Notes and Musings

Paul's thoughts on developing and using WoodForTrees, interesting results, and what they might mean...

WTI: The WoodForTrees Temperature Index

When playing around with temperature graphs, I always found myself having to choose which of the four global temperature sources - HADCRUT3, GISTEMP, UAH, RSS - to use. Since they all have their differences, particularly around short-term responses to extreme events like the 1998 El Nino, I thought it would be nice to have an average of all four...

Hence I've created the WoodForTrees Temperature Index (WTI). This is created from the mean of HADCRUT3VGL, GISTEMP, RSS and UAH, offset by their baseline differences. It covers only the time period where all four series are valid, so begins in 1979 and will only contain the latest month's values when all four sources are in. It is updated from the master sources at 3am GMT/BST each night.

Baseline adjustment

Technically, the series is:

WTI = mean(HADCRUT3VGL-0.15, GISTEMP-0.24, RSS, UAH)

or equivalently:

WTI = mean(HADCRUT3VGL, GISTEMP, RSS, UAH)-0.0975

The adjustment brings down the baseline, so the series is expressed as anomalies from 1979-1999 monthly averages (same as RSS and UAH).

Of course, adjusting the baseline doesn't make any difference when you're looking for trends or cycles.

Choice of sources

The four sources used are the four global sources most often quoted in climate studies, plus there are two land/sea-based (HADCRUT3, GISTEMP) and two from satellite air measurements (RSS, UAH). Hopefully the combination gives better accuracy than any one of them individually.

See the credits page for source information.

WTI compared to original sources

Here is WTI compared to the other four series with baseline adjustments over the full range:

and here is the latest year's data:

Temperature trends - pick a timescale, any timescale!

After many requests, I finally added trend-lines (linear least-squares regression) to the graph generator. I hope this is useful, but I would also like to point out that it can be fairly dangerous...

Depending on your preconceptions, by picking your start and end times carefully, you can now 'prove' that:

• Temperature is falling!
• Temperature is static!
• Temperature is rising!
• Temperature is rising really fast!

Here are all four of the above trendlines plotted together:

What you find can depend on where (or when) you look!

Personally, I prefer the long view, and now we have trendlines, and adjusted anomaly baselines, we can throw it all together into one monster plot:

If you look at the trend data, you can see the current trends in °C, between 0.13-0.17°C/decade, or, if it continues at the same rate, between 1.3 and 1.7°C per century.

Comparing temperature anomalies - getting the baselines right

The main temperature series we have on this site - HADCRUT3, GISTEMP, UAH and RSS - are all expressed as monthly temperature anomalies from a defined baseline period. This means that the average temperature for each similar month (all Januaries, all Februaries, etc.) is subtracted from the monthly value to remove any seasonal cycle, and (in theory) any difference between the absolute starting positions of the series.

Different baselines

Why 'in theory'? Well, the problem arises because the four series use three different baseline periods (UAH and RSS use the same). Here are the baseline periods as reported by each source:

Source	Baseline period
HADCRUT3	Jan 1961 - Dec 1990 (30 years)
GISTEMP	Jan 1951 - Dec 1980 (30 years)
UAH	Jan 1979 - Dec 1998 (20 years)
RSS	Jan 1979 - Dec 1998 (20 years)

Now take a look at all four series from 1979 (the period in common), unadjusted, but slightly smoothed:

Clearly they are very similar, but there is an offset between UAH/RSS (which are roughly the same) to HADCRUT3, and again to GISTEMP.

If you think about the different baseline periods, the reason for this is obvious. GISTEMP has the earliest baseline period, when temperatures were cooler, so its anomalies from this baseline are always higher. HADCRUT is somewhere in the middle, and UAH/RSS have the most recent, warmest baselines, so their anomalies are lowest now.

Estimating the offset

To make a fairer comparison between the series, we need to know what these offsets are, and we can then correct for them. As UAH/RSS are the lowest, and have the same baseline period which also overlaps the other data ranges, let's use their baseline period - Jan 1979 to Dec 1998 - as the period of comparison. We will then calculate the average anomaly of each series during this period to get an approximate offset between them.

Conveniently (in fact, just for this purpose!), the data output for all four series for 1979-1999 will give us the mean values at the end of each data set (look for lines beginning "#Mean:"). Note that we quote 1999 as the end month, because it is "up to but not including", so the last month is Dec 1998.

Here are the means for 1979-1999 for each source, to 2 decimal places:

Source	Mean
HADCRUT3	0.15
GISTEMP	0.24
UAH	0.00
RSS	0.00

The means for UAH/RSS are zero, as we would expect, since this is their defined baseline period. Note that these are annual means, but the anomalies are specified as monthly differences; hence to do this perfectly we'd have to adjust each month independently, but I think this will do!

Shifting the sources

So, now we can shift the sources to the same baseline period, using an 'offset' step:

This clearly removes the overall offset, but there are still differences in range, particularly in the peaks around the 1998 El Nino. My guess is the satellite air temperature sources (UAH & RSS) are more sensitive to short-term influences like this than the land and sea temperature ones (HADCRUT3 & GISTEMP).

Armed with our magic offsets, we can now do a fair comparison of recent history:

This graph will stay up to date with the latest year's values, so feel free to copy the image link to your own site, but please link back to these notes so people can understand it and play with it themselves.