I like to think of my approach as an ‘average of model ensembles’ approach. yuk yuk ]]>

Re: steven mosher (#316),

re 314. Kenneth, I’m waiting to see how well I guessed GISS land surface temp for 2008 before I claim the climate nostradamus crown

Steven, I was going to reply to you, but I thought it would not be of interest to anyone besides you and you, well you, already know what I was going to write. It was going to be a pretty snappy reply wasn’t it Steven.

]]>I played more with the parameters of the Weibull distribution and specifically looking at the MWS incremental failure rates using, not the actual values as I did above, but this time those derived from the Weibull distribution with the fitted parameters.

To make a long story short it was obvious to me that some of the frequency gyrations in the actual results, and particularly at the high MWS end of the distribution, were making the failure rates more difficult to determine. What the fit to a Weibull distribution does is, in effect, to smooth these gyrations in the actual data so that one can calculate incremental MWS failure rates.

What I found was that for all basins, using either Best Track or satellite PCA data, that the failure rate increases as one proceeds from the lower MWS to the higher MWS. The general difference between the Best Track and satellite PCA failure rates is that the Best Track, having the lower value shape parameter and closer to 1 (at 1 the failure rates become constant ), the failure rates increase significantly less on going to the higher MWSs than is the case for the satellite PCA data. A larger value scale factor, on the other hand, spreads out the increase in failure rates with MWS, i.e., needs a higher MWS to reach a final failure rate.

I do not know how clear this explanation is to those reading here, but if it is I would be interested in hearing whether these failure rates (as I defined them previously) have any physical meaning vis a vis the MWS frequencies found in TCs.

]]>Since a Weibull distribution is frequently used with failure times, I quickly looked at the TC duration time relationship with MWS and rejected this approach (see below for evidence).

I then attempted to find a simplistic explanation for obtaining failure rates as an increment of wind speed (or MWS if the TC fails to progress to a higher wind speed) in place of the time increments that are ordinarily used in reliability engineering treatments.

Below I show the relationship of TC duration (to the nearest day) to MWS. While the R^2 shows an overall good correlation, it can be seen that at the lower MWSs the relationship is more linear and when the duration times exceed 12 kts, the relationship breaks down. I, therefore, concluded that the Weibull fit has little bearing on time or failure times.

In looking at the failure rates by MWS increments, I start by noting that all TCs counted for MWS have to go through progressive increments of wind speeds with the terminal speed being the MWS. I ignore for these purposes how quickly or slowly the wind speed incrementally increases or decreases, or, for that matter, whether the wind speed decreases and then increases again to a higher wind speed. In other words, a TC after attaining an incremental wind speed, for example, one between 60 and 70 kts, can survive to the next higher increments of wind speed or diminish in wind speed and eventually die. That death of the TC after attaining that wind speed then fixes the MWS and counts as failure for that increment of MWS and the failure rate is determined using the incremental MWS counts divided by the total TC MWS counts (or total TC counts).

That actual failure rates for the MWS increments for Best Track data shows a progressive increasing rate for a short range at the lowest MWSs up to a point where the failure rate becomes nearly constant and then increases at the highest MWSs. The increase in failure rate at higher MWSs would appear to me to be logically the result of the TC (hurricane) approaching a physically limiting bounds for MWS as noted in papers by Emanuel. Evidently, the lower failure rates at the lowest MWS indicates that it is relatively easier for those TCs to progress to higher MWS than to die than it is for TCs at the higher MWS.

For the Elsner satellite PCA data, the failure rates at the incremental MWS appear to mainly to increase with increasing MWS. The data for both the satellite PCA and Best Track has few TCs counts at the highest MWS and thus obtaining a failure rate in that range is uncertain.

It is my conjecture that, since the satellite PCA does not pickup the MWS measured for the Best Track at the lower MWS, the Elsner PCA is, as was suggested here by Ryan M, measuring something physically different than MWS. That something different can be well fitted to a Weibull distribution, but then again with 3 free parameters maybe that is not all that surprising.

The shape, scale and location parameters for the fitted Weibull distributions for the six TC basins worldwide are listed below when using Best Track and satellite PCA data for fitting.

Best Track:

WP: Shape = 1.3; Scale = 60; Location = -30

NATL: Shape = 1.4; Scale = 43; Location = -30

EP: Shape = 1.3; Scale = 45; Location = -30

SI: Shape = 1.1; Scale = 43; Location = -30

OI: Shape = 1.1; Scale = 26; Location = -30

SP: Shape = 1.25; Scale = 40; Location = -30

Satellite PCA (Elsner):

WP: Shape = 1.7; Scale = 52; Location = -40

NATL: Shape = 2.45; Scale = 40; Location = -30

EP: Shape = 2.9; Scale = 44; Location = -30

SI: Shape = 2.9; Scale = 47; Location = -30

OI: Shape = 2.1; Scale = 36; Location = -30

SP: Shape = 2.9; Scale = 57; Location = -30

How did the rest of the bloggers do on their 2008 predictions for the number of hurricanes and the number of named storms. I predicted 6 hurricanes and 15 named storms but was low on the ACE number[75-80]

]]>Re: steven mosher (#312),

Steven Mosher, do not be bashful about your good fortunes in forecasting. I was unaware of your good luck in forecasting in other areas. Please list all your lucky guesses here as I want the world to know that I have been competing, by diligently putting all my complex models together, with someone with a golden touch.

]]>Don’t worry about being on baby ice. It held its own in the summer death struggles, and now its kicking backside and taking names. That’s a good partner to have in the betting trenches.

]]>Steven Mosher handicaps Gray, while Kenneth Fritsch devises and uses his own forecasting model. You do the math.

Kenneth did not enter the Wizard contest as no self respecting forecaster is going to enter a contest with wizard in the title.

]]>