with increased model resolution in this case represents a practical demonstration of the terra incognita problem for a high-impact weather forecast scenario. Past research (discussed in the introduction) suggests that, when aggregated over many forecasts, few substantive differences arise when going between model grid spacings of 4 and 2 km for parameters such as convective precipitation, but in some cases the forecast can be improved with higher resolution. On the other hand, the results of this study show that in some situations, increased resolution can actually degrade the forecast because of interactions between the representation of PBL processes and the initiation and upscale growth of convection. It is not known how often convection-allowing forecasts are degraded in this way, and in light of previous research findings it is likely an unusual occurrence. Any changes to model configuration will result in changes to the forecast outcome, and the rapid growth of errors on convective scales can make it difficult to identify whether those changes are systematic or simply a result of the limited predictability of deep convection. Yet because there is such a stark difference between the Δx = 4–5-km simulations and the Δx ≤ 3-km simulations, such strong similarities between the simulations within these subgroups, and PBL processes that are shown to be represented differently at higher versus lower resolutions, we can be more confident that these resolution sensitivities are not simply a product of the noise associated with limited predictability. The results provide reason for caution when reducing the grid spacing of a model while maintaining the same PBL parameterization, particularly for local PBL schemes, because reducing the horizontal grid spacing past approximately 4 km brings one more squarely into the terra incognita and may lead to unintended detrimental consequences.

As the explicit modeling and prediction of convection becomes more common, and efforts to accurately predict even smaller-scale phenomena continue, further research into understanding how PBL circulations, drylines (and other boundaries), and convection are simulated by numerical models is needed. In addition to the previously documented differences in the movement of drylines arising from the way in which the PBL is parameterized (e.g., Coffer et al. 2013; Clark et al. 2015), these results show that changing the model resolution can also alter dryline movement via changes in PBL processes. Future work to evaluate the behavior of numerical models in different convective situations could identify how often errors such as the ones highlighted in this manuscript occur, and would help to assess the needs for further improving parameterizations in research and operational models.

6. Conclusions

This study demonstrated that, for the supercell convection and MCS on 31 May–1 June 2013, reducing the grid spacing in a numerical model resulted in poorer forecasts of the initiation and upscale growth of convection. In a forecast with a real-time configuration (horizontal grid spacing of 4 km), the model initiated a supercell very close in space and time to the observed storm, and the convection later organized into a slow-moving, heavy-rain-producing MCS, again very similarly to observations. With higher model resolution (smaller grid spacing), however, the supercell failed to initiate, and a southward-moving squall line developed instead that produced much less rainfall. The differences were attributed to the representation of boundary layer processes to the southwest of the convection initiation region: at higher resolution, a series of gravity waves developed that increased the vertical mixing of moisture beyond that associated with daytime PBL growth, which in turn caused the dryline to move more quickly. With drier air at low levels in the higher-resolution simulations, the environment was less favorable for convection to initiate than in observations or the 4-km forecast. The structure and evolution of the convection during this event, and the resulting spatial distribution of precipitation, were very different between the higher- and lower-resolution simulations. Although the model behavior demonstrated in this manuscript is probably atypical and appears limited to “local” PBL parameterizations, the results provide further evidence that caution is warranted as modeling of convective and PBL processes move further into the “terra incognita.”

Acknowledgments

Discussions with Mike Coniglio, Adam Clark, Jack Kain, Stan Trier, John Peters, and Stacey Hitchcock were very valuable in synthesizing the results presented in this article. The constructive comments and suggestions of three anonymous reviewers and editor George Bryan helped to improve and clarify the manuscript. Thanks go to all involved in collecting radiosonde observations during MPEX, particularly John Peters and Adam Rydbeck who were responsible for data collection on 31 May. Computing resources were provided by Yellowstone (ark:/85065/d7wd3xhc) at the National Center for Atmospheric Research, which is sponsored by the National Science Foundation. National Center for Atmospheric Research. This research was supported by National Science Foundation Grants AGS-1157425 and AGS-1359727.

REFERENCES

  • Bryan, G. H., J. C. Wyngaard, and J. M. Fritsch, 2003: Resolution requirements for the simulation of deep moist convection. Mon. Wea. Rev., 131, 23942416, doi:10.1175/1520-0493(2003)131<2394:RRFTSO>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Burghardt, B. J., C. Evans, and P. J. Roebber, 2014: Assessing the predictability of convection initiation in the high plains using an object-based approach. Wea. Forecasting, 29, 403418, doi:10.1175/WAF-D-13-00089.1.

    • Search Google Scholar
    • Export Citation
  • Chen, F., and J. Dudhia, 2001: Coupling an advanced land surface-hydrology model with the Penn State–NCAR MM5 modeling system. Part I: Model implementation and sensitivity. Mon. Wea. Rev., 129, 569585, doi:10.1175/1520-0493(2001)129<0569:CAALSH>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Ching, J., R. Rotunno, M. LeMone, A. Martilli, B. Kosovic, P. A. Jimenez, and J. Dudhia, 2014: Convectively induced secondary circulations in fine-grid mesoscale numerical weather prediction models. Mon. Wea. Rev., 142, 32843302, doi:10.1175/MWR-D-13-00318.1.

    • Search Google Scholar
    • Export Citation
  • Cifelli, R., N. Doesken, P. Kennedy, L. D. Carey, S. A. Rutledge, C. Gimmestad, and T. Depue, 2005: The community collaborative rain, hail, and snow network: Informal education for scientists and citizens. Bull. Amer. Meteor. Soc., 86, 10691077, doi:10.1175/BAMS-86-8-1069.

    • Search Google Scholar
    • Export Citation
  • Clark, A. J., W. A. Gallus Jr., and T.-C. Chen, 2007: Comparison of the diurnal precipitation cycle in convection-resolving and nonconvection-resolving mesoscale models. Mon. Wea. Rev., 135, 34563473, doi:10.1175/MWR3467.1.

    • Search Google Scholar
    • Export Citation
  • Clark, A. J., and Coauthors, 2012: An overview of the 2010 Hazardous Weather Testbed Experimental Forecast Program spring experiment. Bull. Amer. Meteor. Soc., 93, 5574, doi:10.1175/BAMS-D-11-00040.1.

    • Search Google Scholar
    • Export Citation
  • Clark, A. J., J. Gao, P. T. Marsh, T. Smith, J. S. Kain, J. Correia Jr., M. Xue, and F. Kong, 2013: Tornado pathlength forecasts from 2010 to 2011 using ensemble updraft helicity. Wea. Forecasting, 28, 387407, doi:10.1175/WAF-D-12-00038.1.

    • Search Google Scholar
    • Export Citation
  • Clark, A. J., M. C. Coniglio, B. E. Coffer, G. Thompson, M. Xue, and F. Kong, 2015: Sensitivity of 24-h forecast dryline position and structure to boundary layer parameterizations in convection-allowing WRF Model simulations. Wea. Forecasting, 30, 613–638, doi:10.1175/WAF-D-14-00078.1.

    • Search Google Scholar
    • Export Citation
  • Coffer, B. E., L. C. Maudlin, P. G. Veals, and A. J. Clark, 2013: Dryline position errors in experimental convection-allowing NSSL-WRF model forecasts and the operational NAM. Wea. Forecasting, 28, 746761, doi:10.1175/WAF-D-12-00092.1.

    • Search Google Scholar
    • Export Citation
  • Coniglio, M. C., J. Correia Jr., P. T. Marsh, and F. Kong, 2013: Verification of convection-allowing WRF Model forecasts of the planetary boundary layer using sounding observations. Wea. Forecasting, 28, 842862, doi:10.1175/WAF-D-12-00103.1.

    • Search Google Scholar
    • Export Citation
  • Done, J., C. A. Davis, and M. Weisman, 2004: The next generation of NWP: Explicit forecasts of convection using the Weather Research and Forecasting (WRF) model. Atmos. Sci. Lett., 5, 110117, doi:10.1002/asl.72.

    • Search Google Scholar
    • Export Citation
  • Doswell, C. A., III, H. E. Brooks, and R. A. Maddox, 1996: Flash flood forecasting: An ingredients-based methodology. Wea. Forecasting, 11, 560581, doi:10.1175/1520-0434(1996)011<0560:FFFAIB>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Duda, J. D., and W. A. Gallus Jr., 2013: The impact of large-scale forcing on skill of simulated convective initiation and upscale evolution with convection-allowing grid spacings in the WRF. Wea. Forecasting, 28, 9941018, doi:10.1175/WAF-D-13-00005.1.

    • Search Google Scholar
    • Export Citation
  • Durran, D. R., 1990: Mountain waves and downslope windstorms. Atmospheric Processes over Complex Terrain, Meteor. Monogr., No. 23, Amer. Meteor. Soc., 59–81.

  • Hacker, J. P., 2010: Spatial and temporal scales of boundary layer wind predictability in response to small-amplitude land surface uncertainty. J. Atmos. Sci., 67, 217233, doi:10.1175/2009JAS3162.1.

    • Search Google Scholar
    • Export Citation
  • Iacono, M. J., J. S. Delamere, E. J. Mlawer, M. W. Shephard, S. A. Clough, and W. D. Collins, 2008: Radiative forcing by long-lived greenhouse gases: Calculations with the AER radiative transfer models. J. Geophys. Res., 113, D13103, doi:10.1029/2008JD009944.

  • Janjić, Z. I., 2002: Nonsingular implementation of the Mellor–Yamada level 2.5 scheme in the NCEP Meso model. NCEP Office Note 437, 61 pp.

  • Johns, R. H., and C. A. Doswell III, 1992: Severe local storms forecasting. Wea. Forecasting, 7, 588612, doi:10.1175/1520-0434(1992)007<0588:SLSF>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Kain, J. S., and Coauthors, 2008: Some practical considerations regarding horizontal resolution in the first generation of operational convection-allowing NWP. Wea. Forecasting, 23, 931952, doi:10.1175/WAF2007106.1.

    • Search Google Scholar
    • Export Citation
  • Kain, J. S., and Coauthors, 2013: A feasibility study for probabilistic convection initiation forecasts based on explicit numerical guidance. Bull. Amer. Meteor. Soc., 94, 12131225, doi:10.1175/BAMS-D-11-00264.1.

    • Search Google Scholar
    • Export Citation
  • Lemone, M. A., F. Chen, M. Tewari, J. Dudhia, B. Geerts, Q. Miao, R. L. Coulter, and R. L. Grossman, 2010: Simulating the IHOP_2002 fair-weather CBL with the WRF-ARW–Noah modeling system. Part II: Structures from a few kilometers to 100 km across. Mon. Wea. Rev., 138, 745764, doi:10.1175/2009MWR3004.1.

    • Search Google Scholar
    • Export Citation
  • Lin, Y., and K. E. Mitchell, 2005: The NCEP stage II/IV hourly precipitation analyses: Development and applications. 19th Conf. on Hydrology, San Diego, CA, Amer. Meteor. Soc., 1.2. [Available online at http://ams.confex.com/ams/pdfpapers/83847.pdf.]

  • Lock, N. A., and A. L. Houston, 2014: Empirical examination of the factors regulating thunderstorm initiation. Mon. Wea. Rev., 142, 240258, doi:10.1175/MWR-D-13-00082.1.

    • Search Google Scholar
    • Export Citation
  • Mellor, G. L., and T. Yamada, 1982: Development of a turbulence closure model for geophysical fluid problems. Rev. Geophys., 20, 851875, doi:10.1029/RG020i004p00851.

    • Search Google Scholar
    • Export Citation
  • Mirocha, J., B. Kosović, and G. Kirkil, 2014: Resolved turbulence characteristics in large-eddy simulations nested within mesoscale simulations using the weather research and forecasting model. Mon. Wea. Rev., 142, 806831, doi:10.1175/MWR-D-13-00064.1.

    • Search Google Scholar
    • Export Citation
  • Moeng, C.-H., J. Dudhia, J. Klemp, and P. Sullivan, 2007: Examining two-way grid nesting for large eddy simulation of the PBL using the WRF model. Mon. Wea. Rev., 135, 22952311, doi:10.1175/MWR3406.1.

    • Search Google Scholar
    • Export Citation
  • Morrison, H. G., 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, 9911007, doi:10.1175/2008MWR2556.1.

    • Search Google Scholar
    • Export Citation
  • Muñoz-Esparza, D., B. Kosović, J. Mirocha, and J. van Beeck, 2014: Bridging the transition from mesoscale to microscale turbulence in numerical weather prediction models. Bound.-Layer Meteor., 153, 409440, doi:10.1007/s10546-014-9956-9.

    • Search Google Scholar
    • Export Citation
  • Noh, Y., W. G. Cheon, S. Y. Hong, and S. Raasch, 2003: Improvement of the K-profile model for the planetary boundary layer based on large eddy simulation data. Bound.-Layer Meteor., 107, 401427, doi:10.1023/A:1022146015946.

    • Search Google Scholar
    • Export Citation
  • Nowotarski, C. J., P. M. Markowski, Y. P. Richardson, and G. H. Bryan, 2014: Properties of a simulated convective boundary layer in an idealized supercell thunderstorm environment. Mon. Wea. Rev., 142, 39553976, doi:10.1175/MWR-D-13-00349.1.

    • Search Google Scholar
    • Export Citation
  • NWS, 2014: Service assessment: May 2013 Oklahoma tornadoes and flash flooding. National Weather Service, 63 pp. [Available online at http://www.nws.noaa.gov/om/assessments/pdfs/13oklahoma_tornadoes.pdf.]

  • Peters, J. M., and R. S. Schumacher, 2015a: Mechanisms for organization and echo training in a flash-flood-producing mesoscale convective system. Mon. Wea. Rev., 143, 10581085, doi:10.1175/MWR-D-14-00070.1.

    • Search Google Scholar
    • Export Citation
  • Peters, J. M., and R. S. Schumacher, 2015b: The simulated structure and evolution of a quasi-idealized warm-season convective system with a training convective line. J. Atmos. Sci., 72, 1987–2010, doi:10.1175/JAS-D-14-0215.1.

    • Search Google Scholar
    • Export Citation
  • Potvin, C., and M. Flora, 2015: Sensitivity of idealized supercell simulations to horizontal grid spacing: Implications for warn-on-forecast. Mon. Wea. Rev., 143, 2998–3024, doi:10.1175/MWR-D-14-00416.1.

    • Search Google Scholar
    • Export Citation
  • Schumacher, R. S., and R. H. Johnson, 2005: Organization and environmental properties of extreme-rain-producing mesoscale convective systems. Mon. Wea. Rev., 133, 961976, doi:10.1175/MWR2899.1.

    • Search Google Scholar
    • Export Citation
  • Schwartz, C. S., and Coauthors, 2009: Next-day convection-allowing WRF model guidance: A second look at 2-km versus 4-km grid spacing. Mon. Wea. Rev., 137, 33513372, doi:10.1175/2009MWR2924.1.

    • Search Google Scholar
    • Export Citation
  • Skamarock, W. C., and Coauthors, 2008: A description of the Advanced Research WRF version 3. NCAR Tech. Note NCAR/TN-475+STR, 113 pp. [Available online at http://www.mmm.ucar.edu/wrf/users/docs/arw_v3_bw.pdf.]

  • Stensrud, D. J., and Coauthors, 2009: Convective-scale warn-on-forecast system: A vision for 2020. Bull. Amer. Meteor. Soc., 90, 14871499, doi:10.1175/2009BAMS2795.1.

    • Search Google Scholar
    • Export Citation
  • Thompson, R. L., B. T. Smith, J. S. Grams, A. R. Dean, and C. Broyles, 2012: Convective modes for significant severe thunderstorms in the contiguous United States. Part II: Supercell and QLCS tornado environments. Wea. Forecasting, 27, 11361154, doi:10.1175/WAF-D-11-00116.1.

    • Search Google Scholar
    • Export Citation
  • Torn, R. D., and G. S. Romine, 2015: Sensitivity of central Oklahoma convection forecasts to upstream potential vorticity anomalies during two strongly forced cases during MPEX. Mon. Wea. Rev., 143, 40644087, doi:10.1175/MWR-D-15-0085.1.

    • Search Google Scholar
    • Export Citation
  • Trapp, R. J., D. J. Stensrud, M. C. Coniglio, R. S. Schumacher, M. E. Baldwin, S. Waugh, and D. T. Conlee, 2015: Mobile radiosonde deployments during the Mesoscale Predictability Experiment (MPEX): Rapid and adaptive sampling of upscale convective feedbacks. Bull. Amer. Meteor. Soc., doi:10.1175/BAMS-D-14-00258.1, in press.

    • Search Google Scholar
    • Export Citation
  • Trier, S. B., G. S. Romine, D. A. Ahijevych, R. J. Trapp, R. S. Schumacher, M. C. Coniglio, and D. J. Stensrud, 2015: Mesoscale thermodynamic influences on convection initiation near a surface dryline in a convection-permitting ensemble. Mon. Wea. Rev., 143, 3726–3753, doi:10.1175/MWR-D-15-0133.1.

    • Search Google Scholar
    • Export Citation
  • Wakimoto, R. M., H. V. Murphey, E. V. Browell, and S. Ismail, 2006: The “triple point” on 24 May 2002 during IHOP. Part I: Airborne Doppler and LASE analyses of the frontal boundaries and convection initiation. Mon. Wea. Rev., 134, 231250, doi:10.1175/MWR3066.1.

    • Search Google Scholar
    • Export Citation
  • Weisman, M. L., W. C. Skamarock, and J. B. Klemp, 1997: The resolution dependence of explicitly modeled convective systems. Mon. Wea. Rev., 125, 527548, doi:10.1175/1520-0493(1997)125<0527:TRDOEM>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Weisman, M. L., C. Davis, W. Wang, K. W. Manning, and J. B. Klemp, 2008: Experiences with 0–36-h explicit convective forecasts with the WRF-ARW model. Wea. Forecasting, 23, 407437, doi:10.1175/2007WAF2007005.1.

    • Search Google Scholar
    • Export Citation
  • Weisman, M. L., and Coauthors, 2015: The Mesoscale Predictability Experiment (MPEX). Bull. Amer. Meteor. Soc., doi:10.1175/BAMS-D-13-00281.1, in press.

    • Search Google Scholar
    • Export Citation
  • Weiss, C. C., and H. B. Bluestein, 2002: Airborne pseudo-dual Doppler analysis of a dryline–outflow boundary intersection. Mon. Wea. Rev., 130, 12071226, doi:10.1175/1520-0493(2002)130<1207:APDDAO>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Wurman, J., K. Kosiba, P. Robinson, and T. Marshall, 2014: The role of multiple-vortex tornado structure in causing storm researcher fatalities. Bull. Amer. Meteor. Soc., 95, 3145, doi:10.1175/BAMS-D-13-00221.1.

    • Search Google Scholar
    • Export Citation
  • Wyngaard, J. C., 2004: Toward numerical modeling in the “terra incognita.” J. Atmos. Sci., 61, 18161826, doi:10.1175/1520-0469(2004)061<1816:TNMITT>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Xue, M., and W. J. Martin, 2006: A high-resolution modeling study of the 24 May 2002 dryline case during IHOP. Part II: Horizontal convective rolls and convective initiation. Mon. Wea. Rev., 134, 172191, doi:10.1175/MWR3072.1.

    • Search Google Scholar
    • Export Citation
  • Ziegler, C. L., E. N. Rasmussen, M. S. Buban, Y. P. Richardson, L. J. Miller, and R. M. Rabin, 2007: The “triple point” on 24 May 2002 during IHOP. Part II: Ground-radar and in situ boundary layer analysis of cumulus development and convection initiation. Mon. Wea. Rev., 135, 24432472, doi:10.1175/MWR3411.1.

    • Search Google Scholar
    • Export Citation
1

For many of the analyses shown herein, the 20°C dewpoint contour will be used as a simple illustration of the position of the surface boundaries, because it effectively shows the location of the stationary front and the leading dryline.

Save
  • Bryan, G. H., J. C. Wyngaard, and J. M. Fritsch, 2003: Resolution requirements for the simulation of deep moist convection. Mon. Wea. Rev., 131, 23942416, doi:10.1175/1520-0493(2003)131<2394:RRFTSO>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Burghardt, B. J., C. Evans, and P. J. Roebber, 2014: Assessing the predictability of convection initiation in the high plains using an object-based approach. Wea. Forecasting, 29, 403418, doi:10.1175/WAF-D-13-00089.1.

    • Search Google Scholar
    • Export Citation
  • Chen, F., and J. Dudhia, 2001: Coupling an advanced land surface-hydrology model with the Penn State–NCAR MM5 modeling system. Part I: Model implementation and sensitivity. Mon. Wea. Rev., 129, 569585, doi:10.1175/1520-0493(2001)129<0569:CAALSH>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Ching, J., R. Rotunno, M. LeMone, A. Martilli, B. Kosovic, P. A. Jimenez, and J. Dudhia, 2014: Convectively induced secondary circulations in fine-grid mesoscale numerical weather prediction models. Mon. Wea. Rev., 142, 32843302, doi:10.1175/MWR-D-13-00318.1.

    • Search Google Scholar
    • Export Citation
  • Cifelli, R., N. Doesken, P. Kennedy, L. D. Carey, S. A. Rutledge, C. Gimmestad, and T. Depue, 2005: The community collaborative rain, hail, and snow network: Informal education for scientists and citizens. Bull. Amer. Meteor. Soc., 86, 10691077, doi:10.1175/BAMS-86-8-1069.

    • Search Google Scholar
    • Export Citation
  • Clark, A. J., W. A. Gallus Jr., and T.-C. Chen, 2007: Comparison of the diurnal precipitation cycle in convection-resolving and nonconvection-resolving mesoscale models. Mon. Wea. Rev., 135, 34563473, doi:10.1175/MWR3467.1.

    • Search Google Scholar
    • Export Citation
  • Clark, A. J., and Coauthors, 2012: An overview of the 2010 Hazardous Weather Testbed Experimental Forecast Program spring experiment. Bull. Amer. Meteor. Soc., 93, 5574, doi:10.1175/BAMS-D-11-00040.1.

    • Search Google Scholar
    • Export Citation
  • Clark, A. J., J. Gao, P. T. Marsh, T. Smith, J. S. Kain, J. Correia Jr., M. Xue, and F. Kong, 2013: Tornado pathlength forecasts from 2010 to 2011 using ensemble updraft helicity. Wea. Forecasting, 28, 387407, doi:10.1175/WAF-D-12-00038.1.

    • Search Google Scholar
    • Export Citation
  • Clark, A. J., M. C. Coniglio, B. E. Coffer, G. Thompson, M. Xue, and F. Kong, 2015: Sensitivity of 24-h forecast dryline position and structure to boundary layer parameterizations in convection-allowing WRF Model simulations. Wea. Forecasting, 30, 613–638, doi:10.1175/WAF-D-14-00078.1.

    • Search Google Scholar
    • Export Citation
  • Coffer, B. E., L. C. Maudlin, P. G. Veals, and A. J. Clark, 2013: Dryline position errors in experimental convection-allowing NSSL-WRF model forecasts and the operational NAM. Wea. Forecasting, 28, 746761, doi:10.1175/WAF-D-12-00092.1.

    • Search Google Scholar
    • Export Citation
  • Coniglio, M. C., J. Correia Jr., P. T. Marsh, and F. Kong, 2013: Verification of convection-allowing WRF Model forecasts of the planetary boundary layer using sounding observations. Wea. Forecasting, 28, 842862, doi:10.1175/WAF-D-12-00103.1.

    • Search Google Scholar
    • Export Citation
  • Done, J., C. A. Davis, and M. Weisman, 2004: The next generation of NWP: Explicit forecasts of convection using the Weather Research and Forecasting (WRF) model. Atmos. Sci. Lett., 5, 110117, doi:10.1002/asl.72.

    • Search Google Scholar
    • Export Citation
  • Doswell, C. A., III, H. E. Brooks, and R. A. Maddox, 1996: Flash flood forecasting: An ingredients-based methodology. Wea. Forecasting, 11, 560581, doi:10.1175/1520-0434(1996)011<0560:FFFAIB>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Duda, J. D., and W. A. Gallus Jr., 2013: The impact of large-scale forcing on skill of simulated convective initiation and upscale evolution with convection-allowing grid spacings in the WRF. Wea. Forecasting, 28, 9941018, doi:10.1175/WAF-D-13-00005.1.

    • Search Google Scholar
    • Export Citation
  • Durran, D. R., 1990: Mountain waves and downslope windstorms. Atmospheric Processes over Complex Terrain, Meteor. Monogr., No. 23, Amer. Meteor. Soc., 59–81.

  • Hacker, J. P., 2010: Spatial and temporal scales of boundary layer wind predictability in response to small-amplitude land surface uncertainty. J. Atmos. Sci., 67, 217233, doi:10.1175/2009JAS3162.1.

    • Search Google Scholar
    • Export Citation
  • Iacono, M. J., J. S. Delamere, E. J. Mlawer, M. W. Shephard, S. A. Clough, and W. D. Collins, 2008: Radiative forcing by long-lived greenhouse gases: Calculations with the AER radiative transfer models. J. Geophys. Res., 113, D13103, doi:10.1029/2008JD009944.

  • Janjić, Z. I., 2002: Nonsingular implementation of the Mellor–Yamada level 2.5 scheme in the NCEP Meso model. NCEP Office Note 437, 61 pp.

  • Johns, R. H., and C. A. Doswell III, 1992: Severe local storms forecasting. Wea. Forecasting, 7, 588612, doi:10.1175/1520-0434(1992)007<0588:SLSF>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Kain, J. S., and Coauthors, 2008: Some practical considerations regarding horizontal resolution in the first generation of operational convection-allowing NWP. Wea. Forecasting, 23, 931952, doi:10.1175/WAF2007106.1.

    • Search Google Scholar
    • Export Citation
  • Kain, J. S., and Coauthors, 2013: A feasibility study for probabilistic convection initiation forecasts based on explicit numerical guidance. Bull. Amer. Meteor. Soc., 94, 12131225, doi:10.1175/BAMS-D-11-00264.1.

    • Search Google Scholar
    • Export Citation
  • Lemone, M. A., F. Chen, M. Tewari, J. Dudhia, B. Geerts, Q. Miao, R. L. Coulter, and R. L. Grossman, 2010: Simulating the IHOP_2002 fair-weather CBL with the WRF-ARW–Noah modeling system. Part II: Structures from a few kilometers to 100 km across. Mon. Wea. Rev., 138, 745764, doi:10.1175/2009MWR3004.1.

    • Search Google Scholar
    • Export Citation
  • Lin, Y., and K. E. Mitchell, 2005: The NCEP stage II/IV hourly precipitation analyses: Development and applications. 19th Conf. on Hydrology, San Diego, CA, Amer. Meteor. Soc., 1.2. [Available online at http://ams.confex.com/ams/pdfpapers/83847.pdf.]

  • Lock, N. A., and A. L. Houston, 2014: Empirical examination of the factors regulating thunderstorm initiation. Mon. Wea. Rev., 142, 240258, doi:10.1175/MWR-D-13-00082.1.

    • Search Google Scholar
    • Export Citation
  • Mellor, G. L., and T. Yamada, 1982: Development of a turbulence closure model for geophysical fluid problems. Rev. Geophys., 20, 851875, doi:10.1029/RG020i004p00851.

    • Search Google Scholar
    • Export Citation
  • Mirocha, J., B. Kosović, and G. Kirkil, 2014: Resolved turbulence characteristics in large-eddy simulations nested within mesoscale simulations using the weather research and forecasting model. Mon. Wea. Rev., 142, 806831, doi:10.1175/MWR-D-13-00064.1.

    • Search Google Scholar
    • Export Citation
  • Moeng, C.-H., J. Dudhia, J. Klemp, and P. Sullivan, 2007: Examining two-way grid nesting for large eddy simulation of the PBL using the WRF model. Mon. Wea. Rev., 135, 22952311, doi:10.1175/MWR3406.1.

    • Search Google Scholar
    • Export Citation
  • Morrison, H. G., 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, 9911007, doi:10.1175/2008MWR2556.1.

    • Search Google Scholar
    • Export Citation
  • Muñoz-Esparza, D., B. Kosović, J. Mirocha, and J. van Beeck, 2014: Bridging the transition from mesoscale to microscale turbulence in numerical weather prediction models. Bound.-Layer Meteor., 153, 409440, doi:10.1007/s10546-014-9956-9.

    • Search Google Scholar
    • Export Citation
  • Noh, Y., W. G. Cheon, S. Y. Hong, and S. Raasch, 2003: Improvement of the K-profile model for the planetary boundary layer based on large eddy simulation data. Bound.-Layer Meteor., 107, 401427, doi:10.1023/A:1022146015946.

    • Search Google Scholar
    • Export Citation
  • Nowotarski, C. J., P. M. Markowski, Y. P. Richardson, and G. H. Bryan, 2014: Properties of a simulated convective boundary layer in an idealized supercell thunderstorm environment. Mon. Wea. Re