NSSL WRF performance

The more I looked at the NSSL WRF forecast from yesterday the more interesting it was.

While the model initiated storms off the warm advection cloud band in Iowa after 0100 UTC, which did turn out to be correct, it was not severe nor was it the major player. Observations indicate this was part of an alto-cumulus castellanus area.

The area in NE where storms initiated in the real world along the warm front sharpened until 23-00 UTC where in model land it weakened considerably. There was an indication in the model that reflectivity was small but non-zero when the convergence was strong. This is a good albeit weak signal.

Another area under consideration for convection initiation was along the dryline in KS and northern Oklahoma where one storm formed around 23 UTC. The model had little in the way in convection here until 2 hours later, along the dryline.

Another area was in SW OK, where observations indicated a small, weak storm developed around 01 UTC and quickly died off. The closest model storm was at 05 UTC. Of course, the whole forecast goes awry in these latter two isolated storms as the model initiates convection all along the dryline from KS through Southern Oklahoma. The last 6 hours of the forecast looks little like what happened in terms of storms. I would caution that the model is not entirely wrong, just very aggressive. The cloud fields in the model develop into convection but closely resemble the cloud fields observed.

We are just beginning to harvest the wealth of information contained within such forecasts. I believe we will learn a lot more about these types of forecasts when we get down to looking at cloud fields (not at the grid scale but over substantive areas) and use these to compare the model with observations. This perspective should give forecasters more confidence in the overall appearance, and solidify what to look for when examining fine resolution forecasts. I think information extraction will be much more successful than reflectivity alone.

http://www.nssl.noaa.gov/wrf/110409/

Chasing supercells

Pretty good chase day today. Any chase day is a great day, but this was good because it ended my absence on the Plains after roughly 4 years. It culminated in a splitting supercell with near tennis ball size hail in a marginal low level shear environment, but favorable bulk shear environment.

Hard to say what the issue with convection initiation was, but most of the updrafts that initiated at least marginal vertical cumulus growth seemed to tilt and then result in small turrets on the top. The clusters of clouds that did manage to grow finally did so after 2200 UTC, and by 2300 UTC we had a nice storm to chase at least from its radar presentation. The base was elevated a bit but a clear updraft bell had formed, and when we drove north on I35 to stay ahead of it, noticed pretty good size hail on the ground. Maybe 2-3 inches in diameter, low-density. I even found some that had split in half so you could see the rings. Most of the hail in the grass was unbroken, but the stuff that had hit the pavement was easily smashed.

The model scenario played out fairly well. A moisture pool was left over from the late afternoon mixing of the dryline further south. Last nights NSSL WRF simulation had 1 storm form west of Tulsa and I took this to be a positive sign that CI was probable. Of course the model had some pretty high moisture values so that was a concern. However, afternoon observations indicated that dew points would hold near 64 F. Of course the low level shear was weak as was the SRH, but the bulk shear was up around 20 m/s. Plenty good for supercells, but not so good for tornados.

The focus for initiation was along a convergence zone in Northern Oklahoma that would eventually redevelop rapidly north, at least in model land. I will have to investigate if this actually occurred.  I think its worth investigating from the modeling standpoint, especially the structure of the boundary layer that led to the mixing out event further south but kept enough moisture further north.

The storm we were on is to the east-northeast  of WDG on the image below at the time tennis ball hail was falling while we were on I-35.

Glad to be back chasing storms...

Uncertainty cont.

I have been able to examine the NSSL-WRF, 00 UTC and 12 UTC NAM this morning.
With regard to the trough that is forecast to be the major player, it is likely that significant severe will break out colocated with the strong forcing along the triple point, in the dry punch into MO, and then another round even later in AR.

The main uncertainty lies in OK, where my hopecast suggests storms could try to break out along the dryline. The problem is that it occurs just around 22-23 UTC when the wind profile might be considered terrible for tornadoes. The wind profiles in general become more favorable further east and later on putting the tornado threat into AR but the window for discrete supercells appears to be small. Rather a squall line of some type will form with probably the chance for embedded supercell structures.

Further south however, there are better wind profiles, but the cap is somewhat stronger. The 00 UTC NSSL-WRF forms a squall line there as indicated by the synthetic satellite imagery.

I don't have a good intuitive feel for what may occur given some of the wind profiles I have seen. I do think the overnight models will struggle as they are typically too far east with any convection. They also struggle to produce individual storms ... will only produce storms in stronger forcing. The resolution of the models, the tendency to produce weaker lapse rates, etc all contribute to the storm bias. That said, the main threat appears to Normans east (Tulsa area and north), northwest (along the triple point), and southeast (secondary dry punch). There is still a chance for central OK. It all depends on if any storms attempt to go up along the dryline, which ultimately depends on the relative balance between the depth of moisture in the warm sector and the cap strength.

I am hoping that LMN and OUN will launch 21 UTC soundings, and maybe 18 UTC soundings so we can really examine the wind profile evolution as well as the cap. This will be a good case for analysis either way as it is a strong forcing case that is highly dependent on mesoscale details that our models may not get correct.

Severe weather potential

Interesting forecast shaping up for tomorrow evening. Somewhat strong trough will move into OK tomorrow bringing with i the chance for significant severe weather to central and eastern OK and continuing eastward overnight.

The more certain portion of the forecast is the later period late Sunday night when it appears probable that a severe MCS/squall line will tear up parts MO, AR, LA again. I think its likely as more time will allow for more moisture to advect northward in what amounts to me to be an odd linkage between a pre-existing low level jet well east of the effects of the trough and the trough. The apparent phasing of the two will occur later sunday evening. These highly dynamic environments lead to some interesting MCSs.

The early part of the threat in central OK is less certain. To me, at the moment, the most interesting part of the forecast is that the features of interest appear to slow down before entering it the Plains, then speed up once they do. This makes for a confusing scenario. The prominent features I see this morning are the dryline in western OK, the dryline warm front intersection in NW OK, and the rather strong 700 hPa cold advection over the dryline all at 00 UTC. Now, I have no idea what to expect in this type of slow then fast large scale regime. This fact alone adds uncertainty for my limited experience of significant severe weather in OK during Feb.

What is not clear is if the cold advection aloft will be enough to remove the cap, and if the instability will be large enough to be realized. We just had some rain followed by cold temperatures, and soil water fractions are high apparently. The last 10 and 30 day rainfall maps show that western OK is dry and eastern OK has seen around 2+ inches. So it is possible that strong sensible heating will take place on Sunday in areas that are drier ahead of the dryline. Will this actually matter? I will tell you on Monday.

The other major issue is how the dryline and shear align. At the moment the mean wind for sunday evening appears to be from the SW and the dryline is oriented more north-south. This angle difference will allow storms to move off the dryline. The shear vector should also orient itself across the dryline but the magnitude of that angle will be very important to storm mode.

I guess my inclination at this moment are that all the necessary ingredients can be found, it will be a matter of how they come together and when that will occur and for how long. For now, at least, it is important to get that moisture screaming northward. I could use a good storm chase.

A lot will change. But I think there is a chance for supercells. I am uncertain where convective initiation will occur, and where severe storms will get going, and what that initial mode will be. I will evaluate the Short range ensemble Forecast (SREF) when the 15z run becomes available later this afternoon.  Really this is just me getting my thoughts ramped up ... otherwise known as situational awareness.

More storm discussion

The plot thickens. Not only are subsequent model forecasts increasing the snow for central OK, but the global model has backed way off. The SREF was going crazy with anywhere from 1inch to 22 inches of snow for Norman as of 2100 UTC. The operational NAM was going for 24 inches, and the  GFS was going for 1 inch. Talk about uncertainty. The funny thing is that the NAM gets its boundary conditions from the previous GFS model run.

I speculate that the width and depth of the cold air helped get the GFS a good initialization and as a result a good depiction for the NAM. But maybe that cold air depiction was projected onto too large scale such that the ingredients for this event were misaligned. Also, the NAM ingredients may be too well aligned such that convective feedback may have been playing a role in creating a lot of precipitation but was able to simulate the banded features well.

As of 9:25pm there is the lovely, heart pumping sound of thunder, big lightning streaking across the sky, and ice pellets! Some of which could count as hail. I had to take a break so I could enter some observations for the W-PING project at NSSL. On a related note station KPVJ in OK reported 0.25" of liquid equivalent in an hour (6.3 mm/hr). Impressive rate. If only we had data for the precipitation rates at all mesonet stations!

It looks like the main issues of the day were (in no particular order):
1. where would cold air be, its depth, and the depth and magnitude of the warm air layer aloft,
2. where would thunderstorms initiate and then travel over,
3. as a result of the position of the cold air, where would the surface low be,
4. and related to 3 where would the maximum divergence aloft be (horizontally coupled jets).

I looked at the SREF placement of the Upper Level Jet (ULJ) and looked at the diagnostics from SUNYA* (irrotational and non-divergent wind at 300 hPa; the 850 q vectors with the along and cross front components of the vertical motion). This event looks very similar to east coast snow storms with coupled jets. In this case the irrotational wind center was intense over OK nearly tripling in value over a 6 hour period (18 to 24 hr forecast) from this mornings NAM. The along front forcing for ascent was forecast to come through in 2 waves ... the leading one much weaker than the 2nd. Aloft the forcing for ascent is forecast to be more widespread and last longer into tomorrow late morning.

There were hints that a coupled jet structure was developing at 0000 UTC this evening. But closer inspection of the soundings suggested that a the true jet was near 200 hPa as evidenced by the 150 knot winds near 210 hPa at LZK and the 120 kts at SGF and 135 at LMN and OUN at about 145 kts. Only 0600 UTC balloons at TOP and LMN could help identify if a coupled jet feature did develop.


*Thanks to Kevin Tyle for trying to keep SUNYPAK alive!

Scales of uncertainty

The latest fuss in meteorological circles has focused rather intently on the Storm of Christmas, then the day after, and now the day after the day after Christmas storm. The storm has been "delayed" by 2 days over the course of the last week, and has shifted position numerous times in the forecast (gonna miss far to the east, bounce back to the west for a hit). We call this behavior uncertainty.

So what is uncertainty? "not confirmed", "still undecided", a "lack in confidence"

In modeling terms, the forecast solution has not yet converged. Our current capabilities, however, are quite robust at short range, meaning we expect the solution (over many initialization cycles) to converge. This time period is usually between 48 and 72 hours depending on the modeling system, the scale of the features that need to be resolved well, and the overall scale of the dynamical system which we seek to predict.

Over the years, despite many advances in satellite data, data assimilation systems, and numerical models and their increasingly sophisticated methods ... we still need to wait for the disturbances to enter the well instrumented radiosonde network*. Alas this too can be deceiving as was the case for the Surprise snowstorm of January 2000. You see a classic nor'easter can have origins from the Pacific Northwest which travels in amplified flow towards the gulf coast. The system itself may never leave the well instrumented but the warm gulf, cool temperatures aloft, and strong wind shear may foster convection ... thunderstorms ... or organized noise. This noise can then feedback from the large scale to the small scale ... outside the network and remain outside the network as the coastal low forms and deepens as it travels up the coast.

So, you can see how a "converged" solution of where the storm will be, can help us have confidence on predicting the other aspects of the storms. But uncertainty remains in where the precipitation will be, what precipitation type will fall, and for how long.

This uncertainty about snowfall placement and amount is very similar to the summer forecasting of thunderstorms. We may have a converged solution of the larger scale details but the smaller scale details can have a large effect on where, when, which storms will form and how severe they might be.

This where the art of forecasting kicks in. Where the analog experience of forecasters contributes. Knowing how they were fooled last time, or how they picked up on certain observed details which caused them to do better than the models. The forecaster has the ability to understand these scales of uncertainty. Only in the last few years have methods been developed, like the Ensemble Kalman Filter, which can show us where the uncertainty is for a specified region at 3,5,7 day lead time. In fact the Winter Weather Reconaissance program is designed to use these methods, then collect data in the uncertain regions to see how that can change the forecast and its uncertainty!

Pondering perspective in ensemble modeling

When forecasting anything, one must always consider the perspective one has. This is not easily achieved since our point of view is necessarily biased, either by a previous forecast, previous experience, analogs, or instinct.

Perspective "is the choice of a context or a reference (or the result of this choice) from which to sense, categorize, measure or codify experience, cohesively forming a coherent belief, typically for comparing with another." - From wikipedia.

Note the implied bias: "typically for comparing with another".

This is why ensembles are so neat in modeling the weather. The whole point of an ensemble is provide perspective or perhaps more appropriately predictability and by extension certainty (or uncertainty). This is particularly true even if the range of solutions does not cover the phase space of what is possible. In most instances, the mean of the ensemble is better than any individual member.

In the case where the outlier has the most value (no matter how wrong it is), the forecasters perspective may be the only real clue that it is even remotely likely for it to be correct. That is the value of the human forecaster and their experience is most likely to recognize the value of an outlier. There is significant risk associated with favoring an outlier. For one, you are going against what all the other members of the ensemble are trying to convey. So, you have to have good reasoning and great perspective on why so many members could be wrong.

In my opinion, this is why so many BIG forecast failures have occurred. It is difficult to trust an outlier, in a timely manner, because it takes a long time to discount a lot of members AND analyze the outlier in question in great detail such that you trust the solution. This is a major issue in severe storms research since the forecast period is short, the lifetime of some storms and their hazards are even shorter, and the models we use are shrouded in uncertainty (initial data, model spin up time, resolution, physics, dynamics).

Let us not forget that even ensembles have difficulty in predicting the certainty. Just because all the members are similar does not mean the forecast is certain. The issues we face now are just as much technical as they are scientific. Navigating the world of coarse ensembles and fine resolution ensembles will be fascinating.