Wednesday, September 30, 2015

Joaquin is very sensitive; or why meteorologists are freaking out about this storm

We have a hurricane in the Atlantic at the moment---Hurricane Joaquin, a category 1 storm.  It's currently milling around in the Bahamas, slowly strengthening as it churns over the warm Gulf Stream waters.

Obviously any hurricane near the US east coast would raise concerns, but in this case meteorologists don't know what to do with the forecast.  Why?  Because even with our many state-of-the-art weather models, it's still not clear where this storm is going to go.

Here are some forecast tracks for the center of the storm from the Weather Underground site.  They show many of our major models' predictions for where the storm center will go.
Each colored line represents a different model.  You can see that most of these models take the storm into the Carolinas by this weekend.  One model (the NAM) seems to go a bit off to the east.  This actually looks like fairly good consensus between the models---almost all are agreeing on where the storm will go.  But there's one model missing from this map---the ECMWF weather model, arguably the best global weather model we have today.  Below we have forecasts from the US GFS model on the left and the ECMWF model on the right, both valid at 12Z Sunday morning.

The GFS shows a strong hurricane impacting the Carolinas, like we see in the track map above.  But the ECMWF shows a much weaker storm that actually tracks out toward Bermuda...well away from the US.

And this is what's giving meteorologists the fits.

Just about everyone is agreeing on a land-falling storm this weekend, except for the best model we have.  We can break this uncertainty out even further by looking at ensemble forecasts---several different forecasts produced by the same model, but starting with slightly different initial conditions (after all, we're not 100% certain about the exact status of the atmosphere when we start the model).

Below is a plot (again from Weather Underground) showing the GFS ensemble forecasts from this morning.  These are 21 forecasts using the GFS model.  Most of them take the storm into the coast, though admittedly there are a few that keep it off shore.
The ECMWF also has an ensemble of 50 forecasts.  Here's an animation from Ryan Maue (@RyanMaue) showing the minimum surface pressure centers from all the ensemble members (i.e., where each ensemble member is putting the center of the storm over time) from this morning.  You can see that the different forecasts have the storm centers all over the place...some move it over land, but others keep it out to sea.

So there is a lot of uncertainty.  Adding to the uncertainty is how stubborn these model forecasts have been.  The single operational ECMWF forecast (not the ensemble)---again, our most accurate and trusted model---has consistently taken the storm out to sea over its last four model runs.  Even as the ECMWF ensemble has slowly moved the storm westward on average, that single deterministic ECMWF forecast has stubbornly kept it to the east and away from land.  However, the other models (like the GFS) have increasingly brought the storm further west and towards landfall in the same time period.

So what do you do when a home run would win the forecasting game, but your best slugger comes out and shows a bunt?  Trust their instincts or hope they come around and get in line?

One thing the meteorological community is trying to do to reduce the uncertainty is to get more observations.  As I said above, we're never 100% sure of what the atmosphere is doing at a given time, so our models always start out with a bit of uncertainty.  We can try to reduce that uncertainty by taking more observations.  And that's exactly what we're doing.  You may have seen the "Hurricane Hunter" airplanes which fly into tropical cyclones to get more detailed observations about what the storm is doing.  These planes also release "dropsondes"--think weather balloons without the balloons.  These packages of weather instruments fall through the atmosphere and collect valuable information about the structure of the atmosphere on the way down.  These observations are then "assimilated" into the next cycle of weather model forecasts to reduce the uncertainty.  Below, Alex Lamers (@AlexJLamers) shares the plan for the Hurricane Hunter flight this afternoon.  Each "dot" is a planned location where a dropsonde will be released.

You'll notice that many of the dropsondes aren't even around the storm...they're off to the west in the Gulf of Mexico.  Why are we getting information for areas not around the storm?

As I've blogged about before, hurricanes, despite how powerful they are, are still steered by the much larger ridges and troughs in the atmosphere.  Knowing how these troughs and ridges will evolve is key to figuring out where the hurricane is going to be steered.  One experimental tool we now have is something called "sensitivity analysis" or "ensemble sensitivity analysis".  We can basically use our ensemble model forecasts to look at different times of the forecast and see how changing the atmosphere in one region affects the forecast downstream in another region.  Here's an example ensemble sensitivity plot from Dr. Brian Colle at the University of Albany using last night's ECMWF ensemble.
Here they set the target to be the upper-level pattern over the eastern US 3 days from now.  The plots go back in time through the forecast from Friday evening in the upper left down through Thursday evening in the lower left and then over to this (Wednesday) evening in the middle right.  You may have to click on the image to make it bigger to see the details.  Notice that in the forecast for Friday evening (upper left), there's a lot of dark colors over the eastern US...those are highlighting regions where the pattern is extremely sensitive and uncertain.  We've seen this in the hurricane tracks above.

But let's trace those areas of uncertainty back through the forecast...where do they come from?  You'll note by the time we get back to this evening (+1 day, middle right), the sensitive areas are in these shortwave troughs over western Ontario and the Great Lakes area and into eastern Ontario and Quebec.  That means that it's actually uncertainty in the upper-air pattern over the Great Lakes and eastern Canada tonight that's turning into uncertainty in the hurricane track on Friday!  To improve our forecasts, we actually need more information about what exactly is going on over most of eastern North America...not just around the storm.

The National Weather Service may start doing special weather balloon launches over the next few days to try and reduce that uncertainty.  Each model run that comes out will be able to take advantage of any additional observations to start their forecasts off with greater accuracy.

I haven't even talked about the intensity of the storm, which also varies quite a bit between the models.  Some develop Joaquin into a major hurricane (Category 3) over the next 72 hours, others keep it as a weak hurricane (Cat 1).  Those extra dropsondes released by the Hurricane Hunters in the storm may help refine that aspect of the forecast.

For now, I'll leave you with the latest forecast from the National Hurricane Center (as of 22Z Wednesday) and encourage you to read their discussions over the next few days to see how our forecasts are progressing!