In a recent post, I examined seasonal divergence in troposphere temperature (LT) trends produced by teams at RSS (Remote Sensing Systems) and UAH (University of Alabama at Huntsville). The strong annual cycle in the recent UAH data set has led to a wide divergence of temperature trends depending on time of year.
Since then, the blogosphere has been atwitter over the marked divergence between UAH and RSS for the month of February, which showed the UAH estimated anomaly (0.35) more than a full tenth of a degree higher than RSS (0.23).
Update (April 9, 2008): The March LT global temperature anomalies are out and they show both RSS (0.17) and UAH (0.21) down from February. The divergence has narrowed from 0.12 deg C in February to 0.04 deg C in March, which is exactly the divergence seen for that month in 2004-2008 (see below). So far in 2009, the UAH annual cycle is alive and well.
However, a detailed look at the divergence month-by-month shows that this latest discrepancy is not so surprising. I’ll also take a look at the effect on UAH of the recent switch to the newer AQUA satellite, which has actually resulted in an enhancement of the UAH annual cycle. AQUA has a self-correcting propulsion system, so the UAH annual cycle, and the cyclic component of UAH-RSS divergence, can not be the result of differing methods of correction for diurnal drift resulting from orbital decay.
I start by showing the global temperature trends for each month for the 1979-2008 period for both LT and surface temperature sets.
The three surface trends (HadCrut, NOAA and NASA) seem fairly well correlated, although with varying “bumps” toward the end of the year. And RSS is reasonably close the surface trends, albeit with a slightly different annual pattern.
But the UAH monthly trend shows a smooth, broad annual fluctuation through the year. The February trend at 0.18 deg/decade is above all the other trends, while the May trough at only 0.07 deg/decade is only half that month’s trend in the other data sets.
The difference between UAH and RSS global LT average monthly anomalies for two recent periods since the baseline period (which ended in 1998) shows clearly the annual cycle that results in this extreme UAH trend divergence.
Here we can see that the UAH February peak has diverged more and more over recent years; in 2004-2008 it was an average of 0.o8 degrees higher than RSS. Even more striking is the May divergence, which reached an average of 0.11 degree in the same five-year period.
The trend divergence is even wider for the tropics.
As noted previously, the UAH tropical trend is lower in all months. And there does seem to be a modest seasonal variation in RSS that is not present in the surface data (NOAA).
But the UAH divergence is extreme to say the least, ranging from a high of 0.12 deg. per decade in February to -0.03 deg per decade in June. And, as in the global case, this fluctuation is far from random, but is clearly led by the recent annual cycle.
Recent developments have shed some light on the possible source of the UAH annual cycle.
As I noted in a comment at my previous post, UAH switched last year from NOAA-15 to the AQUA satellite and is apparently using its data exclusively. The change was noted by one of the UAH team leaders, Dr. Roy Spencer:
The biggest adjustment is the fact that we don’t even use NOAA-15 right now…we are using the AMSU data from NASA’s Aqua satellite in the final UAH product.
The extent of the overlap period between the two satellites is not clear from the documentation, but the above statement does seem imply that most of the period from late 2002 on (when the AQUA was deployed) is based on the AQUA AMSU alone. [Update April 9, 2009: The documentation does refer to an overlap period with both satellites being used up to January, 2008. However, there is no clear statement about the overlap period in the current data set, or from which date the current data set reflects AQUA data only].
The UAH team considers this satellite more stable, as it is not subject to “diurnal tempearture drifts”, as noted above.
Spencer has also speculated on the February, 2008 UAH-RSS divergence at Watt Up With That:
“I believe it has to do with the differences in how diurnal variation is tracked and adjusted for. … For that reason, UAH has been using data from the AQUA satellite MSU, and RSS to my knowledge does not, and makes an adjustment to account for it. I believe our data [UAH] is probably closer to the true anomaly temperature, and if I’m right, we’ll see the two datasets converge again when the diurnal variations are minimized.”
In other words, RSS has to make a correction for diurnal drift, while UAH does not. But a comparison of the UAH pre-AQUA LT data set to 2007 with the current data set for the same period shows that the UAH seasonal divergence is stronger now than it was before the change!
I have obtained a copy of the UAH LT data set as of November, 2007, before the merging of AQUA data ( many thanks to cce for sending the data). This archived data set permits comparison of the recent UAH 5.2 data to the pre-AQUA version (also dubbed 5.2) over the 2000-2007 period.
For some reason, there are differences even before 2002, suggesting other processing changes have taken place between 2007 and present. In any event, early year peaks over and above the pre-AQUA data set can clearly be seen in every year from 2001 through 2006.
The result is an increase in trend divergence; the newer UAH data has a stronger peak in February, and a lower trough from May through October.
Since the pre-AQUA LT data was subject to adjustment for diurnal drift, while the current data is not, at least for recent years, the discrepancies can not be due to diurnal drift adjustment in UAH (or RSS for that matter).
Finally here’s a quick look at the T2 (or “mid-troposphere”) from UAH and RSS. (Essentially, these are based on the same raw data as the LT data sets, but have not been adjusted to remove cooling stratospheric influence, and therefore tend have significantly lower trends).
There is an annual cycle in the discrepancy between UAH and RSS in recent years, but it’s less severe than in the LT case (about 0.04 deg peak to trough, as opposed to almost 0.2 deg in the LT set).
The resulting fluctuations in monthly trends can be seen below.
Here both data sets seem to have a moderate annual (or semi-annual) cycle, but there is a much lower peak to trough discrepancy compared to UAH LT. The UAH-RSS discrepancy is still apparent but varies less through the year.
To summarize, there is a severe annual cycle in the UAH LT data set that results in a noticeable divergence in both the global and tropical monthly temperature trends over the 1979-2008 period. This annual cycle and resulting divergence can not be explained by diurnal drift adjustments in the UAH or RSS data sets. Rather, it appears that a moderate annual cycle (of hitherto unknown origin) in T2 channel data has been exaggerated by the UAH LT extraction process.
Why don’t you e-mail Dr. Christy and ask for his insights on your post. Specifically, I would ask him to comment on the general trend difference with RSS, the data adjustments AQUA / pre-AQUA prior to 2002, and the annual cycle.
[DC: I do intend to get in touch with John Christy of UAH (and Carl Mears of RSS). However, I’m waiting until I’m done one way or another with a third post idea that may or may not bear fruit.
I’m not sure there’s much point asking for comments about the general discrepancy between UAH and RSS trends, as this has been known for some time and appears to be largely related to inter-satellite calibration issues. But the remaining questions certainly are pertinent.
By the way, it’s a safe bet that both Christy and Mears are aware of the annual cycle issue, although neither has commented publicly on it to my knowledge. Still, it wouldn’t surprise me if one or both teams were working on the issue. As I’ve said before, I hope someone is. ]
Detrending the RSS and UAH data and plotting the differences would be an interesting post. It would help to visually distinguish the effect of the annual cycles from the trend divergence (which is a separate issue likely due to inter sat calibration like you say). Maybe you are planning something like this in your next post?
The tropical pattern as well as surf tracking more with RSS, makes it more liley that UAH is the one off, not RSS.
Of interest both sattelites seem to have more seasonal difference than the surf. And that difference is in sam months pattern. So, even though UAH seems to have the disease worse, RSS may have a bit of a cold as well…
Detrending by subtracting the least squares line, will not affect the already substantial differences much as far as I can see. The annual half-cycle in UAH-RSS difference ranges from 0.1 to 0.2 deg. C, whereas the difference in trend is only 0.003 deg. C. per year.
TCO: There does seem to be more seasonal variation in the satellite observations than in the surface, but then again there’s greater uncertainty and variance in MSU. Further analysis is needed to determine the source and implications of the moderate annual cycle in the MSU data (for example, by analyzing various zones in T2, or looking for cycles in the T4 stratosphere data set).
DC: It’s just an insight that might help lead to finding the cause. Obviously UAH and RSS have a lot of similarities, but some differences. You could (I guess) list the differences and play with them to see what gives the different result. But maybe it is a feature of both products but one where the knob is turned higher on UAH. I think maybe it needs to be fixed on both, though.
A moderate annual cycle in recent years is apparently in the raw data. This could be random variation of the atmosphere over a short period, or random measurement error, or some combination. The only conclusion I would draw is that MSU data has greater uncertainty and volatility and so is not as reliable as the surface observations.
But when it comes to the LT analyses, UAH is clearly unrealistic, while RSS is more in line with surface observations. The two LT extraction processes are completely different, so it’s not a question of adjusting “knobs.” The UAH LT extraction is clearly broken. You should also be aware that the UAH LT methodology has been criticized on other grounds (see IPCC chapter 184.108.40.206.2, p. 268).
From my previous work, Dr Christy explained to me that the anomalies for UAH are based on an annual cycle of 1979-1998.
This would explain the expanded annual signal in recent years.
Re: Jeff’s comment ,
In general, the more recent the baseline period the smaller the annual cycles should be (as is there is less time for divergence to develop).
And, of course, the selection of the baseline period has no effect on the actual monthly trends, only on the anomalies themselves.
“The only conclusion I would draw is that MSU data has greater uncertainty and volatility and so is not as reliable as the surface observations.”
Not trying to be confrontational, but are you saying this is sufficient to establish which metric has greater uncertainty and which metric is more reliable?
[DC: The uncertainties for global and tropical data sets were calculated in IPCC AR4 220.127.116.11; the error bars (5-95% confidence limits) for the global surface trends are about half those of the two LT data sets (surface are about 0.17 deg C per decade, +-0.04 deg C).
The trends for the MSU data sets also vary more than the surface data sets for the same periods; perhaps I’ll quantify this in a future post.]
If you consider that the annual variation is subtracted based on a pre 1998 basis, any change in the atmospheric response to annualized forcings will result in an increased annual signal post 1998. If the anomaly were calculated for the whole trend, the more recent annual signal will be dampened.
I find this interesting because it seems that there may be a potential to measure the change in atmospheric response to solar forcing and an actual response to increased CO2/CH4.
Long term feedbacks (positive and negative) would be left out of course but it is interesting none the less.
I also spent some time looking at phase evolution of the annual signal. You might consider looking at that.
It’s too bad the data has steps in it.
[DC: Yes, a longer baseline would “dampen” the annual cycle in recent years, but not completely. It would have no effect on the monthly trends though (since all anomalies for a given month would be adjusted by the same amount). But RSS has the same baseline as UAH (1979-1998) and yet does not exhibit anywhere near the same annual cycle in the LT data set, and neither do the surface data sets.
There is data without steps of course – the daily gridded data sets. I don’t think I’ll get around to that any time soon though.]
I didn’t realize that RSS is using the same baseline.
[DC: Yep, that’s why UAH and RSS can be compared directly. The surface sets have earlier baselines, differing from each other; for some kinds of analysis they need to be converted to a common baseline.]
The gridded data contains steps due to transitions between satellites. Dr. Christy has written some papers on the subject. I found that by correcting one step in the middle of the LT series corresponding to a difficult satellite transition. RSS and UAH were almost exact matches. I used surface station data to provide the correction by recalibrating trend and applying some filtering. The math is a bit sensitive to windows chosen but from my results RSS had a positive step in the middle and UAH required less correction according to ground data. This doesn’t of coursev doesn’t have much to do with the annual signal.
[DC: The step difference has been discussed before, and as you say appears to be the major difference between RSS and UAH. I think it’s likely that the correct calibration is somewhere in between, but it may well be closer to one than the other. By the way, have you checked this lately? I have the impression that the step difference may be less right now in the current versions of RSS (3.2) and UAH (5.2), than it was even last year. It’s also interesting to note that overall global trends in the T2 data sets are more divergent, but there is no apparent step in 1992. Yet the LT and T2 are based on the same raw data. Curiouser and curiouser – I think I will need to do a post on this too some time too.]
It’s an interesting post, I’ll stop back when the next in the series is finished.
[DC: Thanks for coming by … to be continued.]
DC: I’m not trying to make a strong point, just something to think about. I agree that UAH shows more deviation from land temps than RSS. But both have similar patterns. Instead of purely thinking RSS good, UAH bad; maybe if you look for a knob that RSS turned up a little and UAH turned up a lot, you will find the feature that causes this issue in both of them.
It’s just a thought starter…
Pingback: saved post to RC « the Air Vent
Pingback: UAH annual cycle continues in 2009 « Deep Climate
Pingback: A first look at UAH 5.3 « Deep Climate