I am pleased to see that the attempt to promote a New Zealand version of “climategate” has more or less foundered. Sure the ACT party and some more extreme opponents of the findings of climate scientists are still campaigning (see for example Auckland Public Meeting: Climategate, NIWA and the ETS). And well know local climate change denier Ian Wishart managed to get international reporting of his slanderous press release (BREAKING: NZ’s NIWA accused of CRU-style temperature faking) in several more conservative and extreme international blogs and papers (for example BREAKING: NZ’s NIWA Accused of CRU-Style Temperature Faking, Climategate Scandal Spreads to New Zealand as MSM Continues Ostrich Act, Oops! Now New Zealand NIWA Accused Of Faking Data, New Zealand’s NIWA Gets Busted ’Tricking’ Their Climate Data and New Zealand Climate Data Shows Clear Evidence Of Fraud). But the New Zealand media has, in general, been more balanced in its reporting. The information from climate scientists at NIWA has been getting through.
Temperature trends for raw data
For example, NIWA’s information on the temperature trends shown by raw data from 11 local met stations (Temperature trends from raw data) has been picked up (Niwa publishes climate data to answer critics). NIWA released this because of the distorted information distributed by the NZ Climate Science Coalition, the Climate Conversation Group and Ian Wishart (Climate change deniers live in glass buildings ).
Some local bloggers picked up and reproduced this misinformation. In the process they have been slandering our local climate scientists and other bloggers who have attempted to correct the misinformation. My personal concern in this is not so much the facts of climate change, but the willingenss of some ideologically driven people to unjustly attack the integrity of honest scientists. And their willingness to distort information with this end in mind.
So it’s worth reproducing some of the latest information released by NIWA. This uses publicly -accessible information (from the National Climate Database). The data used is from the period after 1930 when there were no significant site changes. Consequently the raw data could be used without adjustments to determine temperatures at the individual sites. (You will recall that the climate change denier’s report and press release claimed that NIWA’s adjustment of data to accommodate met station site changes was fraudulent. And that they (the deniers) went ahead to combine data without adjustments and produced a misleading graphic suggesting temperature was unchanged over time. The current data should avoid all issues of adjustments).
NIWA’s information says in part:
We have analysed raw data from these sites directly, with absolutely no adjustments to the numbers from the NIWA climate database. Taking all sites together and averaging the annual mean temperatures (difference from 1961—90 mean at each site) results in Figure 1 below.
Figure 1: Temperature departures from the 1961—90 normal, averaged over the eleven sites listed in Table 1. For years where not all sites are available, the average is over those that do have records.
Note that not all stations have annual mean temperature values for all years in 1931—2008. It is common practice to in-fill isolated missing months, but we have deliberately not in-filled missing data here to keep this analysis as non-contentious as possible. For each year, the available station values have been averaged. In the title of the Figure, the ’p-value’ comes from a statistical test, and indicates the probability that the indicated trend could have arisen by chance.
If the two outlying Island records (Raoul and Campbell Islands) are left out, and the remaining nine records averaged, the result is as shown in Figure 2. In either case, the trend over the 78 year period is close to 1°C.
Figure 2: Temperature departures from the 1961—90 normal, averaged over the nine sites listed in Table 1 that are located on the main islands of New Zealand (i.e., all but Raoul and Campbell Islands). For years where not all sites are available, the average is over those that do have records.
![Reblog this post [with Zemanta]](http://img.zemanta.com/reblog_e.png?x-id=4428b5ff-dcb3-473f-b183-10d8d13aeb60)
Being an overcast Sunday I thought I’d have a closer look at the data now coming out from NIWA. First off I’d been surprised at the time it had taken to put the various responses together; it did seem they had to do it from scratch and that’s a bit of a worry.
One of the initial issues that was being debated was adjusting for station changes. I’d had a little bit of a poke around the literature and found a review article that gave some lie of the land for those of you that are interested (http://www.zamg.ac.at/histalp/downloads/abstract/Peterson-etal-1998-A.pdf).
However that all seemed a bit hard so putting that issue aside I decided to have a look at the “Temperature trends from raw data†(http://www.niwa.co.nz/news-and-publications/news/all/nz-temperature-rise-clear/temperature-trends-from-raw-data) just to see how all this is done without introducing the complexity of station adjustments. I should stress that my Statistics is pretty rusty so E&OE in what follows.
I went to cliflo.niwa.co.nz and pulled down the annual mean temperatures for each of the sites mentioned in NIWA’s PR.
A couple of things to be aware of. The particular stations do not all have data back to 1930 (as can be see from the NIWA release), but in addition when you pull them down there are gaps in the series (i.e. years where a mean wasn’t produced, presumably because some observations were missing). This was particularly acute for Molesworth, and Ruapehu has some significant gaps. I didn’t try and fix this and just worked with the data as it came down.
I then combined the data from slightly changed stations but in the same location into a single series. In a number of cases there were overlapping measurements, and I took the average.
The next step was to average across the 11 stations. The problem here is that the number of sites NIWA has recording from is four for most of the 1930s, 8 through to the mid 1940s, and the full 11 were not on stream until 1950. Therefore you can’t just take an average across the stations because those there is no guarantee that the ones that were recording in the 1930s had the same average temperatures as the final 11 (in fact you can guarantee they didn’t).
NIWA solve this problem by “normalising†the readings to some extent by subtracting the 1961-90 average temperature at each station from the averages produced from that station (at least that’s what I interpreted them as doing).
This kind of adjustment is problematic for a number of reasons. Most obviously because we are looking for a consistent estimate of the temperature in NZ over time, but have a non-random changing set of measures (even if their means have been normailsed). If we have good recordings from 11 locations over a period we could at least make statements with confidence about how the temperature as measured at those locations has moved. We could then investigate whether those locations are suitable representative of the temperature across NZ.
But even if we set ourselves the limited goal of estimating temperatures at the 11 sites, at the beginning of the period we only have a sample of four. Thus at the beginning the accuracy of the estimate of the 11 sites is lower, and the variance in the series is much higher. In practice I think you can see this if you look at the NIWA graphs – the pre-1950 observations seem to have more extremes.
Putting all this aside I did the adjustment as they indicated and ended up with graph that looked pretty much like NIWA’s. I then fitted a simple linear model using date as the independent variable and got a slightly larger coefficient for it than NIWA did. I got +0.0139 degrees p.a. (R2 of 0.4233) or a +1.07 change over the 77 year period from1931-2008 compared with NIWA’s +1.0 degrees change over the same period (I realised at a later stage that NIWA had started at 1931, I had started at 1930 which gave a sharper slope, but this doesn’t completely explain the difference). Notwithstanding this plus an examination of the graphs of observations together suggested we had undertaken broadly the same analysis. (I tried to post the graph but that doesn’t seem possible in replies.)
I should stress that all that is being done here is to fit a model to the observations.
NIWA on the other hand notes that the p test indicates that “the probability that the indicated trend could have arisen by chance†is less than 1 in 10,000. This might leave the reader with the impression that NIWA is saying there is a very high low probability that there is a linear trend in the observations, and that this will continue.
This is not so. The temperature across NZ is driven by complex systems. Sticking a line through a set of observations does little to help understand what is driving the observations. More particularly unless the models that are being fitted have some basis in empirical science the use of trends to predict the future is foolhardy (I noted in an earlier post that the passage of time causes very little).
So the first thing I want to demonstrate is that the choice of assumed model can lead to significantly different conclusions about the future for climate change. When I looked at the graph produced by NIWA my instinct was to say that the rate of temperature increase was declining over time. It looked as though the graph was levelling off.
Sure enough if I fit a logarithmic model to this data (that is one where the temperature is increasing but at a reducing rate) I get a slightly better correlations for the relationship (R2 0.4457) and a curve that looks a better fit (although the RH start point might not look so flash if we had included 1920s data).
Now I’m not advocating this as the interpretation, I am just making the point that any number of models can be fitted to a set of data and tell us little about the underlying process. We need more sophistication if we are to understand the processes, and NIWA’s presentation of the time trend leaves much to be desired.
One specific issue is the impression it leaves about future temperature increases. If you accept NIWA’s implicit endorsement of a linear model then by 2050 average temperatures at these locations will have increased by ~0.56 of a degree. If on the other hand the model is logarithmic (and I’d bet it isn’t anywhere this simple) the increase will be only ~0.15.
The real question is what is driving this increase. To end this post I’ll just include an example to show how if you start to get to what is likely to be driving the model you get better results. Looking at the graphs I saw how they were correlated with Southern Oscillation Index. While I don’t understand the mechanisms one would expect the El Nino/El Nina effects to have an impact on our temperatures (even if not necessarily a linear relationship). I therefore decided to stick the SOI into the model (I got the SOI info from the Australian Met Office and averaged monthly figures to give an annual series – I doubt this is strictly kosher but it will do). This significantly increased the R2 (0.6226 for the linear model, 0.6268 for the logarithmic although it does reduce the degrees of freedom). It also increased the coefficient associated with temperature i.e. including the SO made the underlying upward temperature trends more pronounced.
I’m sure that NIWA knows all this, and also that a linear time trend is unsatisfactory when it comes to describing temperature changes. What I don’t understand is why this complexity is glossed over in their press release.
As I said in earlier posts, pretending this is all simple (and understood) does no one any good. Not providing the detail and the complexity ultimately puts any judgements made off the science into doubt.
And bloggers (on either side) who mindlessly assert the infallibility of the science without understanding what is going on ain’t doing anyone any favours either.