Latest Drought Science Alarming for US

On this blog in the past, I have extensively discussed the drought research of Aiguo Dai and others.  Today I'd like to add coverage of a new paper in Nature that Dai has published.

First, let's quickly review the backstory.  In the past, it was discovered that the Palmer Drought Severity Index (a long-standing popular measure of drought), when applied to a suite of 21st century runs of climate models prepared for the 4th IPCC report, suggested that there would be  increasingly widespread global drought due to global warming:

This was extremely alarming with most of the world's agricultural zones badly impacted by severe drought by mid-century.  However, there was one hopeful caveat - at least to those of us in the US.  Using a statistical procedure (principal components analysis) to extract the dominant trend from the actual observational data (not the models) showed a map of the drying/wetting trends like this:

Here most of the US is in a wetting trend - in contrast to the average model run, which had the US drying out noticeably already.  So perhaps the models somehow get the regional distribution wrong and the US will continue to get wetter in future?

Dai has now explicitly addressed this question, and while I think his case is less than cast-iron at present, the evidence suggests that the answer is no - the US just got lucky in the sixties through the nineties, and its luck is unlikely to continue to hold.

So what is new in this paper?  There are several technical developments.  Firstly, Dai is now using a suite of model runs from the CMIP5 runs (which can probably be viewed for practical purposes as close precursors of the runs that will form the foundation of the modeling section of the AR5 report that will come out next year).  So these are new, presumably better, models.

Secondly, it is well known that episodic droughts in the past have a lot to do with fluctuations in sea-surface temperatures.  If a large section of the ocean off the coast of some particular land region is a few degrees warmer, that can have a large effect on the moisture supply to that region (recall that the water holding capacity of air increases about 7% for every degree celsius increase in temperature).  There are chaotic fluctuations in the ocean circulation that cause multi-decadal changes in where the warmest water is and these can control droughts and wet periods on land.  This motivates Dai in this latest paper to switch from using principal component analysis on the PDSI data alone to using maximal covariance analysis on both the PDSI and sea surface data.  Maximal covariance analysis is a cousin of principal components analysis in which instead of trying to find the main modes of variation in a single data set, we try to find the main modes that cause two datasets to covary together (in this case PDSI - the drought index - and sea surface temperatures).

However, it turns out that making this change results in very similar results - once again the first MCA mode turns out to extract the global linear trend (of increasing PDSI and increasing sea surface temperature), and the second MCA mode captures the El-Nino/Southern Oscillation (ENSO) phenomenon.  And once again the US appears to be getting wetter in the first mode:

Again, this was not true in the ensemble of models:

However, what is particularly fascinating is to compare the first MCA mode for the observed SSTs versus the ensemble of models:

Observations are on top, models below.  Clearly, the average of the model runs shows a much more even, less lumpy warming of the oceans than the observations.  Dai argues this is because there's only been a relatively small amount of global warming so far and so random fluctuations can get incorporated into the trend.  In particular, he shows that some of the individual model runs in the twentieth century also have lumpy distributions that are different from the average, but in quite different ways from each other, and from the observations.  In other words, the observations show a particular set of slow natural fluctuations in the ocean and global warming through 2010 is not strong enough to overwhelm them entirely in the trend.  In particular, for the US, much of the trend is explained by something called the Interdecadal Pacific Oscillation (IPO), which caused the US - particularly the southwest - to get wetter from the sixties to the nineties, though it's now reversing:

(The green line is an index of the IPO and the black lines are increasingly smoothed versions of the PDSI for the southwest US).

However, whereas over the twentieth century, the model runs show a first MCA mode that varies quite a bit from model to model and only explains a small amount of the total covariance, once the 21st century is included, the picture changes greatly.  Now the effect of global warming is so great that all the models have similar MCA first modes, and that mode explains far more of the total covariance.  So Dai's argument is that the same thing will happen with the observations - as global warming proceeds, it will increasingly overwhelm natural sea surface temperature fluctuations, and the overall drying trend that the models show will assert itself in the US also.  Indeed, arguably, this has already begun in the 1990-2010 period in which the US has been drying (culminating in the serious drought this summer).

If this is right, then we can expect massive serious droughts to become an increasingly common feature of US life in coming decades.