Open Mind

Annual cycle in UAH TLT

October 30, 2008 · 11 Comments

In a previous post, we compared estimated global lower-troposphere temperature according to the analysis of RSS (Remote Sensing Systems) and UAH (University of Alabama at Huntsville). I gave my reasons for concluding that the RSS data are preferable to the UAH data.


One of those reasons was the presence of a strong annual cycle in recent UAH data. The data are anomalies, so the average annual cycle during the baseline period (1979-1999) has already been removed. Hence the recent annual cycle represents the difference between the estimated annual cycle, and its average during the baseline period.

It was pointed out in comments that the effect may be real; after all, “standard” global-warming analysis indicates that winter is warming faster than summer, so we would expect that at any given location the annual cycle could change. However, for the globe we expect this effect to be mitigated by the fact that the seasons are reversed between the hemispheres. If winter warms faster than summer, then the northern hemisphere should show more warming near the start of the year (NH winter) while the southern hemisphere shows more warming near mid-year (SH winter). These two factors will partly cancel each other in the global average.

But we still might see a change in the global annual cycle, because the effect of enhanced winter warming should be stronger in the NH than the SH. This is because most of the earth’s land mass is in the NH, therefore the NH shows more warming overall, and we could also expect greater net “seasonal” warming (winter minus summer). So we expect the SH annual cycle change to offset the NH change only partly. Since the NH should dominate, we expect the globe as a whole to show slightly more warming near the beginning of the year and less near mid-year. We also expect the effect to be slight.

These expectations are confirmed by analysis of GISS data for surface temperature. The baseline period for GISS (1951-1980) is earlier than that for satellite estimates, so there’s been more time for changes in the annual cycle to accumulate. We can use a wavelet transform to look for a “residual” annual cycle in GISS data, but it’s only weakly present:

The values of the WWZ (which indicate statistical significance, not physical amplitude) aren’t statistically significant, although the value for the NH around 1990 almost is. If we examine GISS data for land only, however, we get this:

We do have a statistically significant response in NH land areas around 2000 (the wavelet transform is tuned so that at any given moment, the “interval” being studied is a little over 5 years wide). We can also compute the physical amplitude (I’ll plot semi-amplitude, which is just half the amplitude) for land-only in the two hemispheres:

If we do the same for land+ocean (in spite of the lack of statistical significance), we get this:

It’s pretty clear that the response over land is stronger than over land+ocean, and that the effect is stronger in the NH than the SH. We can also look for the phase of the response (the time of year at which it’s maximum) for land-only for the two hemispheres:

As expected, the greatest warming is early in the year in the NH (closer to the beginning of the year for more recent times), but falls after mid-year for the SH. All these results are consistent with the hypothesis that the change of the annual cycle is due to greater warming in winter than in summer.

For satellite data we expect a smaller response because there’s less time difference between the most recent data and the baseline period. Other than that, we expect all the same phenomena: greater response in the NH than the SH, greater response over land than over ocean, and different phase of the response between the two hemispheres. What do the data say?

The cycle is definitely present globally in UAH data but not in RSS data:

That doesn’t mean the effect isn’t present in RSS data, only that it’s too small to be detected with statistical significance. But in UAH data it’s strongly present, in spite of the brief time difference from the baseline period. We can also look at the (semi-)amplitude for UAH data covering the globe, NH extratropics, and SH extratropics:

For the last few years, the response in the SH extratropics is nearly as large as in the NH extratropics, contrary to expectation. Also, the size of the effect is surprisingly large — again, contrary to expectataion — given that we’re less than a decade away from the baseline period.

More surprising still is the fact that the cycle is clearly present in the tropics:

It’s not only clearly present, it’s stronger in the tropics than in the NH or SH extratropics — extremely contrary to expectation, and contradictory to the hypothesis that it’s due to greater winter warming causing a change in the annual cycle.

We also find that for UAH data, in spite of the sizeable land mass in the NH the effect is considerably stronger over the oceans than over land:

Probably the “nail in the coffin” for the “greater-winter-warming” theory is the fact that the phase is roughly the same for both hemispheres’ extratropics, and for the globe as a whole, all peaking near the start of the year:

I find that the annual cycle shown in recent UAH TLT data is implausibly large, is implausibly very strong in the tropics, is implausibly larger over NH ocean than land, and is implausibly of roughly the same phase in both hemispheres. My conclusion is that the hypothesis that this cycle represents a real physical change in the annual cycle of temperature variations due to enhanced winter warming, is untenable.

My final conclusion from the previous post stands: there’s something wrong with UAH TLT data.

Categories: Global Warming

11 responses so far ↓

  • Deepclimate // October 30, 2008 at 6:43 am

    There is definitely something wrong with the UAH data sets.

    The recent wide yearly amplitude in UAH anomalies leads to some very strange results when looking at trends over the entire 30-year period.

    For example, if one compares the trend in the tropics TLT anomalies for December vs June, one gets an increasing linear trend for December (about 0.12 deg C per decade) , but a slightly *decreasing* trend for June.

    http://deepclimate.files.wordpress.com/2008/10/uah-tlt-tropic.pdf

    This wide divergence in trends simply makes no sense, and as you say, can not reflect the real physical changes.

  • Uli // October 30, 2008 at 7:56 am

    Hallo Tamino,

    clearly the UAH data are strange,
    but I have two questions.
    First, the global GISS-data seems not to fulfill a condition like
    GISS_all=0,29*GISS_land+0,71*GISS_ocean
    Why?
    Second, the retreat of sea ice may change the annual cycle for oceans, perhaps much more then the land annual cycle. Does the data show this?

    [Response: I don't have access to "GISS ocean" data, just land-only and land+ocean. Do you know where to get "GISS ocean"?]

  • Uli // October 30, 2008 at 12:20 pm

    Hallo Tamino,

    I used the global mean time series from
    http://data.giss.nasa.gov/gistemp/time_series.html
    You can choose: ‘SST: ocean data only’.
    But this seems to apply also to other data like
    http://data.giss.nasa.gov/gistemp/maps/
    You can choose land and ocean seperatly and download the data.

    [Response: Thanks for the link. I do notice that the data provided there are 12-month running averages.]

  • Ray Ladbury // October 30, 2008 at 12:36 pm

    Hmm! Curioser and curioser. It would appear that the UAH dataset seems to have significant systematic biases and ALL of them seem to be biased to camouflage a greenhouse warming signal.
    Part of the problem here is that joining up datasets across satellites is a very tough problem. I suppose there could be a bias in any one of the individual satellite datasets, but this is very odd.

  • dumskalle // October 30, 2008 at 12:50 pm

    From here we can access UAH’s amsu daily temperatures (the NOAA-15 satellite). From the LT data file <a href=’http://discover.itsc.uah.edu/amsutemps/data/amsu_daily_85N85S_chLT.r001.txt’)here we can compute the anomalies and make a simple plot like this

    Strange indeed. From 2004 we can see a pattern with high anomaly in march, low anomaly in may-june, high anomaly in october-november before a lower anomaly again around newyear.

    We can plot the absolute temperatures like this and see the same pattern.

    What is going on?

  • Uli // October 30, 2008 at 1:19 pm

    Hallo Tamino,

    you can change the mean period, for example to one month.

    [Response: I should have looked more closely.]

  • Gavin // October 30, 2008 at 3:36 pm

    Re: Uli’s comment.

    There isn’t such a straightforward equality because of the different areas that end up being used. The Met-station only index extrapolates over a lot of ocean, and so while the ocean data is not used over land in the land-ocean index, the true ‘land-only’ average is not the same as the met-station-only derived index. You could calculate this though as is done here for the annual averages.

    [Response: Thanks! I was definitely under the wrong impression.]

  • cce // October 30, 2008 at 4:32 pm

    The base period for RSS (and probably UAH) is 1979 to 1998, not 1979 to 1999.

  • Uli // October 30, 2008 at 5:43 pm

    Hallo Gavin,

    thanks. This answers the first question.

    Not only the Met-station only but also the ‘SST:ocean data only’ seems not equal to the data from fig.A4.

  • Phil. // November 6, 2008 at 3:35 pm

    RSS now uses an improved algorithm for matching between satellites (3.2) you might want to redo the analysis with the new product. The paper documenting it is at:
    http://www.remss.com/data/msu/support/Mears_and_Wentz_TLT_submitted.pdf

  • Eric Swanson // November 12, 2008 at 12:35 am

    Tamino,

    All your discussion reminds me of work I did several years ago. I compared the annual cycle of the UAH_LT product with sonde data collected around the Antarctic and noticed a discrepancy. At the time, I suggested that the cause of the difference might be the surface influence on the UAH_TLT. See:

    Swanson, R. E., Evidence of possible sea-ice influence on Microwave Sounding Unit tropospheric temperature trends in polar regions, Geophys. Res. Lett., 30(20), 2040,
    doi:10.1029/2003GL017938, 2003.

    The RSS analysis excludes areas of high elevation, such as over the Antarctic poleward of 70S. Also, they exclude data over the Himalayas and the Andes. How this difference might be reflected in the annual cycles you analyzed, I can not say.

    Christy and Spencer have repeatedly made claims as to the accuracy of their product, claims which have been proven wrong many times over. Douglass and Christy have a new paper to appear in Energy & Environment, in which they again claim that their analysis is better than the RSS product. As I understand it, Christy et al. are now splicing in the AMSU data with the previous time series of data from the MSU instruments. They simulate the MSU measurement to accomplish this result and the ground views are a bit different, as I recall. The earlier MSU algorithm combined the data from some of the scan positions in a single swath into one data value per swath located at nadir, whereas their AMSU algorithm splits the swath into halves, which produce 2 data values per swath. this could easily give different ground registration for the AMSU as compared with the nadir point for the swath used for the MSU data values. Perhaps another source of error??

Leave a Comment