There actually are something resembling El Ninos in the Atlantic ( see here ), driven by wind anomalies, but my understanding is that the Atlantic wind anomalies are small in magnitude and duration, and the amount of ocean potential energy involved is quite small compared to the Pacific.

So, they happen, but they are babies (El Enano?) compared to the Pacific events.

]]>My question is more about the mechanisms for the generation of El Nino conditions off the west coast of SA. What is it about that mechanism that means that El Nino type conditions are not generated in say the Bight of Benin? Is it to do with the Andes, since there is clearly no equivalent in West Africa. Basically what causes El Ninos?

Thanks ]]>

On rare occasions a seedling will survive and form a storm.

]]>I would like to see more details on how Willis E. processed these data, but in my view it is line with changing count efficiencies and cyclical nature of hurricane occurrences.

I first detrended the data in the normal way, removing the least squares linear trend.

Then I used an iterative process (the “Solver” function in Excel) to determine the phase, amplitude, and frequency of the sinusoidal wave that removed the most variation from the hurricane data (minimizing the RMS error between the sine wave and the data).

All the best,

w.

]]>The Cusum plots, while showing a change in mean hurricane counts are not, in my mind, a particularly good analytical tool knowing what we do about the occurence and measurement of hurricanes. We suspect a trend due to counting efficiency (which could be confounded with a natural increase ‘€” but that is contradicted by the steady count of landfall hurricanes) and there is evidence of a reoccurring increase in hurricanes on a multi-decadal time scale.

From Paul Linsay’s Poisson Fit thread, Willis E. detrended and removed a sine cycle from the 1851 to 2006 hurricane data and compared it to a Poisson distribution here:

http://www.climateaudit.org/?p=1022#comments

The Chi square goodness of fit test to a Poisson distribution for the untreated data yielded a p = 0.09, while the data worked by Willis E. to detrend and with cycle removal gave a Chi square goodness of fit to a Poisson distribution of p = 0.87. I would like to see more details on how Willis E. processed these data, but in my view it is line with changing count efficiencies and cyclical nature of hurricane occurrences.

]]>I was also thinking if there is an equivalent of El Nino in the South Atlantic? If not, why so much asymmetry?

Are differences in shorelines and currents sufficient? If an equivalent does exist the

interaction of the two should be interesting. ]]>

I really would appreciate some feedback from someone with a stronger background in statistics than myself. I think I have done the control limits (Decision Intervals) correctly, but I really don’t know. As you can see, I have posted the plot of hurricane numbers from 1944 to 2006. While it looks like a shift in lambda may have occurred in 1995, you really can’t reject the null hypothesis of a constant lambda until 2005.

Here is the log from anygeth that I used to calculate the Decision Intervals for the 1944 to 2006 plot. It may mean more to you than it does to me. I think that an in control ARL of 474 means you can expect one out-of-control point every 474 data points and the OOC ARL of 13.2 means that a run of 14 points above the mean would be another reason to reject the null hypothesis. The Winsorizing constants used mean no Winsorizing was done, whatever that is.:

Program to calculate cusum decision intervals

Copyright 1997, D M Hawkins and D H Olwell

Which distribution do you want?

Poisson

Enter the in-control and out-of-control means

6.000 8.000

The exact theoretical reference value is 6.952

Enter the reference value you want to use :

7.000

What are the Winsorizing constants?

-999.00 999.00

Want zero-start (say Z), steady state (say S) or FIR (say F)?

z

Enter ARL

400.0000

h 10.0000 arls 132.3 119.6 128.2

h 12.0000 arls 252.6 232.8 247.4

h 13.0000 arls 346.5 326.7 340.9

k 7.0000 h 13.0000 ARL 346.51

h 14.0000 ARL 473.73

DI 14.000, IC ARL 473.7, OOC ARL 13.2, FIR ARL 8.2, SS ARL 8.5

Which distribution do you want?

Poisson

Enter the in-control and out-of-control means

6.000 4.000

The exact theoretical reference value is 4.933

Enter the reference value you want to use :

5.000

What are the Winsorizing constants?

-999.00 999.00

Want zero-start (say Z), steady state (say S) or FIR (say F)?

z

Enter ARL

400.0000

h 10.0000 arls 243.1 222.1 237.1

h 12.0000 arls 531.0 497.1 524.2

h 11.0000 arls 360.1 339.1 353.4

k 5.0000 h 11.0000 ARL 360.06

h 12.0000 ARL 531.00

DI 12.000, IC ARL 531.0, OOC ARL 11.7, FIR ARL 6.8, SS ARL 6.8

]]>