Did Katia Go Off?

A few days ago we discussed the anticipated extratropical transition of Hurricane Katia and how the GFS was predicting the central pressure to drop to a remarkable 928 mb over the North Atlantic.

0600 UTC 6 Sep initialized GFS forecast valid 1800 UTC 11 Sep.

We also noted that the GFS was a bit of an outlier amongst all the members of the Global Ensemble Forecast System (GFS), all of which predicted a major cyclone over the North Atlantic, but of varying intensities.

Selected isobars from members of the GEFS ensemble with the GFS wind
speed color filed. 

Well, what happened?  As shown in surface analyses from the Ocean Prediction Center, Katia indeed underwent extratropical transition as it recurved and began to move eastward across the North Atlantic.

Source: NOAA/NWS/OPC

Although the resulting extratropical cyclone broadened in scale, it did not deepen as it continued to move eastward and actually filled somewhat to ~967 mb this morning.  

Source: NOAA/NWS/OPC

We don't have the analysis yet for 1800 UTC, but based on the analysis above for 1200 UTC, the GFS will probably by about 40 mb off on this one!

One could beat up the GFS over such a large error, but let's give this some thought.  It did correctly forecast that Katia would undergo extratropical transition, eventually leading to a powerful cyclone in the North Atlantic.  The timing and position were also quite good, especially when you consider that this was a forecast with a 5.5 day lead time.

Instead, the issue here may not be a problem with the GFS, but the nonlinear, chaotic nature of the atmosphere.  The members of the GEFS ensemble, which start with slightly different atmospheric analyses, showed that there was great uncertainty with regards to the depth of the low center.  In other words, the 5.5 day forecast of cyclone central pressure was very sensitive to the analysis used to initialize the model forecast.

Here's another way to think about it.  Take out a dollar bill and drop it from shoulder height.  Mark where it lands, then do it several times over from the same location.  Even though you release the bill from nearly the same location every time, the bill never lands in the same spot on the floor.  After you've done this many times over, you can probably identify the spot where the bill is most likely to land, but also that there is a fairly wide envelope of possibilities for any given drop.  Why?  The bill's flight is controlled by it's release location and orientation, which are never exactly the same, as well as any motions in the air, which are chaotic and poorly observed (maybe if you add a smoke tracer, you can improve the prediction).

Now, if you want to factor forecast length into this experiment, simply get out a ladder and do it from higher.  The envelope of possibilities will widen.

So, my hypothesis for Katia's extratropical transition is that the overprediction of central pressure by the GFS reflects the lack of predictability in this event instead of an underlying problem with the modeling system.  The GEFS ensemble, which is based on the GFS model, illustrates that there were a wide range of possibilities for the cyclone central pressure given analysis uncertainty and the nonlinear nature of the cyclone evolution.

All of this illustrates the value of looking at an ensemble of model forecasts and not relying on a single model prediction when forecasting the weather.