Guest essay by Mike Jonas
On 10 January, WUWT published my post A new system for determining global temperature. In that post, I described a proposed system for determining global temperature from the historical temperature record.
In that post, I indicated that another objective of the proposed system was to develop an improved model of global temperature : “After a while, the principal objective for improving the model would not be a better global temperature, it would be … a better model.“.
In this post, I explore that objective of the proposed system – a better model – and how the BEST project in particular could contribute to it. [BEST – Berkeley Earth Surface Temperatures].
In order to understand this post, it may be necessary to first read the original post. However, it will not be necessary to understand how the proposed triangulation process works because that is unimportant here (I use a 1-dimensional (“1D”) trial run to illustrate the idea).
The system would still provide global and regional temperature indexes, but these just become by-products (and I think that the size of their error bars might shock some people). The principal aim becomes learning about Earth’s temperatures, weather and climate.
In an earlier post (here) I stated that the climate models were all upside down because they tried to construct climate bottom-up from weather, and that instead the models need to work first and directly with climate. In my 10 Jan post I said that the temperature organisations had their logic upside down because they tried to adjust temperatures to match a model when they should have been adjusting a model to match the temperatures. This post is another step towards changing the mindset.
There is a lot to like about the BEST project.
BEST’s project goals (here) include: “To provide an open platform for further analysis by publishing our complete data and software code as well as tools to aid both professional and amateur exploration of the data“. This is a brilliant goal, which offers a very positive way forward, as I will attempt to explain in this article.
BEST have also, very commendably, put together a much larger set of temperature measurements than is used by other organisations.
What is less good is their mindset, which needs changing. They are using their own notions of temperature trends and consistency to fill in missing temperature measurements, and to adjust temperature measurements, which are subsequently used as if they were real temperature measurements. This is a very dangerous approach, because of its circular nature: if they adjust measured temperatures to match their pre-conceived notions of how temperatures work, then their final results are very likely to match their pre-conceived notions.
We (collectively) should make much better use of the historical surface temperature record (“temperature history”) than simply trying to work out the “global temperature” and its trend. With all its faults, the record is a significant asset, and it would be a shame not to try to make maximum use of it.
The main objective should be to use the temperature history to learn more about how Earth’s temperature / weather /climate system (“TWC”) operates.
The basic idea is simple :
• Build a model of how mankind’s activities and Earth’s weather and climate systems drive surface temperatures.
• Test the model against the temperature history.
• Use the results to improve the model.
Because the model contains both natural and man-made factors, it will be possible – after the model has been refined to be reasonably accurate – to determine the contribution of each man-made and each natural factor to Earth’s temperatures. [NB. Factors may combine non-linearly.].
As the model improves, so we learn more about the TWC. And maybe, along the way, we also get a better “global average temperature” index.
I will illustrate the ideas using a simple 1D trial.
I use January (mid-summer) temperature measurements from Australian stations roughly in a line from Robe on the SE coast of South Australia across Sydney to Lord Howe Island in the Tasman Sea [Bankstown Airport is in Sydney]:
Willis Eschenbach in his recent post used an Altitude and Latitude temperature field to model expected temperatures. If I align Willis’s factors with the measured temperatures along the line in Figure 1 so that their weighted averages are the same, the average difference between the individual temperatures is something like 2.5 deg C over the period 1965 to 2014:
But if I add in two more factors – for coastal/continental effect and urban heat effect (UHE) – then the average difference between temperature field (model) and measured temperature comes down to less than 1.5 deg C:
The two additional factors have been formulated very simply, to demonstrate how much difference it can make to use a better model. I don’t claim that these are the correct formulae for SE Australia in January, or that they can be applied elsewhere, or that others are not using them. They are purely for illustrative purposes. I have also simply combined the factors linearly where a real model might need to be non-linear. The formulae used were:
• Coastal/continental effect: Max temperature rises by an extra 8 deg C over the first 100km from the coast (inland Australia gets pretty hot in summer!). Min temperature falls by 2 deg C over the same distance. No further change further inland.
• UHE: Both Max and Min temperatures are higher by 1 deg C in urban areas. They both rise by a bit more towards the centre of the urban area, with the additional rise being larger for larger urban areas. Formula used for the additional rise at the urban centre is 0.5*ln(1+W) deg C where W is the approx diameter of the urban area in km.
NB. Figures 2 and 3 are actually for 1985, which is reasonably representative of the whole period. I can’t easily use station averages, because of missing temperature measurements:
Applying the Model
The model for Altitude and Latitude only, along the line from Robe to Lord Howe Island, looks like this:
The orange dots are the measured temperatures. It is easily seen that the measured temperatures and the model have different shapes. At this stage, it’s the shape that matters, not the absolute value. Under my proposed system, measured temperatures reign supreme, so the model has to be adjusted to match them:
Because the adjusted model must go through all measured temperatures, you can see that changing the model isn’t going to make much difference to the final average temperature …..
….. or can you??
Here is the same graph, but using the two extra factors – coastal/continental effect and UHE:
Note: The 5th dot from the left is above the line because it is a small urban area (Warracknabeal) that falls between the graphed grid points (5km spacing).
The following graph focuses just on the centre section of Figure 7, so that UHE is easier to see:
Now you can see just how much effect urban stations can have on average temperatures if you don’t treat UHE properly. BEST and others have come in for a lot of criticism (rightly so in my opinion) for the way they treat UHE. The various adjustments, homogenisations and in-fillings that the various organisations have made to temperatures have typically ended up spreading UHE into non-urban areas, thus corrupting their global average temperature index.
In my system, as illustrated above, UHE is catered for in the model, and is clearly confined to urban areas only. The average temperature along the Robe – Lord Howe line in Figure 6 is 28.3 deg C. In Figure 7 it is 27.1 deg C – that’s 1.2 deg C lower. The UHE contribution to the average temperature in Figure 7 is less than 0.1 deg C.
The model can and should be refined and extended to cover all sorts of additional factors – R Pielke Snr for example might like to add land clearing (I suspect that it would be significant between Robe and Cootamundra, where there is a lot of farm/grazing land).
Hopefully, the point is now successfully made that a change in mindset is needed, to get us away from the idea that temperatures must be adjusted, homogenised and mucked around with in order to make them match some idea of what we think they should be. Instead, we need to work with all temperature measurements unchanged. When we have developed a model that can get close to them it will tell us more about how Earth’s temperature/weather/climate system actually works.
Further supporting information is given below, after “Acknowledgements”.
Comments on the original post – especially the critical comments – have been very helpful. Some specific comments are referenced here, but many others were helpful, so I’ll just say “Thanks” to all commenters.
• richardscourtney said of my original proposed system, “That is what is done by ALL the existing teams who use station data to compute global temperature“. Maybe so, but I think not. For example, as demonstrated above, my system prevents UHE from being notionally spread into non-urban areas. I am not aware of any existing system that does that. As always, I’m happy to be proved wrong.
• Nick Stokes said that temperature adjustments made little difference. Since adjustments add to error bars (see below) it may be best to eliminate adjustments altogether. The proposed system makes this easy to do, and to evaluate, but it’s only an option. Note that below I completely rule out some kinds of temperature adjustments.
• Willis Eschenbach likened obtaining a global temperature from the temperature history as putting lipstick on a pig. The same comment may well apply here, but there is still the option of using only the higher quality data (if it can be identified). That would further eliminate any perceived need for adjustments.
• garymount described the open source Wattson project, which may well already be doing everything I describe here and more, but it hasn’t been documented yet.
• It will be interesting to see whether this article is seen as introducing anything new or different.
I say below that the proposed system is suitable for cooperative development. The point is that it is very difficult for anyone to identify the effect of any one factor in the temperature record in isolation. The temperature record is the end result of all factors, so all possible factors need to be taken into account. In a cooperative model, a researcher can be working on one factor but testing it in combination with all other factors. BEST, and maybe Wattson, could usefully act as central collectors, distributors and coordinators.
Changes from the previous post
The proposed system is essentially the proposed system in the previous post, but for those who insist that some temperature adjustments are needed I have included a section for them. The triangulation process is not needed in this post because I am using a 1D example, but it or something like it would be needed in the real system to obtain weightings and ratings and to match the model to the measured temperatures.
The system has three parts:
• The set of all surface temperature measurements, unadjusted.
• Adjustments to temperature measurements (please understand the process before complaining!!)
• Expected temperature patterns (“the model”).
These three parts are dealt with below under the subheadings “Temperature”, “Adjustments” and “The Model”.
Ideally, the new system will actually be implemented and it will be made 100% public for open-source cooperative development. Everything I say below can be interpreted in that light, ie. in a context where everything in the system can be replicated by others, tested, improved, etc.
As before, the actual temperature measurements reign supreme. They must be retained and used unchanged.
Associated with the surface temperature measurements, there needs to be accuracy data (error bars), eg:
• The accuracy of the thermometers themselves. Automated thermometers may be very accurate, but others have limited accuracy plus the risk of human error in reading them.
• Fitness for purpose : the risk that the temperature at the thermometer is not truly representative of what it purports to measure, ie. of the temperature outside. Poorly sited stations in particular would have large error bars.
• Provision for weather station deterioration, eg. failing paint.
• Instances where the measurements themselves are identified as unreliable. For example, some records show if a daily temperature is unreliable because it is estimated from a multiple-day measurement.
There could be blanket error ranges for the various types of equipment over time. So a super-accurate well-sited well-maintained automated thermometer would still be given error bars of say a few tenths of a degree C for “fitness for purpose”. An unautomated thermometer would have larger error bars, and an older thermometer would have even larger error bars.
Sorry, but I didn’t put error bars into my illustration above. It is only an illustration aimed at getting the basic idea across, and I felt that error bars at this stage would not assist understanding.
As before, the system probably cannot completely remove the need for adjustments to measured temperatures. With any adjustments, however, whether they are to fix obviously incorrect individual temperatures or to fix systemic errors such as time-of-day bias, there is always the risk that the adjustments themselves can introduce errors. In the proposed system, adjustments are held separately from the measured temperatures, so that they can be seen, understood, tested, quantified, and if required rejected.
All adjustments are perceived errors in the measurements, so for every adjustment the error bars should be increased. The best general rule is: don’t do them!
Some types of adjustment are not permitted, and these include many of the adjustments that are currently being done by BEST and others. These include things like infilling, station moves, “outliers” from some expected pattern, adjustment from surrounding stations, etc.
People are so used to station moves being adjusted for, that it may seem odd that adjustments for station moves cannot be allowed. The reason is that every temperature measurement has a location, so when a station moves the only change to iits known data is its (x,y,z) location and if that doesn’t change then there’s nothing to change. To single out the station for special treatment might itself introduce a bias. In my system, the “before” station, the “after” station, and all other stations are treated identically. In fact, all temperature measurements are treated identically. As I said in my previous post, a station with only a single temperature measurement in its lifetime is used on an equal footing with all other temperature measurements, and even moving temperature measurement devices could be used.
Here’s a really significant thought: If a particular station move is regarded as being significant even though it is a trivial location change, then it would be reasonable to estimate the amount of temperature change that the move could generate and then ensure that the “fitness for purpose” error bars are at least this large on all stations. The rationale is that there must be this much potential error at other locations.
The really important part of the system is the model. The set of temperatures is there for the model to be tested against.
The model contains all factors affecting temperature that anyone wants to test. The factors are combined to predict a pattern of temperature around the world’s surface over time. The pattern is then tested against the measured temperatures. Rather than try to describe it all in words, I hope that the simple 1D illustration above will suffice to get the ideas across.
Mike Jonas (MA Maths Oxford UK) retired some years ago after nearly 40 years in I.T.
Data and workings
All data and workings are available in a spreadsheet here:.ModelTrial.xlsx (Excel Spreadsheet)
BEST – Berkeley Earth Surface Temperatures
C – Centigrade or Celsius [If you don’t like “Centigrade” please see Willis’s comment on the original post]
TWC – Earth’s temperature / weather /climate system
UHE – Urban Heat Effect
1D – 1-dimensional