Browsing by Author "Gary M. Lackmann, Committee Chair"

Now showing 1 - 8 of 8

Analysis of Model QPF Errors During the 2-4 December 2000 Snowstorm in North Carolina
(2006-01-31) Caldwell, Raymond Jason; Gary M. Lackmann, Committee Chair; Jerry M. Davis, Committee Member; Allen J. Riordan, Committee Member
Model forecasts of an early season snowstorm for 2 – 4 December 2000 followed the historical blizzard of 24 – 25 January 2000 that dumped 20.3 inches of snowfall at the Raleigh-Durham International Airport. Much like the January 2000 storm, operational models exhibited a significant lack of skill, particularly in the realm of quantitative precipitation forecasting. As early as 1800 UTC 1 December, operational models from the National Centers for Environmental Prediction (NCEP), including the 32-kilometer Eta, generated liquid equivalent precipitation totals approaching two inches for the Raleigh–Durham metropolitan area. Later forecasts indicated as much as 2.77 inches of precipitation would fall. In reality, only a trace of precipitation was observed at the Raleigh–Durham airport. A local, real–time version of the fifth-generation, mesoscale modeling system (MM5) was operational at the time of the event and provided a much-improved forecast scenario compared to the NCEP Eta model. Remarkably, the initial conditions and lateral boundary conditions in the MM5 were identical to those used to initialize the Eta model at 1200 UTC 2 December. In this study, an examination of the potential sources of error in the quantitative precipitation forecast is performed to challenge prior studies that suggest that data quality issues with sea surface temperature analyses led to spurious precipitation generation. The study includes a case study of the 2 – 4 December snowstorm, model sensitivity experiments, and quasi-geostrophic analysis to identify and diagnose the quantitative precipitation errors in the Eta model and the superior forecast guidance available from the local MM5 model. The case study showed that several potential sources of model error existed including missing upper air soundings, sea surface temperatures, model design, and misdiagnosed topographic flow. This study will test the hypothesis that errors at the 500–hPa level led to limited precipitation early in the period and, hence, produced errors in the cold air damming, coastal front, and cyclogenesis in later periods responsible for the heaviest precipitation in model forecasts. Results from sensitivity experiments with the MM5 model failed to exhibit significant differences in the representation of topographically induced phenomena or the westward extent of the precipitation shield into central North Carolina. The Eta model produced an anomalously strong 850 hPa jet at the North Carolina coast which transported warm air and moisture inland over the region. Better representation of the initial 500–hPa shortwave trough and associated vorticity maximum in the MM5 model is shown in the results to strengthen the low-level damming episode and shift the coastal front farther offshore. The results of this study provide basis for further investigation into both models and concludes that the effect of the upper–level forcing on the evolution of the low-level topographically induced flow and the surface–based forcing of upper–level dynamics can be of equal magnitude and importance in winter season precipitation forecasting. Results of this study will be coupled with local efforts to improve forecasting through conceptual model development by providing operational forecasting with the knowledge that individual models can have independent and opposing response to initial condition errors based on the physical and dynamical make–up of the mesoscale modeling system.
The Effect of Upstream Convection on Downstream Precipitation
(2005-07-22) Mahoney, Kelly Marie; Gary M. Lackmann, Committee Chair; Allen J. Riordan, Committee Member; Lian Xie, Committee Member
Numerical weather prediction (NWP) models have demonstrated a weakness in the ability to accurately forecast precipitation amounts downstream of strong, organized convection in the Southeast US. This weakness has also been communicated by operational forecasters, as past events have exhibited reduced downstream precipitation amounts relative to the model quantitative precipitation forecast (QPF), particularly when the upstream convection (UC) feature moves rapidly eastward. Conventional forecaster wisdom evolving from such events thus advocates reducing downstream QPF in the presence of UC. The purposes of this study are (i) to identify the physical processes by which downstream precipitation may be reduced or enhanced in the presence of UC; (ii) distinguish UC cases in which downstream QPF should be reduced from those in which it should be enhanced; (iii) understand why operational models are challenged to produce an accurate downstream forecast in these situations; (iv) identify synoptic settings associated with different types of UC events; (v) find ways in which human forecasters may anticipate and improve upon erroneous model forecasts; and (vi) investigate optimal model configurations for the representation of UC and downstream QPF. A brief climatology of National Centers for Environmental Prediction (NCEP) Eta model QPF error revealed that the UC problem is observed to occur in three relatively distinct synoptic settings, in which different QPF biases tended to occur. The three scenarios are based on the orientation and movement of the UC. Scenario 1 (S1) is characterized by UC that is oriented parallel to mean flow, and the organized convective system propagates quickly eastward relative to the primary synoptic system in a direction perpendicular to flow. These cases were generally characterized by a positive model QPF bias. Scenario 2 (S2) features UC that is oriented parallel to mean flow as in S1, but that propagates slowly (or not at all) with respect to the primary synoptic system. It was hypothesized that in these cases a diabatically-enhanced low-level jet (LLJ) would act to enhance moisture transport ahead of the synoptic system, and these cases were therefore hypothesized to yield increased downstream precipitation amounts, albeit with smaller QPF errors. Scenario 3 (S3) is characterized by UC oriented perpendicular to flow, usually aligning with a coastal and/or warm frontal features. S3 cases were found to often transition into an S1 or S2 event. They were consequently treated as precursors to S1 or S2 cases and are not investigated in great detail here. Several potential physical mechanisms of downstream precipitation alteration were examined for each scenario. These physical mechanisms include (i) moisture consumption; (ii) stabilization of the downstream environment; (iii) alteration of lower-tropospheric moisture transport through interruption (S1) or enhancement (S2) of the LLJ; and (iv) alteration of synoptic dynamics. Two main case studies were undertaken to investigate S1 and S2. For the S1 case it was found that a high-resolution (4-km grid-spacing) convection-resolving forecast using the Weather Research and Forecast (WRF) model vastly improved the downstream QPF relative to the large overprediction of precipitation produced by the operational Eta forecast. Comparisons were made between the successful WRF run, analyses, and the erroneous operational Eta forecast. Errors in the low-level wind and horizontal moisture flux fields from the operational Eta forecast revealed that the delayed speed of the model-forecasted UC did not permit an accurate representation of low-level moisture transport (physical mechanism (iii)), causing the model to allow too much moisture to infiltrate the downstream area, and likely contributing to the model overprediction of precipitation. The S1 case showed that despite excellent operational Eta forecasts of the placement and intensity of the synoptic system, errors in model-forecasted convective motion resulted in large operational QPF errors downstream. The S2 case study revealed that UC does not always lead to a decrease in downstream precipitation. Potential vorticity (PV) diagnostics were used to demonstrate that the diabatic influence of the UC acted to enhance the low-level jet (LLJ) and increase moisture transport (physical mechanism (iii)). A quasi-geostrophic PV inversion showed that the diabatic cyclonic QGPV anomaly associated with the UC feature contributed significantly to the southerly LLJ preceding the convective line. S2 proved to be an important counter-example to the notion that UC generally reduces downstream precipitation, as this case study demonstrates a physical process associated with the UC that instead acts to enhance downstream precipitation. The findings of this study may be helpful to forecasters in anticipating future UC events, as well as to numerical modelers facing decisions regarding what type of model run will yield the best results for future UC events. The results are further analyzed in the context of the ability of cumulus parameterization (CP) schemes to accurately represent convective movement, and the implications of such a shortcoming for UC forecasts. Future work in this area is suggested.
The Effects of Latent Heating on Cold Frontal Speeds and Accelerations from a Potential Vorticity Perspective
(2003-02-12) Reeves, Heather Dawn; Yuh-Lang Lin, Committee Member; Roscoe R. Braham, Committee Member; Gary M. Lackmann, Committee Chair
The effects of latent heating on frontal speed are investigated. It is conjectured that the existence of prefrontal latent heating leads to faster translation speeds and that the development of latent heating in the prefrontal zone can lead to frontal acceleration. A case study of a cold front where the attending precipitation band propagated into the prefrontal zone is presented. This front accelerated at the same time the precipitation moved into the prefrontal zone. Through inspection of the potential vorticity tendencies due only to latent heating, there is evidence that latent heating did alter the wind flow in the prefrontal zone, which may have contributed to positive frontogenetic tendencies in the prefrontal zone. A sensitivity test was conducted comparing a control simulation of the case study to a simulation ignoring the effects of latent heating and evaporative cooling (a 'fake dry simulation') for the same event. The front in the fake dry simulation moved slower than the front in the control simulation. This is in agreement with the hypothesis that latent heating leads to faster frontal translation speeds. However, the individual contributions of latent heating and evaporative cooling could not be determined from this experiment. An additional simulation which included the effects of latent heating, but not evaporative cooling was performed. Although the intensity of the front was considerably reduced in this simulation, the speed of the front was nearly identical to that in the control simulation: suggesting that latent heating effects are more important in dictating frontal speed than evaporative cooling effects.
The Formation and Impact of an Incipient Cold-Air Precipitation Feature on the 24-25 January 2000 East Coast Cyclone
(2005-07-22) Brennan, Michael Joseph; Gary M. Lackmann, Committee Chair; Allen J. Riordan, Committee Member; Lian Xie, Committee Member; Sethu Raman, Committee Member
The 24–25 January 2000 East Coast cyclone was characterized by a major operational forecast failure. In an effort to understand why short-range operational numerical weather prediction (NWP) model forecasts were so poor, the impact of a cold-air incipient precipitation (IP) feature that developed prior to the rapid cyclogenesis on 24 January is investigated using potential vorticity (PV) analysis. The IP was poorly forecasted by the operational NWP models, and these models failed to produce heavy precipitation far enough inland over the Carolinas and Virginia later in the cyclone event. Here the formation of the IP is examined from an observational perspective, the impact of the IP is quantified using PV methodology, and the ability of a NWP model to simulate its formation is tested by varying model physics, initial conditions and grid spacing. It was hypothesized that latent heating associated with the IP that formed over the Gulf Coast states early on 24 January generated a lower-tropospheric PV maximum that was important to the moisture transport into the Carolinas and Virginia and the track and intensity of the surface cyclone later in the cyclone event. Calculations from a PV budget and piecewise PV inversion found that the IP was associated with the genesis of a lower-tropospheric PV maximum and that the balanced flow associated with the PV maximum contributed significantly to moisture transport into the region of heavy snowfall. Operational NWP models that failed to forecast the IP did not generate the PV maximum or the heavy precipitation over the Carolinas and Virginia. Observational analyses and radar imagery showed that the IP formed in a region of elevated convective symmetric instability (a mixture of gravitational conditional instability and conditional symmetric instability) where forcing for ascent was provided by an approaching upper-level trough/jet streak. Short-range forecasts from NWP models under-forecasted the strength of the forcing and instability, and were unable to generate the IP in the region where it was observed. An 18-member mesoscale model ensemble with 20-km horizontal grid spacing varying initial condition analyses and model physics was unable to generate the IP feature. Variance associated with the cyclone?s sea-level pressure and precipitation distributions due to initial condition variation was larger than that due to variations in model physics, although significant variation was due to poor performance by ensemble members initialized from the Global Data Assimilation System analysis. A high-resolution model simulation with 4-km grid spacing showed that the IP initially formed within a layer of elevated CSI, consistent with analyses. Buckling of absolute geostrophic momentum surfaces indicated adjustment to slantwise convection at later times. Simulations with 12-km and 20-km grid spacing degraded the representation of these features, suggesting that models run with even coarser grid spacing would be unable to capture the initial formation of the IP. Other simulations initialized only three or six hours later showed a marked improvement in the representation of the IP, the cyclone track and intensity, and the final precipitation distribution, confirming the importance of properly representing the IP feature in successful simulations of this event. The current configuration of operational models with CP schemes and grid spacing insufficient to properly resolve the effects of slantwise convection suggests that future cases may occur where NWP models fail to capture the impact of a cold-air precipitation feature (possibly associated with elevated gravitational and slantwise instability), resulting in poor forecasts of downstream moisture transport and cyclone track and intensity. Operational forecasters should be aware of this possibility and be able to anticipate the potential feedbacks from precipitation (in NWP models and in reality) onto atmospheric dynamics. Available observations and high-frequency model analyses can be used to evaluate NWP model forecasts of precipitation and the lower-tropospheric PV distribution, allowing forecasters to recognize instances when model guidance can be adjusted to improve forecasts of high-impact cyclone events.
The Precipitation Mass Sink in Tropical Cyclones
(2004-07-22) Yablonsky, Richard Michael; Allen J. Riordan, Committee Member; Yuh-Lang Lin, Committee Member; Gary M. Lackmann, Committee Chair
Conservation of atmospheric mass is one of the fundamental concepts used in the study of meteorology. Any time precipitation occurs, however, atmospheric mass is not conserved. As precipitation is removed from the atmosphere, the hydrostatic pressure in the precipitating region is reduced. In a tropical cyclone, the heaviest precipitation occurs in the eyewall, and this localized region of heavy precipitation leads to mass and moisture convergence towards the center of the tropical cyclone. The Coriolis force deflects the converging air, thereby contributing to vorticity generation and increasing the cyclonic wind speed. Moisture convergence can also enhance precipitation. In addition, potential vorticity (PV) increases as a result of the precipitation mass sink. To assess the significance of the precipitation mass sink, several hypothesis tests are performed. The MM5 model is used to create a physically realistic dataset of Hurricane Lili (2002) from which mass and PV budgets can be performed. The mass budget reveals that in a 100-km radius cylinder around the model storm center, the 5-h total and 1-h average pressure-equivalent mass loss due to precipitation during forecast hours 30 to 35 are -7.25 hPa and -1.45 hPa h⁻¹, respectively, while the 5-h total and 1-h average model surface pressure change during that time are -2.29 hPa and -0.46 hPa h⁻¹, respectively. Although the surface pressure change in the cylinder is mostly due to large cancellation between strong low-level convergence and stronger upper-level divergence, the continuous removal of mass via precipitation represents a non-negligible effect in the mass budget. The PV budget reveals that in the same cylinder, the average hourly instantaneous mass sink PV tendency during forecast hours 30 to 35 is 0.42 PVU day^-1, while the average hourly instantaneous diabatic PV tendency is -2.19 PVU day⁻¹. The diabatic PV tendency exhibits large spatial and temporal cancellation, so the small but continuously positive PV contribution from the precipitation mass sink has a non-negligible effect on the PV budget. In addition, according to the PV tendency terms, an air parcel rising through the troposphere in the eyewall should experience nearly continuous PV generation via the precipitation mass sink but both PV generation and destruction via latent heat release, leading to large cancellation from the latter. Therefore, the parcel upon reaching the upper troposphere can likely attribute a non-negligible amount of its PV to the precipitation mass sink, but trajectory computations would be needed to quantify the relative contributions of the PV tendency terms on a given parcel. In addition to the mass and PV budgets, the workstation version of the Eta model is used to perform multiple sensitivity experiments with and without the precipitation mass sink. The most realistic of these sensitivity experiments reveals that the precipitation mass sink generally reduces the central pressure by ~5-7 hPa, increases the wind field by ~5-15 kt, and increases the precipitation rate by ~5-25 mm h⁻¹, but the rainfall rate difference in particular exhibits large spatial and temporal variation. PV and geopotential height cross sections show maximum precipitation mass sink-induced PV increase (>5 PVU) and geopotential height reduction (>4 dam) near the surface and near the melting layer. The results of this study suggest that the precipitation mass sink should not be neglected in tropical cyclones. Further research possibilities include detailed analysis of the impact of the precipitation mass sink on the tropical cyclone track, as well as the importance of the precipitation mass sink in heavily precipitating systems other than tropical cyclones.
Quantitative Precipitation Forecast Sensitivity to Microphysics Parameterization and Sea Surface Temperature Source over North Carolina during Two Cold Season Events
(2007-05-16) Haglund, Nicole Lynne; Brad S. Ferrier, Committee Member; Sandra E. Yuter, Committee Member; Gary M. Lackmann, Committee Chair
In the southeastern United States, some of the most dramatic model quantitative precipitation forecast (QPF) failures in recent years have been associated with winter precipitation events. For example, the Eta model predicted nearly three inches of total liquid equivalent precipitation over most of central and eastern North Carolina for 2-3 December 2000, while less than 0.10 in. (2.54 mm) of liquid equivalent precipitation actually fell over the majority of central North Carolina. While the over-prediction of precipitation for the 21-22 January 2003 event was not as significant, the predicted precipitation nevertheless might have led to a higher impact case, if it had verified. Despite a forecasted liquid cloud with cloud top temperatures warmer than -15°C, the Eta model produced excessive QPF for both cold season events. The purposes of this study are (i) to determine whether sea surface temperature data source (1° by 1° weekly Reynolds SST vs. 1.27-km CoastWatch daily SST) could have significantly impacted the 2-3 December 2000 QPF; (ii) to test sensitivities associated with the Ferrier microphysics scheme by studying the effects of various ice nucleation and total glaciation temperatures on QPF; and (iii) to investigate sensitivity of QPF to sea surface temperature data and to choice of microphysics scheme to determine which change yields a more significant contribution to QPF differences. In an effort to understand why the Eta model over-predicted precipitation in the 2-3 December 2000 and 21-22 January 2003 winter events, sensitivity tests were conducted using the Weather Research and Forecasting model (WRF). These sensitivity studies included testing the QPF differences due to choice of microphysics parameterization scheme and to choice of sea surface temperature (SST) data source for the 2-3 December 2000 case, while only the sensitivity of QPF to choice of microphysics parameterization scheme was tested for the 21-22 January 2003 case. It was hypothesized that by cooling the ice nucleation and total glaciation temperatures, better QPF (less precipitation with a cooler ice nucleation temperature, more precipitation with a cooler total glaciation temperature) would result in both cases. Since the cloud top temperature in the 21-22 January 2003 case was below the original total glaciation temperature (-10°C), cooling the total glaciation temperature was not expected to change the QPF. Additionally, for the 2-3 December 2000 case, it was hypothesized that changes in SST data source would have a greater impact than the choice of microphysics parameterization scheme on QPF. Major findings in this study include: (i) Surface low tracks and total precipitation patterns were not significantly different between the runs using Reynolds SST and CoastWatch SST data; (ii) By cooling the ice nucleation temperature in both case studies, better (closer to analyzed) QPF resulted in the 21-22 January 2003 case. With a cooler total glaciation or ice nucleation temperature in the 2-3 December 2000 case, no clear QPF difference pattern emerged; (iii) While a qualitative analysis of total liquid-equivalent precipitation differences between SST data source and microphysics parameterization scheme runs indicated that SST data source had a greater impact on QPF than choice of microphysics scheme, area-averaged total liquid-equivalent precipitation in three regions showed that choice of SST data source led to QPF biases on the same order of magnitude as the QPF biases due to choice of microphysics scheme; and (iv) Since there are more similarities between simulations run with a particular version of WRF (V2.1.2 vs. a subsequent version) than there are between the two versions using any microphysics scheme, choice of microphysics scheme has less of an impact on QPF than convective parameterization (CP) scheme activity for the winter storm cases studied here.
The Role of the Great Lakes in Northwest Flow Snowfall Events in the Southern Appalachian Mountains
(2007-11-06) Holloway, Blair Sterling; Yuh-Lang Lin, Committee Member; Sethu Raman, Committee Member; Gary M. Lackmann, Committee Chair
Northwest flow snowfall (NWFS) events are a regional forecasting challenge that affects much of the southern Appalachian Mountains. These events can be defined as snowfall accompanying upslope flow and low-level northwesterly winds in this region, and typically feature irregular snowfall distributions and highly variable total accumulations. Previous research done by Perry and Konrad (2004—2007) provides an excellent climatology of NWFS events, and shows that NWFS accounts for nearly 50% of mean annual snowfall along the higher elevations of the southern Appalachians. Additionally, through analysis of backward air parcel trajectories, their research shows that NWFS events that featured a Great Lakes connection exhibited increases in composite mean and maximum snowfall totals. This body of work clearly suggests that the Great Lakes can enhance snowfall in NWFS events by warming and moistening the low-level airmass upstream of the southern Appalachians. The specific objective of this study is to quantify and evaluate the role of the Great Lakes in NWFS events for select cases via model experiments using the Weather Research and Forecast (WRF) model. The selected cases occurred 5–6 March 2001, 18–20 December 2003, and 10–11 February 2005, and were investigated using a case study approach. In order to determine the effect of the Great Lakes on NWFS precipitation in these cases, two experimental runs were designed to isolate the role of the lakes. First, surface fluxes of heat and moisture were set to zero across the entire model domain (NOFLX). Second, surface fluxes of heat and moisture were set to zero across only water points (LKNOFLX). The sensitivity of the selected NWFS events to planetary boundary layer (PBL) scheme was also tested (MYJPBL). Overall, it was found that the Great Lakes play an important role in some NWFS events and can be responsible for 20–30% of the precipitation that occurs in these events. Of the selected cases, the March 2001 and February 2005 events showed large decreases in precipitation in the LKNOFLX model run compared to the control (CTRL) run. In these two events, the role of the Great lakes was to destabilize the upstream airmass and increase the Froude number. At a point roughly halfway between the Great Lakes and the southern Appalachians, the LKNOFLX model run in the February 2005 event had an average 950?850 hPa Froude number of 0.99, which was 0.40 less than the CTRL value of 1.39. Similarly in the March 2001 event, the LKNOFLX model run had an average 950–850 hPa Froude number of 1.28, which was 0.42 less than the CTRL value of 1.70. In both cases, the reduced average low-level Froude number in the LKNOFLX run compared to the CTRL shows that when the effect of warming and moistening of the low-level upstream airmass caused by the Great Lakes is removed, a more stable upstream airmass occurs which reduces the Froude number and reduces NWFS precipitation.
Sensitivity of WRF Simulations of Hurricane Ivan to Horizontal Resolution
(2007-08-01) Gentry, Megan Suzanne; Gary M. Lackmann, Committee Chair; Fred Semazzi, Committee Member; Anantha Aiyyer, Committee Member
As finer resolutions become possible in numerical modeling, it has become increasingly common to turn off the cumulus parameterization scheme in favor of explicit simulation of convection. To the author's knowledge, the grid spacing at which it is appropriate to do so in a tropical cyclone (TC) case has not been systematically investigated. Therefore, this study examines the sensitivity of explicit model simulations of Hurricane Ivan (2004) to changes in horizontal grid spacing, when grid spacing between 12 and 2 km is used. As grid spacing decreases, the minimum central pressure of Ivan deepens, dropping by approximately 20 hPa as grid spacing decreases from 4 to 2 km. However, the 8-, 6-, and 4-km simulations have intensity differences of only around 10 hPa between them. The structure shown by model-simulated radar, as well as model-simulated satellite infra-red (IR) temperatures, shows that the eyewall of the coarser resolution simulations (12- to 6-km) is highly asymmetrical and elliptically-shaped, with two large maxima (minima) in reflectivity (cloud top temperature) rotating about the TC center. The 4- and 2-km runs have more circular eyewalls, with more numerous and larger maxima (minima) in reflectivity (cloud top temperature) embedded within the eyewall, as well as better developed spiral bands. Temporal and spatial averaging, done at a given radius over azimuth, show the system-averaged quanitites in cross-section and reveal differences in the structure of the TC core and eyewall. The finer resolution simulations have larger updrafts and more subsidence within the eye. However, the warming of the eye, relative to the other runs, is confined to the upper levels of the troposphere. The eyewall of the TC in the finer resolution runs slopes radially outward less with height, as the horizontal temperature gradient changes little with height, compared with the coarser simulations. This lack of warming in the lower- and mid-levels of the TC eye indicates a ventillation mechanism at work in the finer resolution runs, acting to mix high potential temperature (θe) air from the eye into the eyewall. Such air could act as a fuel source for buoyant convection within the eyewall (Persing and Montgomery 2003; Eastin et al. 2005b; Yang et al. 2007). Fine-scale eyewall and eye features are examined at high temporal resolution in order to further analyze changes in the TC structure as grid resolution increases. Wind, θe, and potential vorticity (PV) anomalies in the finer resolution simulations tend to be smaller in size and larger in magnitude, especially in the 2-km simulation. The PV field in the 2-km simulation appears to have several wave-like features moving throughout the eyewall, suggesting that smaller-scale processes, such as vortex Rossby waves (VRWs) and buoyant convection, areat least partially resolved at this grid spacing. VRWs, waves that propagate along a PV gradient, are further explored as a possible ventillation mechanism acting in the lower TC eye. The presence of VRWs is tested by visual analysis, as well as by a subjective estimate of the motion of PV features and a PV budget. Both of these analyses show the properties of these PV features to be consistent with the theoretical and observed properties of VRWs. A spectral decomposition of kinetic energy shows that the higher resolution simulations distribute energy to specific wavenumbers where organized wave motions are simulated. However, the coarser runs distribute lower amounts of power over more wavenumbers, some of which are not even fully-resolved at that grid spacing. There is some convergence in the model solution for the basic TC structure and intensity at grid spacings between 8- and 4-km, suggesting that these grid spacings might be appropriate for an operational NWP environment. For research purposes, where the time needed for numerical integration is less constrained, 4-km is the largest grid spacing that could be considered appropriate to partially resolve physical process within the eyewall. However, as the minimum central pressure of the 2-km simulation is significantly deeper than all other simulations, small-scale physical processes important to the intensification.