Temperature variation at Heathrow over the past year

Posted: August 31, 2015 by tchannon in Analysis, climate, methodology, Natural Variation, Surfacestation, UHI, weather

This article is part of preparing the way for later revelations about instrumentation defects.



Figure 1 (upper), Figure 2 (lower) computed mean insolation for horizontal surface at this exact location and weather parameters, no cloud.

Figure 1 (upper), Experimental work[1] showing nearly daily temperature variation from expected, specifically designed to exclude diurnal but include detail variation at the fastest scale feasible. Time graticule at 10 days, data points at 12 hours. Surprisingly the July 1st hot period has vanished. Plots of other sites show a similar effect. The most frequent warm and cool periods of weather are brief and readily seen.

This computation will produce different values from the mean values computed from thermometer minimum and maximum data because data shape at other times is taken into account, min/max does not. The filter used is also windowed, leakage is negligible.

Extracting a climatology from just over a year of hourly data, the whole archive length here, is impossible by conventional means, nor are these available for all Met Office Datapoint stations, including that some are recent stations. The solution uses existing innovative software by the author which will non-discrete Fourier match data, in this case locked on the almost wholly dominant two terms, 12 and 6 months. Checking the results against published climatology on a few stations shows good agreement. No attempt has been made to extract an absolute climatology (a mean temperature) but is nevertheless in good agreement.

Evidence not shown here (Met Office data) points to high maxima tending to occur earlier in the year than the peak mean temperature.

Insolation is highest during June (figure 2). A time delay of a couple of months between insolation and mean peak is typical of sites. Exceptional temperature perhaps tends to be more about high elevation sunshine and clear skies: as the year progresses rainfall increases and sunshine decreases, see other works by the author on UK parameters where the annual variation is shown.

The shape of the annual variation varies considerably across the UK but this is a new work on 170 stations so much more can be investigated. For example in the far north the annual insolation curve is significantly asymmetric and so is the temperature.

Several oddities have turned up, unexpected variations which seem to have a pattern relating to station and station geographical location. Some of this may be related to UHI and similar thermal mass / water effects.

Heathrow shows some degree of one of the effects, the shape of the temperature builds up during the year. Bear in mind that high maxima from sunshine is immediate whereas high mean comes from thermal mass heat, so this kind of effect is normal but might sometimes be unnatural. A phase delay.

1. Dinurnal removal filter. This is a compromise giving reasonable rejection whilst not rejecting at two days and with tolerable impulse response. Interpolating over single or a few missing data points is accepted as-is, longer periods have been linearly interpolated. These are human calls where the intended result context must be considered.


Output data sample points at 12 hours, therefore Nyquist requires zero at >=  <= 24 hours. Response at 48 hours is 0.86, at 60hr, 0.97

Readers who are not familiar with filtering:, counter intuitively no data is lost as such, the process here is a trade between time resolution and amplitude resolution which becomes finer, if invisible on plots. This process is not communicative, irreversible.

The intended effect is reducing a huge number of data points and removing the large day/night change. What remains is the general variation taking into account the daily change, eg. a hot day with cold night cancels and so on.

Post by Tim

  1. tchannon says:

    This article is fairly uninteresting so I am not expecting much comment or interest. The surprises come later when I start showing what was and is going on in detail in the instrumentation. This raises questions which need answering.

  2. E.M.Smith says:

    Well I found it quite interesting. I did something of an “eyeball filter” on a monthly data graph set for a transect of the USA here:


    and was pondering more mechanical, analytical methods to quantify the visual impression. Now I have an example, looking at the annual cycle, to model upon…

    What I saw was very similar. Daily cycle min temp set by latitude (during one month, so seasonality removed) with max set by water (proximity of oceans or presence of rain & humidity) and cold excursion from rain events.

    I think the latitude correlation would be even closer with a lapse rate adjustment.

    So here you find a nice match to solar angle AS the season changes in one place, reinforcing the solar correlation. One site as season changes, while I did one season as longitude changes, and a smaller set on latitude changes. Very similar answers.

    Now I just need to figure out how to apply your technique to the geography domain… or you could do it for me 🙂

  3. tchannon says:

    Odd you should mention geographic variation, just finished a quick look at ancillary data from the UK work. The results are a little unexpected, such as the east / west effect on annual cycle correlation, maybe… Or perhaps as much on distance from the sea.

    If I wanted to follow this up one tack would be to show the data to classification software. This would I think pick out classes. Interesting? Possibly not other than tending to automate librarian work, state the obvious.

    Another ploy I used was adjust the mean for station altitude, which has dramatic effects on rankings. Why we don’t normally do this is a mystery but there again I have long bemoaned the terrible state of world data where both station altitude and station pressure are missing. GHCN having gone as far as refusing to produce a pressure dataset. My reasoning is based on knowing instances of station moves revealed by altitude change and the temperature data changed too. (this is about WMO station subsets)

    If anyone wants to have a play I can upload a small CSV
    WMO name latitude longitude altitude r2 temperature for the 150+ UK sites used

  4. tchannon says:

    Continuing, I nearly commented last night EMS. This was a very different take about the pointless nature of “back radiation”, explaining what is going on and pointing out there is no instrument network able to measure this minuscule misdescribed signal and yet this is a primary data for AGW. Missing data and no-one can be bothered to collect it?

    If you want to compute the insolation I can give you the code. This is a private library, a wrapper around a more or less public library. You can handle C, just needs a makefile, pretty portable (you use linux so Lua is just install). Alternatively roll you own driver code. Minimum input is date and lat/long, add altitude, temperature and air pressure if available.
    The snag is if you want an annual curve, insolation moves around, not quite simple. I actually took hourly values and low pass filtered, this automagically produces the annual. Sun angle is another output parameter, depends what you need. It might be enough to take the midday solar angle.

  5. ivan says:

    Ah, back radiation for which we have no reliable instrumentation outside the lab. So they make up figures that fit their theory rather than fitting the theory to the non existent figures.

    I will be watching where you are going with this series with my engineers hat on. What you are doing is what the climate scientists should be doing and publishing but the wont because it doesn’t fit the requirements of the green religion.

  6. E.M.Smith says:

    I would love to have a look at the code. Been putting off a sun angle / insolation code writing session and would not mind putting it off forever in favor of already written code…

    FWIW, a session of graph inspection seems to show coastal and island stations summer min limit set by water but winter max looked clipped too. Like summer sun driving hot excusions for a few hours while in winter nightly radiative is the excursion while max returns closer to long tern average. I think a solar source, ground / water heat store model is likely best.

    At very high latitudes, things go wonky. Looks like days to weeks on end driven by mobile polar air masses and few daily sun cycles. Only looked at Aug and Jan though. Nome Alaska is rain and snow modulated even in August. So may also be an ice and snow switch at high latitudes coupled with a more annual than daily sun cycle.

    To me it looks like polar has an ice point hump, but once frozen you can just radiate to space to extreme cold, that then slides toward the equator, while equatorial is hot limited by thunder storms and hurricanes (water cycle heat pipe to altitude), and dry temperate is a ground heat store solar pumped with winter max aprox equal to summer min. Excursions each way representing heat in or out of the storModeling that looks devilish. ..

  7. Tim, SOI is calculated from pressures. You should be able to find other data for Tahiti and Darwin. Further these are long term data.

  8. tchannon says:

    There are two freely available codes for insolation.

    1) newer fancy one but is licence encumbered so I won’t look at it.

    2) An old perfectly good plain C version for year 1950 .. 2050 with a more sensible licence but there was fun over download file format, reported to no reply. (IIRC mixed line ends, such as Apple, easy when you figure it out)
    This is classic old C, core and console wrapper.

    I’ve written a simple Lua binary library wrapper for the core. Point an ANSI C89 or similar at the code, link with the correct Lua core library, etc.

    What to do and my code is in this massive .zip https://tallbloke.files.wordpress.com/2015/09/lua_solpos-01a.zip (5kB)
    No doubt broken, quickly knocked up.