Sunday, December 13, 2009
Peer Review for the Climate "Science" Coalition
A couple of weeks ago it got a bit more serious. The New Zealand Climate "Science" Coalition (C"S"C) put out a press release [PDF] that accused NIWA of scientific fraud. If you missed the story my sciblings Gareth and Ken have covered it in detail. The short version is the C"S"C collated temperature records for each of the regions that contributed to NIWA's national long term climate data series and found no significant trend in the raw data despite the fact NIWA's series has New Zealand warming by almost a degree in the last century. So, presumably in an attempt to put a local angle on the "climategate" story they released that extraordinary press release. I reproduce some of the highlights below.
... the station histories are unremarkable. There are no reasons for any large corrections. But we were astonished to find that strong adjustments have indeed been made.
The shocking truth is that the oldest readings have been cranked way down and later readings artificially lifted to give a false impression of warming, as documented below. There is nothing in the station histories to warrant these adjustments and to date Dr Salinger and NIWA have not revealed why they did this.
We have discovered that the warming in New Zealand over the past 156 years was indeed man-made, but it had nothing to do with emissions of CO2 it was created by man-made adjustments of the temperature. It's a disgrace.
New Zealand's contribution to the global statistics is now under a shadow, so there could be regional or even global implications of these disgraceful adjustments which should be investigated.
Now we must ask: do we really need an ETS? For, if all that nasty carbon dioxide and methane we are pumping into the atmosphere has utterly failed to increase our temperature until now, why ever should it do so in the future?
Got that? There has been no warming in New Zealand, NIWA scientists purposefully changed the data to fit their agenda, this is a disgrace (and indeed shocking, disgraceful and astonishing) now that we know that there has been no warming in New Zealand we can dump the ETS. Since that release they've walked their claims back somewhat, in that annoying "oh, we were just putting it out there" way people that say outrageous things usually fall back to but it was the claims in their press release that started the story. The basis of all those fervid claims, that the adjustments made to the NIWA data were arbitrary, can only be described as a lie. Since the C"S"C stitched together the data for each region from multiple stations they knew very well that the some of NIWA's adjustments were made to take into account the change in recording station used to take the region's temperature. Moreover, as Ken has pointed out, if the C"S"C really wanted earn the "S" in their name they could have looked at the "flat" data they were presenting to see if stitching together data from those different stations without accounting for the change was more or less "shocking" then NIWA's adjustments.
Since the Wellington series has the simplest station history let's use that as an example. Before 1927 Wellington's temperature was recorded at Thorndon, just above sea level. In 1928 the station was moved to Kelburn about 120m above sea level. It will be news to no-one that as you get higher is gets cooler and, in fact, the C"S"C found a cooling trend in the Wellington of 0.7° per century. But even without knowing about the alititude of each station we can test for an effect of the station move in the raw data. Statistical tests have a horrible way of boiling down to a series of p-values - when you run tests on the raw data it turns out that the change of station has a significant effect on the temperature recorded but what does that mean? Perhaps the easiest way to understand just how wrong it would be to draw conclusions from the raw data without including a station effect is to compare how well each of a set of different models explain the raw data.
The C"S"C's -0.7°C trend comes from a simple model - a single linear trend over the length of observation. You could also imagine a model, lets call it the "altitude effect" model , in which a single linear trend applies over the length of observation but temperatures at the Kelburn station are offset by some factor (also estimated as a best fit) as a result of that stations elevation. You might even imagine another model "two slopes" in which as well as being offset by some factor temperatures at Kelburn were following a different trend than those down at sea level. I ran those regressions and got a bunch of numbers:
|Comparing our models|
|Model||Rate (95% CI)||Variance explained||AIC|
|C"S"C||-0.21 (-0.43 — -0.02)||3%||213.18|
|Altitude Effect||0.72 (0.37 — 1.07)||25%||179.82|
|Two Slopes||Before move||0.11 (-0.50 — 0.72)|
|After move||1.01 (0.27 — 1.75)||28%||175.80|
And indeed some pretty graphs:
The first thing that you note when you look at the table above is I didn't reproduce that -0.7°C trend even after going to *shudder* excel to check - I'm sure there is nothing untoward but going on but it is strange. For our purposes the values in "variance explained" and "AIC" are more important. The first of those is fairly self explanatory - how much of the total variance in the dataset is explained by the given model, AIC is a measure of how well the model fits scaled for the number of parameters used in the model ( which is a good thing to check because you can almost always explain some of the variance in dataset by adding another variable whether its meaningful or not) lower numbers represent a better fit. Between the table and the graphs we can see what it means when I say the station move has a significant effect on the temperature recorded - not including it in our estimate of the trend ignores a major source of variation in the data. When the move is incorporated we estimate a trend of 0.72°C per century ( and we can be 95% sure that value falls between 0.37°C — 1.07°C ) over the length of observation. Interestingly our third model is actually the best of the bunch. This is likely an artefact of trying to put a straight line through a trend that really isn't linear. Warming in the second half of the 20th Century was about twice as fast as earlier warming, since only the Kelburn observations include this accelerated warning fitting a line through that section separately fits the data better (though the uncertainty in our estimate of the trend is much larger since their are less data points making it up)In reply to comments from Ken the C"S"C, who had called their press release a "research paper", claimed that a process equivalent to peer review would not have invalidated any of scientific claims that they made. In fact, even a very simple look at their data have shown that drawing conclusions of the sort the C"S"C did from the raw data would be terrible science and accusing NIWA scientists of fraud on that basis really is disgraceful.