Notes for Convective Weather Refresher
Usage recommendation: Print these notes, then view the presentation in "Slide Show" mode using the notes as a guide. The notes here are identical to those attached to each slide in the Powerpoint file.
|
Slide, topic |
Explanation (self-explanatory pages and self-evident main points omitted) |
|
4 Background |
Here is an example of what happens when the convective parameterization fails to relieve instability fast enough. On the far left is a model sounding showing a saturated layer with a lapse rate steeper than moist-adiabatic. This layer is produced by strong mesoscale convergence and lift in the resolved grid-scale motions and is a characteristic signature of such lift in a moist environment. The plot next to it shows vertical velocity at the same time in green, an hour later in blue, and two hours later in magenta. The saturated unstable layer is at the top of the region of strong ascent. The model continues to lift this region, forming a grid-scale cumulonimbus rising over several hours. The extreme vertical velocities clearly violate the model’s hydrostatic assumption, making the pressure and velocity forecast evolution physically unrealistic. The strong convergence and vertical motion bulls-eye results in a locally-intense vorticity maximum. On the US map, the pre-storm ambient vorticity is shown, with typical values of 3 to 12 x 10-5 s-1. The upper zoomed-in area shows 3-hour precip contours and shaded vorticity (same color scale) at the time of the sounding shown in the left-most plot, which was for a grid point to the southeast of the precipitation bulls-eye at that time. The vorticity has now increased to over 20 x 10-5 s-1. The lower zoomed-in area shows the same for 3 hours later, with the vorticity maximum exceeding 39 x 10-5 s-1 co-located with the precipitation bulls-eye near the location of that sounding. Precipitation at that sounding location is listed in the table. The convective parameterization did trigger, but it could not relieve instability fast enough to prevent the model from making a thunderstorm updraft core on the grid scale. Most of the precipitation coming from the grid-scale scheme in a convective situation is a classic characteristic of a model grid-scale blow-up. A grid-scale thunderstorm is desirable in a very high resolution nonhydrostatic model (the storm would grow more rapidly and have a narrower updraft), but is exactly what must be avoided in operational NWP models that are not at high enough resolution to resolve a convective updraft. The purpose of the convective parameterization is to avoid this scenario. |
|
5 Background |
Note how the sharpness of the trough as well as the wind speeds and CAPE are different in these two model runs. The models were the same except for the convective parameterization. The model convection affected not only the precipitation but also the model mass, horizontal wind, and vertical motion fields. |
| 6 Background |
The
skill of model precipitation forecasts in convective situations is poor,
even for rain vs. no rain, not only for the amount of rain.
The numbers shown here use the Equitable Threat Score (ETS), which essentially measures the overlap between the predicted and observed areas of precipitation exceeding the specified threshold amount over the specified time period. A score of 1 is for perfect overlap, zero is the same as a random chance (no skill) forecast, and negative is worse than a random forecast. A threshold of .01” corresponds to a rain/no-rain forecast. The ETS numbers are very sensitive to the length of time period considered (e.g. precip falling over a 3-, 6-, or 24- hour period), but are far less sensitive to resolution, convective parameterization, and even which model is used. However, not using a convective parameterization at resolutions of 10 to 30 km is worse—the score gets cut in half. ETS is a very limited verification metric—it shows
no skill for a perfectly detailed squall line forecast that is off by
a few hours or a few grid boxes, while it shows some skill for a much
wider broad-brush smoothed squall line that has some overlap with the
sharper observed squall. As models move to higher resolution and can provide
more detailed and useful information even when point values are incorrect
(or their timing is incorrect), other measures for assessing model skill
and model utility are needed. |
|
7 Background |
Suppose a single grid column is the size of the red box. This red box includes an entire MCS with leading convective line, trailing stratiform region, and mesovortex, as well as some isolated cells ahead of the MCS. A climate model grid box is this large. Are the two major assumptions listed above appropriate? It looks fine, though if the convective system slides over so it straddles the grid box line, it could be a problem. |
|
8 Background |
Now, take the same picture but consider a smaller grid box, say the size used in the GFS or first few days of the global ensembles, or even the SREF. Here we show two possible locations for a grid box of around that size. In the one on the right, the individual cells are entirely confined to the grid box, and any compensating subsidence in their immediate vicinity fits in too. So that looks like it meets the assumptions just fine. In the grid box in the middle of the picture, the convective updraft and any subsidence under the forward anvil overhang is contained in the grid box, but the MCS itself is divided across grid boxes. If the model grid-scale motions could generate the mesoscale features, which include the rearward ascending current creating the stratiform rain part of the MCS and the descending rear inflow current that contributes to evaporative cooling creating the mesohigh and gust front, then the convective parameterization only needs to represent the part of the MCS covered by the leading line convective towers. Thus, the MCS representation gets split into a portion taken care of by the convective parameterization and a portion taken care of by the grid-scale resolved motions and grid-scale microphysics parameterization. However, the resolution is too coarse to fit enough grid boxes across the MCS to resolve it well, so the model will do a poor job trying to account for the part not handled by the convective parameterization. |
|
9 Background |
Further shrink the grid boxes, say to the scale of the operational Eta or RUC. Now how do the assumptions look? For the grid box on the far right, the entire cell does fit in the grid box, but it now occuppies a significant fraction of the grid box area, and the grid box may not encompass the whole region of significant compensating subsidence. The middle grid box faces a similar predicament to that discussed on the previous slide, except now the advance of the gust front clearly needs to be handled by the grid-scale motions. If the convective parameterization can stimulate the model grid-scale variables to generate a gust front and if the convective parameterization trigger function responds to the gust front, then the model grid-scale variables can provide a link between adjacent columns and enable the MCS to propagate forward. Unfortunately, no convective parameterizations exist that can reliably stimulate such behavior in the grid-scale variables. The left-most grid box illustrated is within the MCS stratiform region and slices through the mesovortex. The convective parameterization should be doing nothing here, but should have stimulated the grid-scale variables to generate the mesoscale flows through the MCS to make a stratiform region. The resolution is fine enough that the mesoscale flows through a large MCS can be resolved, though they will only be generated if the convective parameterization and other parts of the model, including especially the grid-scale microphysics parameterization, interact well. |
|
10 Background |
As a result of assumption 2), all convective parameterizations have difficulty with MCS propagation. As will be seen later, the RUC has a trick for getting around this, but the jury is still out on how well it works. Research models with no convective parameterization have the model dynamics and vertical motion explicitly and properly coupled with the convective heating, and as a result do a far better job with propagation even at resolutions of several km (still too coarse to resolve cloud details and updraft cores), though they are prone to overpredict system intensity and longevity. |
11 Background |
Physical processes that trigger convection in nature operate on smaller scales than can be resolved in NWP models. The convective parameterization has no forecast data on such processes, so it has to make a guess. Boundary layer rolls have been observed to play a key role in initiation of surface-based convection in nature. This plot shows boundary layer rolls in a model with 1 km grid spacing over a section of the domain 30 km across. Imagine all of this actually happening within one model grid box where the model has one sounding and one wind profile and one vertical velocity profile. The parameterization would have no information about whether these rolls exist or how strong they are or whether they are merging anywhere to lift parcels to their LFC. Convective parameterizations are similarly handicapped for any other initiation mechanism also unresolved in an NWP model. Basically, we are asking it to perform magic, inventing information where it has none, so we ought to be more surprised by its successes than its failures. |
|
12 Background |
NWP models have a hierarchy of forecast skill for different variables on different scales. As a forecaster, you should make the most use of those parameters most skillfully predicted by the model and should not rely on poorly predicted parameters.
Therefore, to make an accurate forecast of convection, great attention must be paid to observations, including mesoscale structures that may not be correctly represented in the models. You must synthesize your own understanding of the processes at work in the atmosphere and compare that to the model forecast evolution. Again, rely most heavily on the model fields that have best skill, after correcting for discrepancies against observations and observed trends, and do not rely on model fields with poor skill. |
|
13 Table |
Items in this table will be discussed in more detail in individual slides
next, going through the table column by column. The schemes are summarized
for intercomparison here, with color coding indicating red flags, yellow
caution, and green for good. The parameterization schemes range from simplest
on the left to most complex on the right. The two simplest schemes are
adjustment schemes, pushing the model grid column toward a specified temperature
and moisture profile when convection is triggered. The other three schemes
are mass flux schemes, which use a 1-dimensional cloud model to determine
the temperature and moisture changes in each model layer in the column.
The last row indicates whether the precipitation generated in convective
situations in the model shows up primarily in the output fields of convective
precipitation or grid-scale precipitation, as will be discussed in more
detail later. |
|
14 NGM-Kuo |
Triggering essentially requires deep moisture convergence with any—small
or large—amount of instability. The scheme does not check for a
CIN. But when there is sufficient deep moisture convergence for the model
to trigger convection, there is usually sufficient vertical motion to
lift out a cap if a cap was present. |
|
15 NGM-Kuo |
Simplest possible temperature adjustment you could think of—slap
on a moist adiabat for the unstable parcel and push the pre-convective
sounding a fraction of the way toward the moist adiabat, same fraction
in each model layer. But remember, the model grid box needs to represent
all the air around the storm, not just the convective cloud. |
|
16 NGM-Kuo |
Kuo schemes are famous for making artificial wave-CISK, which is essentially a propagating wave that couples vertical motion, convergence, and convective heating. This results in model convection propagating away from its source region (at the wrong speed if it should propagate), and it affects the pattern of convergence and vertical motion in the forecast. |
|
17 Eta-BMJ |
The BMJ scheme looks at precipitable water in the layer between the LCL and the equilibrium level and it requires some (small or big, any amount) of instability. Thus, soundings with lots of mid-level moisture content or rich moisture extending well above the LCL are favored to trigger. Also, it only checks over the lowest 200 hPa, so it may miss elevated convection. |
|
18 Eta-BMJ |
Instead of pushing the preconvective sounding toward a moist adiabat as the Kuo scheme does, BMJ creates a “reference profile” that differs slightly from a moist adiabat. It starts with a first guess reference profile, shown here, and then modifies it as shown on the next slide. Note that the reference profile starts at cloud base for the unstable parcel – it does not directly affect air below the LCL. The temperature profile below the freezing level has a lapse rate slightly steeper than moist adiabatic, while above the freezing level the profile is a little more stable than moist adiabatic. The moisture profile is set so that a parcel in the reference sounding at the level of the convective cloud base has a deficit saturation pressure (DSP) of 39 hPa, meaning it has to rise 39 hPa to reach saturation, with DSP also specified at the freezing level and the convective equilibrium level. (These DSP values are reduced (moister profile) for nearly saturated preconvective soundings and when convection has been going for a while at the same location.) Thus, convection pushes the model toward a sounding that is not saturated, with temperature-dewpoint spreads of a few degrees and the driest air at the freezing level. |
|
19 Eta-BMJ |
The first guess reference profile in the previous slide is adjusted to account for making sure the amount of heating and drying in the sounding corresponds to the amount of precipitation dropped. The precipitation is simply the difference in precipitable water between the preconvective and post-convective soundings. The amount of latent heat released from condensing water to make this amount of precipitation must correspond to the net amount of heating in the column between the preconvective sounding and the reference profile. The first guess profile might not satisfy this constraint, so it needs to be adjusted, as shown on the next slide. |
|
20 Eta-BMJ |
This slide and the three that follow form a sequence illustrating the adjustments to the first-guess profile. The profiles essentially slide to colder or warmer temperatures until the net heating in the column corresponds to the amount of precipitation produced through drying. If the preconvective sounding is dry above the LCL of the unstable parcel, no fit will be found, because the reference profile will be too moist unless the temperature profile is pushed far to the left on the skew-T to cold temperatures. In that case, the scheme fails to trigger deep convection on the basis that there is not enough moisture to support convection. The precipitation produced comes from moisture above the LCL—it does not tap moisture from below the LCL. In
this slide, the profiles have been shifted far to the left on the skew-T.
If these profiles were used, the post-convective sounding would be far
drier, thus a lot of precipitation would be dropped, but the temperatures
would also be colder, inconsistent with the required heating. |
|
21 Eta-BMJ |
Since the profiles were too cold and dry, let’s see what happens if we
slide them to the right to make them warmer and moister. Now, the postconvective
sounding has more precipitable water than the preconvective sounding.
This violates the physical principle of conservation of water and is not
allowed. |
|
22 Eta-BMJ |
Reducing the temperatures and dewpoints of the reference profile somewhat still leaves a mismatch. The precipitable water is slightly reduced from the preconvective amount, which allows a little precipitation to be dropped. But the amount of heating in the column is more than would be created by condensing out that small amount of rain. |
|
23 Eta-BMJ |
Cooling (sliding to the left) a little more, the profiles are adjusted just right. Now, the amount of heating in the column is consistent with the amount of precipitation dropped as the precipitable water is removed from the grid column during convection. This is the profile which the BMJ scheme will use. This process shows that different preconvective soundings will get different matching BMJ reference profiles, but all the reference profiles look fundamentally similar. |
24 Eta-BMJ |
The
reference profile state would be achieved in roughly 40 minutes, except
that new profiles are computed every 9 minutes based on the updated sounding.
The important result for interpreting model output is that soundings resembling
reference profiles actually show up in the model forecast. The scheme does not produce deep convection if any of the following occur:
In these cases, the scheme switches to “shallow convection” instead of deep convection. Shallow convection sometimes has a large, incorrect effect on forecasts of surface dewpoints, CAPE, and CIN as explained by the diagram, which shows a temperature and dewpoint sounding represented by the black curves on the left and physical processes represented by the schematic on the right. The problem comes in communication between boundary layer turbulent mixing and the shallow convection, which is trying to extend the boundary layer upward due to small or moderate cumulus or stratocumulus. Turbulent mixing is parameterized using turbulent kinetic energy (TKE), which is generated by slightly superadiabatic lapse rates, such as from surface heating during the day, and also can be generated by strong shear. TKE mixes surface heating and moistening through the depth of the boundary layer as shown by the red adjustments to the black sounding. When the shallow convection turns on, it transfers heat downward from above the top of the boundary layer, as shown by the blue adjustments to the black sounding. This downward heat transfer and upward moisture transfer is an attempt to emulate cumulus-generated eddies entraining warmer potential temperatures into the boundary layer and the evaporation of the cumulus themselves, cooling and moistening the air above the boundary layer. These changes weaken or lift the capping inversion and reduce the lapse rate in the upper portion of the surface-based mixed layer, inhibiting TKE mixing at that level. The TKE mixing shuts down in the upper part of the boundary layer, leaving further influx of moisture trapped in the lower part of the boundary layer, which results in forecast soundings with too much CAPE and not enough CIN. |
|
25 Eta-BMJ |
These soundings show the impact of shallow convection on soundings in the Eta model. The yellow sounding in the upper plot shows well-mixed moisture and winds through a deep boundary layer. A short distance away at the same time, it has the moisture trapped in a thinner layer with higher dewpoints, and the winds show more directional shear. In the lower diagram, we see a similar result comparing the Eta run with the BMJ parameterization in red (temperature) and green (dewpoint) to the Eta run with the Kain-Fritsch scheme (purple). Note the sounding in the BMJ run looks like the moisture should be mixed over a deeper layer than it is, the cap has been lifted, and the winds show a little more directional turning than in the KF run. The cap held and the CAPE in the BMJ run was too high. Other convective parameterizations and other models also have shallow convection, but this problem with boundary layer mixing does not usually happen in other operational models nor in the Eta using the KF parameterization, though it has shown up in an NCAR version of WRF (not operational). However, when the Eta model is run without shallow convection, its precipitation verification scores deteriorate. Some changes are being worked on, and if successfully tested, will get implemented in the fall 2004 Eta change package or may instead work their way into the nonhydrostatic model that will replace the Eta by the end of 2005. |
|
26 Eta-BMJ |
In addition to items discussed earlier, the BMJ scheme in the Eta model tends to trigger too early in the day and it also tends to trigger the first few hours of the forecast, drying out the model atmosphere as water is rained out. The remainder of the forecast period has a dry bias. This sequence is illustrated by the green observed precipitation (note diurnal cycle) and blue predicted precipitation in the graphic. Also, because its trigger is sensitive to deep moisture rather than to CIN, the BMJ scheme tends to trigger nearly simultaneously over a broad area. Replacing preconvective structure with reference soundings over a region has two effects:
Lastly, the unsaturated
reference profiles inhibit interaction with the grid-scale microphysics,
though often the result is moistening in the upper troposphere, which
is the desired trend. Remember, to effectively reproduce an MCS in a model
with 12-km grid spacing, the model needs to explicitly generate the trailing
stratiform region and associated mesoscale circulations, which requires
interaction between the convective and grid-scale schemes. |
|
27 Eta-KF |
The KF scheme averages model variables over layers 50 hPa thick to obtain potential source parcels for convection. Since model layers near the ground may be only a few hPa in thickness, the KF source parcels may be more representative of the air going into a storm than simply taking the most unstable model layer. The triggering
function responds to any amount of CAPE—large or small—but
only allows initiation where CIN is small. Even with CAPE present, it
does not trigger where the resolved vertical motion is downward unless
the sounding has a superadiabatic layer. It only checks for CAPE over
the lowest 200 hPa, so it may miss elevated convection. This 200 hPa search
depth can easily be changed for versions of the KF scheme used in other
models such as MM5; here we are reporting only on the implementation at
NCEP. |
|
28 Eta-KF |
The KF scheme, like all mass flux schemes, assumes the changes to grid column soundings are caused by air detrained out of convective updraft cores, compensating subsidence in the environment, and convective downdraft outflow. It ignores the contribution from updraft air, on the assumption that updraft cores occupy only a tiny fraction of the grid box area. The
KF scheme assumes the cloud contains various mixtures of parcel and environment
air. These mixtures have various equilibrium levels at which they detrain.
It uses a one-dimensional cloud model to determine the resulting cloud
properties and detrainment amounts at all levels as well as the amount
of moisture condensed into precipitation. |
|
29 Eta-KF |
Unlike adjusting toward a moist adiabat or reference profile, the sounding is not pushed toward any specified profile. Instead, tendencies are calculated based on the warming or cooling (and drying or moistening) rates from the detrainment, compensating subsidence, and convective downdrafts. This leaves a lot more degrees of freedom for how the sounding is changed than when a profile is specified. Therefore, in any situation, it can be much harder to recognize the influence of model convection on the sounding. Typically, the main impacts are slight warming and drying at mid-levels and pronounced cooling at the bottom. These temperature and moisture tendencies are multiplied by an amount that, if nothing else happened except convection, would remove all the CAPE from the source layer in 30 minutes. |
|
30 Eta-KF |
The sounding changes actually seen in the model are very strongly affected by the dynamic response of the model to the convective heating. The red and green temperature and dewpoint profiles are the same as shown on the previous page, while the purple and blue temperature and dewpoint profiles are soundings actually seen in a model run. The deep upward motion acting over hours creates a deep saturated profile, stimulating grid-scale precipitation. |
31 Eta-KF |
Eta model runs for a squall line case using two different convective parameterizations are shown in this and the next 5 slides as a series of soundings and wind profiles at the same location where the squall line passed through. The run using the BMJ parameterization is shown in green while the KF is shown in blue. The image in this
slide is for a 12-hour forecast, valid in the morning. At this time there
already is considerable difference in moisture at 500-300 hPa due to upstream
convection in the model and differences in upper-tropospheric winds, but
the soundings are very similar between 1000 and 500 hPa. |
32 Eta-KF |
Four
hours later, KF is maintaining a capping inversion and has not convected
yet at this location, while BMJ looks like it is convecting weakly. |
| 33 Eta-KF |
Now
at 18 UTC in the middle of the day, the run with KF is still not convecting
and is still capped, allowing more surface heating to build up more CAPE.
The BMJ is already convecting, producing soundings that look like BMJ
reference profiles. Also, note the wind differences at all levels, including
more backed winds in the boundary layer in KF. |
|
34 Eta-KF |
One hour later, the cap in the KF is eroding but still holding while moisture advects in aloft. The BMJ parameterization is still convecting. Wind differences aloft are increasing. |
|
35 Eta-KF |
At 21 UTC, the KF run is now convecting. Note the cooling near the surface. Also, the run using KF has pumped up a lot of moisture through the column – the moistening is due to the dynamic response, not the direct changes from the convective parameterization. The BMJ is still convecting and still showing a reference profile. |
|
36 Eta-KF |
Twelve hours after the first slide in this series, the convection has
now ended in both runs, and the stronger winds in the BMJ have advected
in drier air behind the squall line, while in KF this dry air lags further
to the west. Recall, at the start of the series, the soundings were nearly
identical below 500 hPa and different above. Now the moisture and winds
are different throughout, but the temperatures are nearly identical above
500 hPa except for the tropopause break being sharper in the KF run. |
|
37 GFS-SAS |
Moving on to our next scheme… the Simplified Arakawa Schubert used in the GFS takes the most unstable parcel and entrains environment air up to the LFC. It then calculates the CAPE of this mixture. It also considers a downdraft starting at the level of minimum wet bulb potential temperature and finds the downward CAPE of the downdraft. It takes the sum of the updraft and downdraft CAPE and checks to see if it exceeds a threshold value. This threshold value varies as a function of cloud-top height and resolved vertical motion, as shown in the graphic. Note that the threshold value is higher for deep clouds, meaning that a tall cumulonimbus won’t be triggered unless the CAPE is higher, while low-topped convection can be triggered with less CAPE. The cloud-top level is not taken at the parcel equilibrium level; rather, it is taken as a random function that varies between mid-levels and the equilibrium level, effectively creating a sort of ensemble of cloud heights in an area spanning several grid boxes and over time. Also note that the CAPE threshold deceases as vertical velocity is more negative (upward). Thus, convection is triggered with less CAPE in regions of large-scale ascent. Also, because the CAPE is computed for a mixture that includes environment air up to the LFC, if the LFC is above a capping inversion with dry air in the capping layer, then the mixture will have far less CAPE than a pure parcel of near-surface air; thus the scheme will be less likely to trigger. |
|
38 GFS-SAS |
Conceptually, SAS is similar to KF except that it detrains only at one
level—at the cloud top. Like in KF, the grid column is assumed to
be affected by detrainment from the convective tower (though now only
at the top), compensating subsidence, and the convective downdraft dumping
into the low levels. In the GFS, the winds are also changed through these
same three mechanisms, as indicated on the diagram. Additionally, the
GFS evaporates some of the precipitation on its way down, starting at
the top. |
|
39 GFS-SAS |
As with KF, the convective parameterization changes the model soundings
in a complex manner with no characteristic recognizable signature except
cooling at low levels. The magnitude of the changes is determined such
that CAPE would be reduced to the threshold value in roughly 30 minutes.
Also as with KF, the changes caused by the model’s dynamic response can
be quite large and different than the changes produced directly from the
convective parameterization. |
|
40 GFS-SAS |
SAS in the GFS tends to respond inadequately in cases of strongly focused forcing with strong ascent impinging upon an unstable region. The model then makes an enormous grid-resolved cumulonimbus the size of one or several states, such as the outline of high RH in the middle cross section. The tick marks are 2 degrees of latitude, with total distance across the cross section being more than 1500 km. The top panel shows a vertical velocity couplet and a region of high vorticity, exceeding 30 x 10-5s-1, below the level of and slightly lagging the vertical motion peak. The bottom panel shows the characteristic marker of this model phenomenon – most of the very heavy precipitation is falling from the grid-scale scheme (difference between the curves) rather than the convective parameterization (red curve). Every convective parameterization fails in this manner once in a while, but it has been a particularly frequent and vexing problem in the GFS, as the processes producing this problem seem to help the model in predicting tropical cyclones. The size of the resulting grid-scale disturbance and its longevity and synoptic impact are reduced as model resolution is made finer. |
|
41 RUC |
The RUC actually uses an ensemble of approaches rather than a single scheme. Each ensemble member triggers differently and produces a different set of changes to the sounding. However, unlike making an ensemble of different model runs that each would evolve toward different forecasts, the parameterization combines all the results into one prediction for convection at each time step. For now, all the convective parameterization variations are averaged at equal weight. Work is underway to objectively determine more optimal weights so that the variations most appropriate for a given situation can be weighted more heavily in the average than members generally less skillful in that situation. That is a major motivation for going to this ensemble approach, but it is a long term project that may take several years. In the meantime, equal weighting of all members will be used. Like SAS in the GFS, the trigger requires CAPE for a mixture of environment and low-level unstable air. Like KF, the CIN is explicitly considered among the trigger requirements, though instead of using the magnitude of the CIN, it uses the depth of the CIN layer. Additionally, triggering requirements used in various other schemes are included for some members. |
|
42 RUC |
In the background section at the beginning of this presentation, we noted
that parameterizations consider each grid column independently, making
prediction of MCS propagation difficult. Some members of the RUC convective
parameterization ensemble enhance their response when there is convection
in the upstream grid box. The enhancement is stronger when the downdraft
is stronger, and the downdraft is stronger when shear is stronger. It
isn’t clear whether this makes a large improvement in predicting MCS propagation.
So you may want to watch how the RUC does this summer, and if it seems
to be doing pretty well over a series of cases, weigh that in your considerations. |
|
43 Real Case |
Now we shall examine an actual case of a strongly-forced squall line, the sort of case we would expect the models to perform better than in an average case. This will give a sense of how the characteristics compare for precipitation produced using the different parameterizations and also how the parameterizations affect the model soundings. This slide shows observed one-hour accumulated precipitation, from the Stage IV multisensor product, which is a quality-controlled blend of radar and gauge data. An 18 UTC sounding was taken at Slidell, LA, ahead of the squall line. The forecasts will begin at 12 UTC this same day, so we will be looking at 0-6 hour forecasts. Note that one-hour amounts exceed 25 mm along the squall line from southwest LA to central MS. The forecast plots in the next slides will show 3-hour accumulations. Therefore, the squall line swaths should not appear quite as narrow. However, since the squall line was moving slowly (e.g., compare the positions for the hour ending at 15 UTC with that at 18 UTC), we’d still expect a good forecast of 3-hour accumulation to show an elongated linear pattern of heaviest precipitation. |
|
44 Real Case |
The NGM is too fast and the maximum amounts are too light, but considering its very coarse resolution, it is showing a somewhat elongated feature with the right orientation. Most of the precipitation is falling from the convective parameterization. It has considerable instability over southern LA, where it failed to produce much convection, though the squall line was observed to pass through that area. |
|
45 Real Case |
The operational Eta made a diffuse area of light precipitation, giving no indication of a squall line. The output grid used is 20 km, so coarse-resolution output is not the cause of the diffuse pattern shown. Almost all of the precipitation over LA/MS/AL is from the convective parameterization. It has CAPE over southern and central LA greater than 500 J/kg with small CIN, CAPE of greater than 1000 J/kg in places, but it did not lead to convection in the model. Much of where the model did produce convection was too far east – it triggered too early. Remember, this poor prediction is valid at only 6 hours into the forecast! The blue dot in the upper left panel shows the location of the Slidell 18 UTC sounding. The model produced some convection there. We will also look at the model sounding at Tuscaloosa, AL, indicated by the blue + symbol, where the model produced more convective precipitation. |
|
46 Real Case |
The GFS produced a squall-line-like signature but with the heavier precipitation occurring too far north and the heavy precipitation coming entirely from the grid-scale scheme. GFS output is now available on a ½-degree grid every 3 hours for the first 36 hours. The upper left panel shows the precipitation on the 1-degree grid that has for years been the standard output grid from NCEP. The upper right shows the same but on the higher-resolution ½-degree grid. The model is the same at T254, just different output grids. The increase in detail depicted on the higher resolution grid is coming entirely from the grid-scale precipitation area, not where the model has convection. The convective parameterization appears to have triggered over the right area but didn’t produce enough rain. An area of moderate CAPE and small CIN remains over southeast LA after the model convection has been going. The location of the 18 UTC Slidell sounding is shown with the blue dot in the upper left panel while the location of a more active grid point to be examined shortly is shown by the blue + symbol. |
|
47 Real Case |
The RUC did a good job showing an elongated linear feature, but its amounts are too light. Hourly grids are available for the RUC, so one-hour total QPF is shown in the upper right, with corresponding lower amounts than the three-hour totals. The output grid is 40 km, which explains amounts less than the peak observed amounts shown from the 4.8-km Stage IV data but not as low as shown here. The progression of the band is too slow in MS but looks well-placed in LA. Note that most of the precipitation is from the grid-scale scheme except along the leading edge, where amounts are lighter and most are from the convective parameterization. The RUC shows a large swath of CAPE exceeding 500 J/kg, almost all with small CIN, and CIN has been completely removed over southwest MS. The blue dot in the upper left panel again represents Slidell, while model soundings will also be examined for the location of the blue + symbol, where the model convection was more active. |
|
48 Real Case |
These are Eta forecast soundings at Slidell, LA, with the observed 18 UTC sounding in yellow on the left. The green is the forecast valid at 18 UTC in both panels, while the magenta is at 15 UTC and the cyan at 17 UTC. At 18 UTC, the Eta overall fits the shape of the observed sounding though it is too cold at the very bottom, too warm up to 800 hPa, and too dry between 700 and 600 hPa. However, most pronounced is the signature at 15 UTC – it looks like a BMJ reference profile, and sure enough, the model was convecting. The mid-level drying trend seen from 15 to 17 to 18 UTC is from advection. The moist reference profile from 15 UTC has also advected downstream. |
|
49 Real Case |
The left panel shows Eta forecast soundings at Tuscaloosa, AL at 12, 15, and 18 UTC in green, cyan, and magenta, respectively. The sounding starts out dry aloft and by 15 UTC has moistened up with something somewhat resembling a reference profile for convection rooted around 900 hPa. Over the next three hours, the primary change is further warming and moistening in the middle and upper troposphere. Did model convection cause these changes? The right panel shows hourly convective heating – the hourly change in temperature profile caused by the convective parameterization alone. The times are the ending times of the one-hour periods, with shades of green for ending times of 13, 14, and 15 UTC and shades of blue for 16, 17, and 18 UTC. It shows that the model convection was warming the sounding from around 900 to 500 hPa by 1°C/hour the first three hours of the forecast (green curves) – it triggered right away and kept going and going though rather weak (1 °C/hour is not strong for convection). Then it increased to 2 °C/hour in the 700-600 hPa layer for each of the next three hours (blue curves). However, the sounding did not warm up that much – vertical motion compensated the warming. The vertical motion was caused by the model dynamics, some from the synoptic wave and perhaps some additional from the response to the convective and grid-scale heating distribution. The yellows and oranges are heating rates from the grid-scale precipitation parameterization, showing the generation of a stable precipitation anvil above the convection, supported by vertical motion supplying the source of water. Heating rates reached 3 °C/hour at 18 UTC, and it was this heating above the convective equilibrium level that was responsible for the warming of the forecast soundings above 500 hPa. Also, note the grid-scale scheme has negative values – cooling – below 600 hPa. The cooling spike just above 700 hPa is from the melting of falling grid-scale precipitation – notice that it is at or just slightly below the 0 °C level in the soundings. |
|
50 Real Case |
On the left are GFS soundings at Slidell at 15 UTC in cyan and 18 UTC in green, along with the 18 UTC raob in yellow. The forecast is too warm below 900 hPa and too moist above 600 hPa but otherwise fits the raob pretty well. The forecast soundings changed little between 15 and 18 UTC except moistening above 500 hPa and warming above 350 hPa, neither of which is likely to be due to the convective parameterization. As shown a few slides earlier, convection did occur in the model, but as we noted before, sounding changes in the GFS are dominated by other processes. On the right are GFS soundings at Jackson, MS at 12 (magenta, initial time), 15 (cyan), and 18 (green) UTC. Jackson was in a region of heavier model precipitation between 15 and 18 UTC, with some coming from the grid-scale scheme. The primary change over time is deep moistening by 15 UTC, all the way to the grid-scale precipitation scheme’s threshold for condensation, which is 90% RH with respect to ice for this sounding above the freezing level. As convection occurs and the model drops a half inch of precipitation at the ground between 15 UTC and 18 UTC, the sounding changes little except for cooling and drying around 700-600 hPa. |
|
51 Real Case |
The left panel shows observed (yellow) and RUC forecast (green) 18 UTC soundings at Slidell. The six-hour forecast is too moist above 600 hPa and misses the inversion at 600 hPa at the base of the dry layer. The right panel repeats the 18 UTC forecast sounding in green and shows the 15 UTC and 17 UTC soundings in magenta and cyan, respectively. It shows that the inversion and dry layer were captured better in the 3-hour forecast, and it appears that ascent has lifted and weakened the inversion and moistened the dry layer. The model did produce some weak convection at Slidell during this time, but the only hint of it in the soundings is the slight warming between 500 and 300 hPa; other processes dominate the sounding changes. |
|
52 Real Case |
The left plot shows soundings for a location in northeast MS at 16, 17, and 18 UTC in green, cyan, and magenta, respectively. The convective parameterization was active only between 16 and 17 UTC, producing the heating amounts shown in cyan on the right. Notice the spike in the convective heating at the nose of the inversion, corresponding to a slight lowering in the base of the inversion and relatively dry layer in the soundings between 16 and 17 UTC. This is from compensating subsidence in the convective parameterization – downward advection strongly warms and dries the air immediately below the inversion. The other noteworthy feature is the convective cooling near the surface due to convective downdrafts, shown in the right panel (spike of negative values at the bottom). This does not show up in the sounding changes between 16 and 17 UTC on the left – rather, the sounding changes show warming from a little solar insolation making it through the clouds, a little mixing with the warmer air above, and a little longwave radiation warming from clouds in the warm layer above the surface. This is a typical example of how the effects of model convection on the sounding are masked by a variety of other processes at some levels but show up at other levels, and how detailed structure is not lost by a mass-flux convective parameterization such as in the RUC and GFS. In contrast, the reference profile approach used by the BMJ scheme in the Eta eliminates detailed structure and produces recognizable signatures in the model soundings. The orange curves on the right are heating rates from the grid-scale precipitation scheme. They are generally small at this location, with some erratic spikes of cooling in subsaturated layers where light precipitation was evaporating and one notable warming spot between 600 and 500 hPa at 17 UTC, which was associated with upward motion. There is no obvious or coherent connection between the grid-scale and convective activity at this location. |
|
53 Guidance |
Switching gears away from the convective parameterizations, we will now discuss products that can help assess the model solutions. Some of these are fields not available in AWIPS presently. Most (or similar parameters) could be generated from local models. All fields that we have seen so far, except the convective and grid-scale heating rates, are available in AWIPS. The BUFR sounding files, which you view with BUFKIT, are loaded with many other variables that BUFKIT doesn’t show but could be modified to show. Some of these are sounding variables, including for the Eta and RUC, the model cloud water (or ice) content in each layer – directly from the model’s microphysics - and the model cloud fraction in each layer – the same as used by the radiation parameterization in the model. Other variables are surface fields, including skin temperature, surface fluxes that show how much solar energy is reaching the ground in the model and how much of that is being used for evaporation and how much for surface heating. |
|
54 Guidance |
The link shown includes graphics of Eta model cloud bases and tops. Because there are no colors plotted where there are no clouds of the type indicated (deep convection, shallow convection, or grid-scale), it reveals where shallow convection is occurring in the model, and thus where soundings are being affected. Since shallow convection does not produce precipitation, your only other way of identifying its influence is by happening upon a sounding location in BUFKIT that reveals the telltale sign of moisture being mixed only around halfway up through the boundary layer. The plot on the left is an example of deep convective cloud tops, which are based on temperature equilibrium levels for source parcels from the most unstable model layer. This plot reveals that the convection over VA/WV/PA/NY has very low tops, and even for the southern squall line, tops are higher up than 300 hPa only along a sliver at the immediate front end of the line. The plot on the right is an example of grid-scale cloud tops. It shows a cirrus plume across the Gulf of Mexico, extending up the eastern seaboard, a mid-level deck behind it from LA to OH, and a deck of low clouds socking in most of the central US, while a mid- to upper-level band crosses Iowa. |
|
55 Guidance |
Same as previous plots except showing cloud bases instead of tops. This could be helpful for aviation forecasts, except the scale is pressure, not height above ground. For instance, note the higher bases in WV might not be higher or much higher above ground. |
|
56 Guidance |
More Eta plots of convective parameters are at the web site listed. Note that it includes helicity in the lowest 1 km in addition to the usual 3 km and that it includes three types of CAPE and CIN:
|
|
57 Guidance |
Similar fields from the RUC are shown at the FSL web site and most of these fields should be available in AWIPS on the 40 km grid #236. The parcel the RUC uses for its CAPE and CIN is the single most unstable model layer, except the bottom seven layers are mixed together to form one parcel representing the bottom roughly 50 hPa. The field identified as the convective cloud top is not from the convective parameterization. Rather, it is diagnosed from the same sounding analysis used for CAPE. Further, the convective cloud top is not the top of the positive area on a skew-T. It goes higher until the negative area above – with the overshoot, possibly above the tropopause – equals the positive area. Studies show this corresponds better with radar-measured cloud tops. Visible cloud tops are observed to extend even higher. |
|
58 Guidance |
An example of a useful RUC field not in AWIPS is the boundary layer depth,
measured as the height at which the virtual potential temperature exceeds
that at the surface. Surface-based convection would only be expected when
the LCL is within this layer, so a quick glance at this map can help show
where and when surface-based convection should shut down if the RUC temperature
forecast is good (or continue elevated but fade away as little more is
triggered). |
|
59 Guidance |
The SPC real-time mesoanalysis page is very useful for assessing the current conditions, with juxtaposition of fields and contour colors and intervals ideal for examining most convective environments. What is different than data available in AWIPS is that these SPC plots blend SPC’s surface objective analysis of current observations with a 1-hour RUC forecast. The list of available parameters is shown here. The emphasis is on diagnosing atmospheric processes affecting triggering, maintenance, mode, and intensity of convection. Note the absence of model precipitation forecasts |
|
60 Guidance |
Example from SPC page. Other fields for the same time are shown in the next slides. |
|
61 Guidance |
Example from SPC page showing the amount of detail and structure in the RUC moisture fields, which can identify corridors where convection can be fed deep moisture and other areas where the boundary layer could dry out if surface temperatures heat up enough to mix to 850 hPa. |
|
62 Guidance |
Example from SPC page showing where the boundary layer, not just the surface condition, has high CAPE and small CIN. However, in this case it did not lead to convection because there was no forcing, the cap was strengthening instead of weakening, and the air between the LCL and LFC was dry. |
|
63 Guidance |
Continuing the example from the SPC page, we now see that the area of high CAPE and low CIN over north-central Oklahoma seen in the previous slide has a low RH between the LCL and the LFC. This means building towers are likely to entrain dry air and evaporate. |
|
64 Guidance |
Example from SPC page showing potential for intense convective downdrafts if storms were to form. |
|
65 Guidance |
Another source for information about convection characteristics are internal parameters within a convective parameterization. The example here is from the KF parameterization. It shows the upward mass flux as a percentage of the mass in the unstable source layer in the grid box, so 100 means the model is overturning all the air it possibly can – extraordinarily vigorous convection. It seems to correlate well with updraft strength and severe weather potential but not necessarily with rainfall. It may also correlate with intensity of lightning activity, since updraft velocity correlates with lightning frequency. Similar information can be extracted from any mass flux convective parameterization, including those in the GFS and the RUC, though those models have never been programmed to output this internal information. This NSSL version of the Eta stopped running. However, SPC wants NCEP to provide this field for KF members of the short-range ensembles (SREF), so output of the convective mass flux might resume if NCEP programs its version to output this information. If you express interest in using it, that might help move things along, especially if your NWS Region can list it as a requirement |
|
66 Guidance |
Here is another example from the NSSL version of the Eta using KF. This shows the pressure of the source air that the convective parameterization used in triggering. It is showing surface-based convection off the southeast coast while the convection in the model moving across the northern plains is elevated, with roots at around 800 hPa. But the convection in Idaho might be surface based since it’s over elevated terrain (e.g., those pressures might be about the same as surface station pressures). |
|
67 Guidance |
The operational SREF unfortunately is not in AWIPS but you can find plots
at the link shown. |
|
68 Guidance |
List of fields available
on the NCEP SREF web site.
|
|
69 Guidance |
The orange contours are for the RSM members using SAS (like the GFS), and the red and green are Eta members using BMJ and KF convective parameterizations, respectively. Note how they clump into clusters by parameterization. The composition of the ensembles will be changing, with more variations of convective parameterizations but still 15 members in total, and the resolution will be improving to 32 km (48 km now). |
|
70 Guidance |
MOS thunderstorm probabilities were derived on a 48 km grid—not related to any model grid—based on using the lightning detection network for verifying the occurrence of a strike within each grid box. The same grid was used for severe reports (the usual—50 knots, 3/4–inch hail, and tornado) to verify the occurrence of a severe event. The process is the same as for developing other MOS equations, except these apply for grid boxes instead of at station locations. This means it has the same caveats—rare events will be poorly forecast because of small sample size in the development dataset, forecast behavior may vary by regime, and forecasts are increasingly weighted toward climatology at longer time ranges. The products include probabilities of occurrence for 6-, 12-, and 24- hour time periods for the Eta to 60 hours and GFS to 72 hours, and for 12- and 24- hour probabilities for the GFS out to 192 hours. The thunderstorm probability is the probability of a lightning flash anywhere within the 48-km grid box. The severe storm probability has two types: conditional and unconditional. The conditional probability is the probability that if a thunderstorm verifies, a severe weather report will occur within the 48-km grid box. The unconditional probability is just the probability of the severe event occurring, and it is the same as the thunderstorm probability times the conditional severe storm probability. Shown here is a GFS forecast of thunderstorm probabilities (left) and unconditional severe storm probabilities (right) for a 6-hour period ending 18 hours from the start of the forecast. |
|
71 Guidance |
This is the same except that it is based on the Eta instead of GFS, and the severe storm probabilities are conditional. When the severe storm probabilities are multiplied by the thunderstorm probabilities, the values are all very low, with hardly any spots exceeding 5%. For instance, it is showing that the probability of any thunderstorms that form becoming severe is higher than 20% north of Dallas but the probability of even having a thunderstorm is very low there. Comparing against the previous slide, we see that using grids from the higher-resolution Eta compared to GFS results in more structure in the thunderstorm probabilities. |