Thursday, November 29, 2012

How to say "It's going to be a hot one"

If there is anywhere where it would be hard to predict the weather, it would have to be here in Kansas. We are just too far from the sea for anything to have strong influence on our weather.

Yet, it seems like it must be a great temptation to try to predict the weather. Because many people have said to me in casual conversation some variation of, "It's going to be a hot summer." Whatever the weather is today (or the past few days) people seem to project forward a couple of months.

Is there any basis for this?

If it's hot this month, is it likely to be hot next month? Or dry?

Here's how you test that.

First you generate climate anomalies. In a seasonal climate like ours, it's simple to know what the climate will be like in a month or 6 months or a year.

If it's June, in the grand scheme of things, it's likely to be hot next month cold in 6 months and hot again in 12 months.

But, what we're interested here is in the climate anomaly and how they are correlated over time.

So we aren't saying it's going to be hot next month, but hotter than average next month.

To evaluate this, you look at past records of data and run autocorrelation analyses. This analysis looks at how different temperature or precipitation over a certain length of time and then examines the correlation between the current period's anomaly (hotter/colder, wetter/drier) vs. next month's.

Here are some data from May weather over the past 50 years for Manhattan KS.

I went to and accessed their climate explorer. The site runs autocorrelations for any weather station in the world.

Here's what the autocorrelation analysis for daily maximum temperatures look like for 30-d periods throughout May and June (start dates of May 1 to May 30).

What is important in this graph is the correlation after 30 d. Essentially it says that the correlation coefficient between the 30-d starting in May and the following 30-d is less than 0.2.

How helpful is that?

Here's data for Konza (close to Manhattan) for 1984-2010, which I've been using for analyses lately.

X-axis is mean maximum temperatures for day of year 120-149 (roughly May) vs. same temperatures in the next 30 d.

Correlation coefficient of this is about 0.3 (similar to autocorrelation analysis).

You can see that there is about 9 °C variation among years in May temperature. If it's the hottest May on record, June is likely to be just 1°C hotter that average. That's not much predictive capacity.

And the 30-d after that? r = 0.2 and no significant predictive capacity. It could be equally likely to be the hottest month on record as the coldest.

For precipitation, it gets even worse.

here's the same autocorrelation analysis for precipitation during that period.

Essentially for the 30-d following May, there is no significant predictive capacity (maybe slightly drier).

 Precipitation between the two 30-d periods at Konza: flat.

As a caveat, i should say that these patterns have nothing to do with general trends over long time periods associated with global warming. Only the structure of seasonal variability.

That said, it's probably more certain to say that the future is going to be hotter than it is to say that the next month is hotter (or colder or wetter or drier).

At least here in Kansas.

No comments:

Post a Comment