Regime Change

Last week, a headline in the Bangor Daily News read: “Heat grips much of US, so get used to it.” It was for a story from AP about the unusual early-June heat wave which afflicted much of the U.S., and includes the report that “A new study from Stanford University says global climate change will lead permanently to unusually hot summers in the coming years.”

The new Research is by Noah Diffenbaugh and Martin Scherer at Stanford University. They studied both the output of computer models for the 21st century, and historical temperature data as well. Their results don’t just indicate that seasonal temperatures will enter a new regime, but that they already have. “It turns out that when we look back in time using temperature records, we find that this extreme heat emergence is occurring now, and that climate models represent the historical patterns remarkably well.”

They took a different approach than most who study temperature change. Instead of simply looking at the expected change in average temperature over a variety of regions, and over various seasons, they studied how temperature is expected to change relative to the amount of variation it exhibits now. This can dramatically alter the outlook, because different regions and different seasons show differing amounts of variability. If a given season in a given region shows little temperature variability, then even a small change can force it into a new regime; if a season for a region shows large variability naturally, then a larger change is required to bring about regime change.

For example, I took all the temperature data in the GHCN for stations located between the equator and latitude 10 deg.N, aligned them (using a modifed Berkeley method), and generated a combined data record. Then I averaged June/July/August for each year to get an annual series of summer (JJA) values, and December/January/February to get a series of winter (DJF) values. Here’s the summer result, together with a modified lowess smooth:

Note that this is not to be taken as a latitudinal temperature history because it’s not an area-weighted average — the purpose is to estimate the variability rather than the mean value. The range is 1.42 deg.C from lowest to highest, the standard deviation is 0.29 deg.C. We can also remove the secular trend by subtracting the lowess smooth, giving this:

and now the range is a mere 0.95 deg.C, the standard deviation 0.19 deg.C. Clearly, the tropical summer doesn’t show very much variation.

If we look at the same data sets for winter (DJF) rather than summer, we get this:

Note that the range is now 2.51 deg.C from smallest to largest, more than twice the range shown in summer (JJA), and the standard deviation is 0.29 deg.C, nearly twice the summer value. Likewise if we de-trend the winter data:

The range is now 1.42 deg.C (nearly 50% larger than the summer value) and the standard deviation is 0.28 deg.C (also nearly 50% larger than in summer).

The arctic is expected to warm much faster than the tropics, but it also shows much larger natural variation. If we combine GHCN stations from 60 deg.N to the pole, we get this for summer (JJA):

and when detrended, this:

For the raw data, the range is a whopping 8.02 deg.C and the standard deviation 1.88. Detrending reduces the range to 2.21 and standard deviation to 0.40. All these values are much higher than their counterparts for the tropical region graphed previously.

The same is true in winter:

and when detrended:

The raw range is 9.84 deg.C, standard deviation 1.62, while the detrended range is 4.66 deg.C, standard deviation 0.97. All this is hardly a rigorous study of the change in variation with region and season, but clearly the amount of year-to-year variation in seasonal temperature is strongly dependent on those factors. In fact for either raw or detrended data, the standard deviation of the Arctic winter is greater than the entire range of variation of the tropical summer.

The smaller variation in tropical regions, and during summer, makes those regions and that season more susceptible to regime change simply because a smaller net temperature increase exceeds the present range. Diffenbaugh and Scherer found, in fact, that not only is summer likely to enter a new regime sooner than winter, the tropics are likely to enter a new regime sooner than other parts of the globe, in spite of greater warming outside the tropics. Here, for instance, is their estimate of the fraction of seasons (either winter DJF, or summer JJA) which, at future dates, will be hotter than the hottest such season during the 1980-1999 period:

Note that by the end of the 21st century, over a vast swath of the globe we expect every summer (even the accidentally mild ones) to be hotter than the hottest summer from 1980-1999. Most of that happens where I wouldn’t have expected it to: near the equator.

Here is their estimate of the time at which the “regime change” will be complete, in the sense that the average season will be hotter than the “spread” (mean plus one standard deviation) of that season during the 1980-1999 period:

Again it’s noteworthy (and frightening) that sizeable regions will have entered a new regime as early as 2030.

Diffenbaugh and Scherer point out that for regions with lower variability, smaller temperature change can not only introduce a new temperature regime, but by doing so usher in significant biological and social hazard. The fact that the tropics don’t warm as fast as other regions doesn’t protect them from harm, because they already show less variation so both human and natural systems are less adapted to temperature variation.

They also point out that certain systematic biases which may be present in the computer models they used, rather than increasing hope for a safe future, tend to lean toward making their forecast too conservative. It’s entirely possible that their outlook is too optimistic — if such a description can seriously be applied.

So I’ll contradict the news story from AP which was carried by the Bangor Daily News. Don’t get used to it — i.e., what’s happening now and in the next decade or two. Because if you get used to that, you’ll be ill prepared for the worse that is soon to follow.

UPDATE: I goofed

A reader asked about the unexpected trends in the seasonal averages given above for extremely large regions. I think I goofed — because instead of combining station *anomalies*, I combine station raw data. That’s OK *if* the seasonal cycle is nearly the same for the stations being combined. But the regions are so large there are drastic differences between the seasonal cycles of different locations within the region.

When the station data sets are aligned, they’re offset to make the averages as close as possible during their time of overlap. If one station has a small seasonal cycle while another has a large cycle, then their *average* will be as closely aligned as possible, but during winter and summer their data will diverge the most. Hence stations with a large seasonal cycle will tend to make summer seem warmer and winter seem colder, while those with a small seasonal cycle will have the opposite effect.

If different stations cover different time periods, then when a station with a larger-than-average, or smaller-than-average, seasonal cycle enters or leaves the data record, it will have a disproportionate effect on the winter and summer seasonal averages. This can cause a false winter cold/summer warm trend, or a false winter warm/summer cold trend.

And that’s what seems to have happened with the above grid averages — especially since the grids are *huge* and we really can’t expect all the stations to have nearly the same seasonal cycle. After all, the tropical grid stretches around the globe so some points are 20,000 km apart, and the arctic grid cover 60N to the pole so some points are over 6,500 km apart.

I recomputed the huge-grid seasonal averages using anomaly (to remove the seasonal cycle) rather than raw temperature. It still shouldn’t be taken as a regional average because it’s not area-weighted, but at least it no longer shows the bizarre trend which really is due to differing seasonal cycles between stations. Here’s the tropics during summer:

The tropics during winter:

The arctic during summer:

The arctic during winter:

By the way — it’s still the case that the arctic shows much more variation than the tropics, and that winter shows more variation than summer, so the essential point is unchanged. But now the huge-grid averages don’t show that freaky, unexpected time trend.