# Global Warming Stopped in 1997, Yeah Right

## An Eccentric Anomaly: Ed Davies's Blog

David Rose, writing in the Mail on Sunday claims Global warming stopped 16 years ago, reveals Met Office report quietly released... and here is the chart to prove it. Lots of people have already bashed this but I think I can provide a perspective that might be helpful to some.

Yes, if you do an ordinary least-squares regression on that time period, using the HadCRUT4 global monthly data, the line is pretty flat:

Actually, by my calculations the trend is not quite zero; it's about 0.003 °C/year. That's not a lot and I can well believe it's not statistically different from zero. On the other hand it's worth emphasising that, once autocorrelation is taken into account, it's not statistically different from the overall global warming trend either (says Tamino, who seems to understand these things a bit better than most).

To be fair it seems that we ought to plot the trend for the early part of the modern global warming era as well:

Humph, that's odd - the temperature seemed to jump up by about 0.14 °C during 1997. Do we see articles in the Daily Mail titled something like “Decade's worth of global warming in one year - we're all doomed”? Thought not.

If we're going to pretend to think that something of interest happened in 1997 (other than the beginning of a strong El Nino which pushed the global temperature up a bit) then perhaps we ought to do a least-squares fit of the trends before and after 1997 but eliminating the rather implausible jump in the temperature by constraining the two trends to meet in that year.

That's not so easy with off-the-shelf software that I'm familiar with so I wrote a bit of Python code to do it. It does the OLS fits shown above as well and fires up Gnuplot to produce the graphs shown here.

(This code is a bit over-engineered for the job it does. It was quicker to just re-purpose some existing code I had rather than write something minimal.)

The output it produces for the two methods (OLS for each of the two time periods independently) then a least squares fit constrained to having the two trends meet is:

```
Start temperature anomaly (1975-08): -0.070129 °C
First temperature trend (1975-08/1997-07): 0.016065 °C/year
First temperature anomaly (1997-08): 0.283293 °C
Temperature 'jump' at 1997-08: 0.142112 °C
Second temperature anomaly (1997-08): 0.425405 °C
Second temperature trend (1997-08/2012-07): 0.003155 °C/year
End temperature anomaly (2012-07): 0.472461 °C

Start temperature anomaly (1975-08): -0.098340 °C
First temperature trend (1975-08/1997-07): 0.019920 °C/year
First temperature anomaly (1997-08): 0.339900 °C
Temperature 'jump' at 1997-08: 0.000000 °C
Second temperature anomaly (1997-08): 0.339900 °C
Second temperature trend (1997-08/2012-07): 0.011730 °C/year
End temperature anomaly (2012-07): 0.514873 °C
```

When we plot the constrained trends we get:

This shows that, with the jump eliminated, there's even less difference between the trends before and after 1997.

In summary, if you want to draw straight lines with this data looking only at fairly short periods of time then you really have to also admit to the resulting jumps in temperature where the periods join. Or you have to look a bit more widely.

More importantly, of course, you also have to consider whether your periods are long enough to get statistically significant results. Working out the likely error bounds for this constrained fit of two trends is way beyond me. If somebody else can do that then fine, but this post is mostly meant to just illustrate one pitfall of looking at a too-short period of data.