Sunday, October 9, 2016

Hurricane Matthew

Been testing out some concepts on hurricane Matthew over the last few weeks.

The hurricane cone overlaid with GEFS forecasts provides a dual ensemble approach. The GEFS for NWP and the accumulation of cones for the human and/or time lagged ensemble. Integrating the components of NHCs' forecasts along with an ensemble can add insight, and linking older forecasts with newer forecasts gives an unspoken sense of uncertainty. Seeing the forecasts in their entirety allows a glimpse into the minds of forecasters and what they wrestle with ... which is exactly what the recipients of forecasts wrestle with: Where will this Hurricane go? Will it be close to me? How certain are you?

The heatmap was added not soon after to show the GEFS max wind speed over time. Clark Evans was nice enough to point out the GEFS effective resolution was probably not fine enough to get to cat 4-5 intensity. So this is a different way of getting at the intensity (through dot size on the map) in aggregate. The time trend is also interesting but with a small inset it has to be dramatic to be noticeable. What I did notice was the discontinuity at times between the initialized intensity and the first 6 hour bin. Again effective resolution at work,  playing a role in model start up. Model land isn't identical as the real world, even when you use data assimilation. But its worth trying and doing, because it can make a difference.

I also played around with wind radii forecasts but due to their issuance every 6 hours makes for an unpleasant and odd plot.  So I switched to just the best track wind analysis, advisory positions, and the latest cone. You can get a feel for how big the wind field is relative to the early cone. For this storm the TS wind radii extend beyond the cone maybe for Day 2/3 like errors used to build the cone. Here is an example:


Lastly I will turn toward verification of Matthew cones and GEFS.  Down there is an animated gif of the sequence of 5 day cones and t+5d advisory positions. We talk about skill and cones in the abstract. NHC has a great 5 year running mean error, which at times draws some ire because "it doesn't use an ensemble to adapt to THIS storm". Much as in severe warnings or Convective outlooks, draw big or go home. Its easy to say you can narrow it down. But try it. Draw small and suffer at the hands of the unknown, or the random. The weather events that we forecast are not smooth. They may not all be predictable. We think we have answers down on time scale of minutes or hours (choose your phenomena) but that is a tall order. Just cause you know "its going to turn" or loop or whatever doesn't mean you know when or where. That is the uncertainty principle we should acknowledge here.  Ancient verification wisdom: "Cones should be made as small as possible, but not smaller."


Edit:
Below is the same image but with the GEFS tracks and NHC center line of the cone (purple). There is a lot here but I dont think I can summarize it right now. Enjoy: