Last month I discussed Joe Romm’s nearly hysterical reaction to the fact that NOAA thought the Southern Oscillation was about to oscillate back to El Niño (as it often does after La Niña.) This month, NOAA, whose staff write with somewhat less emotion than Romm updated their forecast:

Model forecasts of SST anomalies for the Niño-3.4 region (Fig. 5) reflect a growing consensus for the continued development of El Niño (+0.5°C or greater in the Niño-3.4 region). However, the spread of the models indicates disagreement over the eventual strength of El Niño (+0.5°C to +2.0°C). Current conditions and recent trends favor the continued development of a weak-to-moderate strength El Niño into the Northern Hemisphere Fall 2009, with further strengthening possible thereafter.

The June and May forecast from various models are illustrated below:

Click for larger

So, it looks like the Southern Oscillation will continue to oscillate as it always has. La Niña is followed by La Nada, and then El Niño, then La Nada then La Niña all in a somewhat irregular pattern, with occasional skips.

What does this mean for testing of AR(4) models.

As readers know, each month, I show the current trend in surface temperature since 2001 along with uncertainty intervals corrected for red noise, (i.e. AR(1) ) and compare those to projections from the IPCC AR4. The comparisons have been indicating that the data we actually gathered since 2001 was inconsistent with the mean model trend for the AR4 being correct at a confidence of 95%. So, if beginning the test in 2001 is appropriate (or more specifically, if the choice was truly random) these results suggest either a) the model-mean trend for surface temperatures from the AR4 are probably too high, b) the statistical tests rejects at to much frequently during periods without volcanic eruptions (something I doubt for reasons I have entirely discussed or c) the weather event as an outlier. That is: we know that things that happen less than 5% of the time do happen. In these cases, statistical test will correctly recognize that the event happens less than 5% of the time, and we will reject at p=95%. However, it will turn out the rejection was incorrect. (This happens with all statistical tests under frequentist rules. It is not an analytical mistake and has nothing to do with using short time periods.)

The upcoming, regularly scheduled El Niño is important because it will provide more data to indicate which of the above possibility is true.

The question one might ask is: “How will we know whether (a), (b), or (c) is correct?” Well, in some sense we can never be 100% sure. Just as we reject or fail to reject a model mean trend at some confidence level, we would also diagnose the above at some confidence level. But here is an incomplete list of simple things I am will be looking at:

1. Will the observed trends begin to fail to reject? I anticipate this will happen as El Niño progresses. Oddly, I expect it to happen whether or not the projections in the AR4 are too low. The reason I think it will happen is that I sincerely doubt the projections are off by more than a factor of 2. When an incorrect hypothesis is tested constantly as data trickle in, it is typical to see the “rejections” go “fail to reject….. ….. ….”, “reject.”, “fail to rej..”, “reject..reject “,”fal”, “reject, reject, reject…” until finally we see consistent rejections at some level of “p”. The precise amount of time it takes to get to “reject, reject, reject…” is unpredictable, but, unless models are way, way, way off, it’s possible to show we are not yet at the point where we should expect to be getting a persistent run of uninterrupted ‘rejects’ at p=95%.

   (For those wondering about short time effect, if the hypothesis is absolutely, positively bang on correct and we test at p=5%, we will in 5% of cases, and the probability of rejections does not change over time. So, the ‘short time’ aspect of the test doesn’t mean we get to downpay rejections. If a test is properly constituted, these should happen equally rarely for tests over short periods and long periods.)

2. Will observed trends over take and exceed IPCC trends and then reject on the high side using the current method? If they do, that’s a darn good indication that the statistical method selected does reject too often. (If that occurs, expect some chortling with glee by “others”.) But, if that sort of rejection does occur, at that point, I’ll of course have to admit it happened, and we’ll have enough data to better characterize the spectral properties of “weather noise” when there are no volcanic eruptions occurring. ( Even if no high-side rejection occur, we’ll be monitoring for information to suggest the method of computing uncertainty intervals during periods without volcanic eruptions should be modified. )
3. If the IPCC trends do not end up rejecting on the high side, we’ll be monitoring to see the observed trends drop back down and re-reject during or even before the next La Niña.

Will I be adding other tests?

Yep! As El Nino deepens, I’m going to be watching the uncertainty intervals so I can report when the test does show statistically significant warming. The “rejections” for testing the “no warming” hypothesis are going to oscillate back and forth from “fail to reject” to “to reject” in exactly the same in exactly the same way those for testing the IPCC AR4 oscillate– and for the same reason. So, even though they oscillate, people like to know the current status using any particular test, and I plan to mention when we break that threshold.

I anticipate we will certainly see a “reject no warming” as we approach the top of El Niño.

I’m now waiting for RSS…. I posted RSS on the 4th last month. Technically, we don’t have AR4 projections for the troposphere; I’ll show uncertainty intervals, and test both “no warming” and “about 2C/century” when it comes in.

If any of you see RSS report before I do, let me know! (Just to let you know: I expect they will be up from May.)

References: NOAA’s most recent ENSO updates are available at ENSO Discussion