6.1 MCSs Vary in Scale, Models Vary in Resolution

It should be clear to you now that MCSs come in many sizes from small, isolated bow echoes to systems as large as a very long squall line or MCC. Mesoscale models are typically thought of as those with 30 kilometer grid spacing or less. All other factors being equal, the higher the model resolution, the better it should be at representing MCS structures. But because most operational models are notoriously poor at predicting the onset and exact location of convection, they are in many ways unreliable for predicting MCSs. For example, convective parameterization schemes vary in their ability to generate convective downdrafts and mesoscale cold pools. Important details such as these are tremendously significant in determining any particular model’s ability to forecast MCS evolution.

In this hypothetical example, a model with 30 kilometer gridpoint spacing may initially resolve a midsized MCS; however, it is clear that with this kind of resolution the model cannot sufficiently depict significant details such as the updrafts, downdrafts, and other embedded precipitation processes to properly forecast the MCS's continued evolution.

6.2 MCS Processes and Model Performance

Model errors forecasting convection and MCSs arise from at least two common sources. One source involves the “initial conditions.” If there are errors in the initial conditions, then it is very difficult to correctly forecast convective details. The second obvious source comes about as a result of having to “parameterize” convective processes in all hydrostatic mesoscale models and even in non-hydrostatic models with grid spacing of more than a few kilometers. Convection and other important non-hydrostatic precipitation processes in MCSs occur on scales much smaller than can be resolved by most operational mesoscale models. Thus, these processes are parameterized to simulate their effects on the larger-scale environment. Parameterization of these details is better than not trying to account for them at all, but is still quite problematic and can result in significant departures from reality when it comes to forecasting thunderstorms with models.

6.3 Limitations of Convective Parameterization Schemes

It is important to know and understand the limitations of any particular model’s convective parameterization scheme. Generally speaking, the stronger the large-scale forcing, the better the convective parameterization will do with the timing, location, and amount of heavy precipitation. However, if there is elevated instability and the environmental forcing is strong, sometimes the model convective parameterization will not be able to keep up with the supply of moisture and lift, resulting in the grid-scale precipitation scheme trying to make a cumulonimbus with an updraft as wide as an entire grid box! Obviously such a wide, intense updraft produces too much latent heating, too much precipitation, too low surface pressures, and other inappropriate feedback.

Consider these examples from the 22 kilometer operational Eta Model.  It began making precipitation bull's-eyes over the plains in mid-July 2001 and continued generating such episodes even after the 24 July 2001 model upgrade. These precipitation bull's-eyes with amounts over four inches, mostly in a six-hour period, were being generated primarily by the grid-scale scheme in high-CAPE environments with strong mesoscale forcing. These forecast events were spurious and unphysical but occurred under conditions favoring strong convective systems with heavy rains. These grid-scale storms became the model's focal point for convergence, vertical motion, and instability release, causing the model to miss the forecast where the MCS actually occurred. 

6.4 What Can Go Wrong?

There are five aspects of convection to address in a forecast: 1) convective initiation – including both timing and location of the onset of convection, 2) convective system evolution – including the size of the area affected and the propagation and duration of the system, 3) the total amount of precipitation, 4) the accompanying weather hazards, and 5) the impact on other downstream weather elements such as clouds and temperatures.

The accuracy of the first three of these factors in the model forecast is affected by the synoptic scenario, the accuracy of the initial conditions, the resolution of the model, and the particular convective parameterization used.

The fourth factor isn’t predicted by any NWP model.

The fifth factor is predicted to some extent by the models, but is limited by errors in the second factor.

No parameterization is routinely “better” than another. Each has a different mix of strengths and weaknesses in handling these different aspects of the forecast. None is consistently good.

In contrast, a model forecast of important features such as fronts, jet streaks, capping inversions, and the presence of a sea breeze or dryline will be more accurate than the model forecast of convection itself (although feature position may be off).

Thus, you should rely more heavily on the forcing and inhibiting features predicted by the model than on the convection forecasts themselves.

6.5 Review Question: Factors Affected by the Synoptic Scenario

The accuracy of which of these factors in the model forecast is affected directly or indirectly by the synoptic scenario? Select all that apply.

a) Hail size
b) Convective initiation
c) Total amount of precipitation
d) Maximum outflow winds
e) Impact on downstream temperatures
f) Convective system evolution

Review Question Feedback

The correct answers are b), c), e), and f).

The accuracy of convective initiation, convective system evolution, and total amount of precipitation is directly affected by the synoptic scenario. Impact to downstream weather elements, including temperature, is predicted to some extent by the models. However, accuracy is limited by errors in predicting convective system evolution. Accompanying weather hazards, such as hail size or maximum outflow winds, are not predicted by NWP models.

6.6 Review Question: Model Forecast Adjustments

Imagine you are forecasting for a location downstream of an MCC that is currently weakening. Model fields show development of a mid- to upper-level circulation center associated with the MCC propagating over your area within the next 24 hours, along with moderate-to-heavy precipitation amounts. However, there is no evidence on satellite and radar observations that a vortex exists and no other features are evident to help focus or enhance the precipitation. What adjustments to the model forecast may be necessary? Select all that apply.

a) Decrease forecast precipitation amounts

b) Increase forecast precipitation amounts

c) Decrease cloud cover and increase daytime temperatures

d) Increase cloud cover and decrease daytime temperatures

Review Question Feedback

The correct answers are a) and c).

The formation of a mid- or upper-level circulation in the model is a clue that the model’s convective parameterization scheme kicked in and created the vortex in response to the latent heat being released by the convection in the model. Even though the convective parameterization scheme may have “functioned” correctly, the resulting circulation may be an artifact propagated downstream in the model that is acting to generate the subsequent cloudiness and precipitation in your area. If you’re sure the vortex isn’t really there, then it is probably appropriate to decrease forecast precipitation amounts and expected cloud cover and to increase daytime temperatures (as long as other precipitation-enhancing factors are not present).

6.7 What about ~ 10 km Resolution?

Click the image above to view animation.

Most MCSs are of sufficient size to be depicted but not fully resolved by a 10 kilometer model. You may recall from the “How Mesoscale Models Work” module that an NWP model can show the presence of a feature five grid spacings across, or in this case, 50 km.  However, a feature would have to be 7 to 10 grid spacings in size to be more properly handled. Thus, a 10 kilometer model with suitable physical parameterizations can generate many features of a realistic mesoscale convective system. It is important to remember that even at 10 kilometer grid spacing, the model timing and placement of convective initiation will vary depending on the convective parameterization being used. In fact, the timing and track of model MCS forecasts have little skill. Sometimes they’re good, sometimes they are not, but they are definitely not reliably good! One major contributor to poor MCS evolution forecasts is the inability of even a model of this resolution to properly generate and evolve the system gust front and cold pool. However, at 10 km resolution, models may do a good job of signaling the likelihood of an MCS in the general region.

6.8 MCS Prediction with Non-Hydrostatic Models

Only in the recent history of NWP models have computational resources and new non-hydrostatic model code allowed routine investigation and modeling of convection at extremely high resolution without convective parameterization schemes. Examples of this new generation of storm-scale models with grid spacing of no greater than 5 km include ARPS (the Advance Regional Prediction System) and some configurations of MM5 and WRF (the Weather Research and Forecasting model). Studies examining ARPS’ explicit convection (in other words, without parameterizing convective processes) indicate that storm-scale models with state-of-the-art model physics can do a credible job of modeling convective storms. With 2 kilometer grid spacing, the ARPS model has clearly demonstrated the ability to generate new cells along a gust front and depict MCS evolution. However, even these cutting edge models are much better at forecasting convection initiation and evolution in strongly forced weather patterns than they are in the weaker, less obvious cases. It is important to remember that despite the remarkably realistic-looking detail produced by these advanced models, accurately forecasting the placement, timing, and amount of precipitation depends on details in the initial conditions that often are not adequately observed.

6.9 Review Question: Convective Parameterization

True or False? Convective parameterization in the WRF model is routinely better than that in the Eta Model.

• True
• False

Review Question Feedback

The correct answer is "False." No convective parameterization is routinely better than another.

6.10 Review Question: Comparing Models

Although higher-resolution models provide more detail, we must be careful when using them to forecast convection. As an illustration, let’s compare reflectivity forecasts from the 10-km and 4-km WRF models to the radar loop showing what actually occurred in the central U.S. during the night of 25 June 2003. Please review the data below and answer the question that follows.

Click the image above to view animation.

Click the image above to view animation.

How well did the models do? Select all correct answers.

a) The 4 kilometer WRF was better than the 10 kilometer WRF in forecasting the squall line from eastern Missouri to northwest Texas.
b) Both models missed the convection over Illinois.
c) The 10 kilometer WRF was better than the 4 kilometer WRF on forecasting disorganized thunderstorms over the Texas Panhandle.
d) Both models were correct in forecasting bow echoes along the line from Arkansas to the Red River area of Texas/Oklahoma after 0600 UTC.
e) The 10 kilometer WRF was better than the 4 kilometer WRF on timing of the line in Oklahoma and northwest Arkansas at 0600 UTC.

Review Question Feedback

The correct answers are a), b), and d). This example demonstrates that higher resolution models are not necessarily perfect. They may often be good at forecasting the anticipated storm structures and mode of convective organization, but can, with great detail, miss the actual location or timing.

6.11 Using Mesoscale Models to Forecast MCSs: A Summary

At this point, there are several reasonable approaches for applying mesoscale models to MCS prediction. The first and most important is to use model output to look for favorable synoptic and mesoscale patterns and associated buoyancy and shear profiles conducive to MCS formation. Of course, care must always be taken to be alert for synoptic positioning errors and particular kinds of model biases.

(View in-depth discussion on this topic below.)

Second, the model parameterizations themselves may yield useful diagnostic fields. For instance, the operational Eta and RUC produce fields of convective cloud tops. Glancing at those can give a general idea of where there is a risk that if convection occurs, it will be tall rather than low-topped. The Kain-Fritsch parameterization scheme outputs a convective mass flux field that indicates how vigorous it expects the convection to be. Where the mass flux is near the maximum value, cloud-scale updrafts may be expected to be unusually strong and severe weather is possibly more likely, even if the model-predicted precipitation is light.

Third, forecasters should watch out for exceptionally large vertical velocities colocated with very heavy precipitation in model forecast runs that use convective parameterization. Such features indicate that the model is trying to make an unphysical large-scale cumulonimbus! However, it also means you need to inspect the situation carefully because even if the model feature has large errors, an MCS may be likely in the general vicinity.

And fourth, studies have shown that high-resolution mesoscale models without convective parameterization (less than or equal to 6 kilometer grid spacing) are quite good at forecasting the occurrence of convection in a region, though not at a particular point. They are even pretty good at forecasting the mode of convection in strongly forced situations. This means a well-constructed model of sufficient resolution can give good guidance as to whether storms will be isolated, multicellular, or evolve into larger convective systems. Again, this is especially true in strongly forced weather patterns. The good news for forecasting MCSs is that the larger the system, the more likely the model is to forecast it, even out to 48 hours in advance. Unfortunately, in the weakly-forced scenarios, which are more difficult for humans to predict, the models also have a much more difficult time. In these scenarios forecasters must place more emphasis on conceptual models and pattern recognition.

In conclusion, the best advice is to realize that if the highest-resolution mesoscale model is predicting a large area of long-lived convection, the situation likely warrants close watching by you, the forecaster, even though the predicted timing and location of the system may be off.

End of Section 6.0: MCSs and NWP