I recently showed a couple of breakpoint comparisons for satellite data: RSS versus NOAA and RSS versus UAH. Today, I’ll show a similar comparison for UAH versus NOAA, again stratifying by Land, Ocean and All. (Unfortunately, I was unable to extract a satellite comparison for other major food group: the CRU_TAR (airport tarmac).
Again, the breakpoints are calculated using strucchange and line segments are fitted between breakpoints.
Figure 1. UAH minus NOAA, by firma.
At this point, I’m simply plotting results using a relevant algorithm as an exploratory analysis and am not asserting that any of the breakpoints are “Significant”. Having said that, the breakpoints (which have CIs attached to them) plausibly relate to the (first) termination of NOAA-6 (1983-11), NOAA-10 (1988-10), NOAA-12 (1998-12) and the introduction of AMSU units (2004-12). Here is one more interesting graphic from Christy et al 2000 showing the differences between instrument temperature for different satellites up to 2000 – something that has to be adjusted for. Could errors of 0.2 deg C arise in this standardization? Seems possible to me. Could three errors all be of the same sign? Sure, they could.
The breakpoint structure point to an interesting statistical issue in respect to trend estimation, which may have important implications for the ongoing issue of satellites versus observation. Take a look at the bottom panel (Land). The intra-segment trends are all flat or slightly positive, while the trend without breakpoints is negative – this negative trend is the T2LT warming less than surface (NOAA) warming inconsistent with model expectations.
However, suppose that the breakpoints analysis has identified real breakpoints at important satellite transitions. Is it possible that three inter-satellite adjustments were each off by 0.2 deg C or so (and all in the same direction)? I don’t know enough about the inter-satellite adjustment procedures to comment on this, but my quick perusal of the literature hardly leads me to exclude that possibility.
Let’s also suppose that there were three breakpoints of the sort shown and that otherwise the fluctuations around the trend are white-to-low order red noise. If the breakpoints are not recognized, this will ARIMA-model out as quite high order red noise.
Also if there were breakpoints of the form shown, note that the “true” trend is the composite of the intrasegment trends, which is positive i.e. troposphere warming slightly more than the surface in line with models. So quite a lot turns on breakpoint analysis – an issue that I do not recall being addressed in the CCSP or IPCC reports on this topic. (FWIW, I’ve reflected a bit on this matter, about which much has been made in “skeptical” literature and will try to re-visit it some time. After some reflection, the idea that the tropical troposphere would warm more than the surface doesn’t seem to me to be a particularly unreasonable model property. On the other hand, surface histories of tropical tarmac may have flaws of their own. As also histories of SST with complicated transitions from buckets to engine inlets to buoys.)
Climate Audit 
19 Comments
Note that many of the break points also correspond to known climate events, which suggests they may be artifacts.
Right after the 1998 El Nino, After El Chicon and the concurrent El Nino. A La Nina in the late 80’s.
The 2005 breakpoint is interesting, and doesn’t appear to result from anything like the others.
Re: Andrew (#1), I’ve noticed that the lower satellite data has numerous saw tooth events, with sudden drops in temperature followed by gradual warming. A natural explanation for this behavior is very plausible. GISS seems to show this behavior much more than Hadley CRUT3. If you squint (or have some imagination), the upper satellite data shows drops in temperature corresponding to volcanic eruptions in which the subsequent warming is extremely slow.
Re: Jason Lewis (#5), Just to clarify what I was getting at with my previous post:
If the upper atmosphere has an extremely slow recovery from drops in temperature, and the lower troposphere (from satellite data) has a moderately slow recovery period, then it’s plausible that the surface has a very short recovery period. If so, then this would be a natural explanation for the jumps in the graphs above (MSU-surface) as John Christy suggests.
Thanks for the post. One expects the most problems over oceans and in the tropics because of lack of in situ data (radiosondes and surface measurements) to help with the ill posed inversion of the integral equation (exactly as you have now noticed). And in the presence of clouds, the inversion process is even more dicey. Note that exactly where satellites are supposed to be most helful (in areas like the southern hemisphere where there is very little in situ data), they have the most problems. Strange, yeah?
Steve:
My 2007 paper deals specifically with the tropics. Note that radiosondes also show the lack of coherence with the surface – due to physical processes. The surface and the troposphere are somewhat coupled, but respond differently to forcings, so you would expect difference time series as you show to display these type of differences – due to physical properties of the two media.
Another point – at a transition from one space craft to another, there is also a base satellite operating, so the time series is not a collection of disjoint time series – there is tremendous overlap of continuing satellites. When apples to apples comparisons are shown (i.e. satellite vs. balloons) one does not see the discontinuities implied by the image above. The surface and troposphere are two different animals.
John C.
Re: J Christy (#3),
Dr. Christy,
One question – By overlap, are you saying that the series does not make a sudden calibrated transition to a new satellite but rather an averaged smooth transition occurs or are you saying the overlapping data is used to calibrate the satellites well and a sudden transition is made??
The overlapped calibration time period and sudden transition is used in sea ice data which seems to have very good quality transitions.
Re: J Christy (#3),
John, thanks as always for your comments. The observation that
is obviously something that you’ve discussed in many articles. This is a point of view that seems at odds with the tight coupling of surface and troposphere in the models and is obviously a very big and important question.
I’ll plot up some HadAT series for comparison and clip out a plot from your 2007 article for comparison as well.
Re: Steve McIntyre (#11),
Steve, you might also try plotting comparisons with the GISS data. The GISS data seems to track the satellite date better than Hadley CRUT3. See:
http://www.junkscience.com/MSU_Temps/MSUvsGISTEMP.html
http://www.junkscience.com/MSU_Temps/MSUvsHadley.html
Steve,
Could you please post R the code for this thread. I am using strucchange for similar reasons and am new to it so your code would be helpful in understanding it.
The limitation of all “breakpoint” analyses is the applicabilty of their underlying assumptions about the mathematical structure of the “unbroken” series to the record at hand. There are many real-world processes that produce remarkably abrupt changes in the apparent regressional slope of the data over irregular intervals of time. (Commodity prices and ocean current vectors come immediately to mind.) Thus the “breaks” do not necessarily indicate any flaw in measurements. Linear regressional methods have never been proper time-series analysis tools and they cannot be relied upon to provide any definitive determinations in series with complex–and most often uknown–structures. Fundamental physical considerations must be invoked in searching for measurement irregularities.
Re: John S. (#8),
I agree with comments about limitations of breakpoints methods – we’ve discussed this from time to time, mostly in connection with USHCN breakpoints methods and GISS.
It seemed interesting to experiment with these methods with the satellite data.
As one plots the results, one sees that the breakpoints methods are definitely attracted to big El Nino’s, which also more or less coincide with satellite changeover in some cases.
Re: Steve McIntyre (#10),
GHCN global series may provide even more-revealing examples of breakpoints, because of the station-shuffling involved at various time-junctures of the index construction. USHCN has enough century-long records to mute this effect.
As another way of looking at this I suggest that you also plot the actual anomalies from a common base period aganinst each other (as opposed to a difference plot). I have done this for both hemispheres and for the globe for both land and ocean areas. As it turns out, the tropospheric data has larger up and down “jags” as might be expected because of the lower thermal inertia of the troposphere as compared to the ocean surface. Many of the “discontinuities” seem to be where the “jags” cross over. More interesting, the UAH MSU data over the oceans has an almost identical long-term (1979-2008) trend as does the ocean surface data (0.107 degrees per decade MSU vs. 0.109 degrees per decade NOAA surface). Over land, the surface data has a considerably greater trend than does the tropospheric data (0.293 degrees per decade surface vs. 0.162 degrees per decade MSU). This implies that the land surface data is probably contaminated by UHI and/or other surface measurement problems, as most of us suspect, increasing the observed warming considerably over the real value. If we then take the UAH MSU data as the most accurate observation of the global temperature trend since 1979 and correct for volcanic effects to get a “no-volcano” natural trend it comes out to be about 0.1 degrees per decade vs. the IPCC “consensus” value of 0.2 degrees per decade, based on their climate models…. a very real and observable disconnect.
This is already a fascinating thread. Hopefully we’ve hardly started. Thanks for all the insightful contributions.
To #6
In general there are two satellites operating at any given time, so for each day we have two values to average. When a replacement is launched for the older, the newer continues to operate as now the new older satellite. E.g. NOAA-10 (0730/1930 orbit crossing) and NOAA-11 (0130/1300 orbit crossing time) operated together with NOAA-10 the first to go up. When NOAA-12 was launched to replace NOAA-10, NOAA-11 was operating across both – hence we have a backbone to carry the data forward without a discontinuity. It’s a bit more convoluted than that (which leads to the RSS shift in 1992) but that’s the basic idea. If you look at the hot-target temperatures shown in Steve’s post you will get an idea of how this worked.
John C.
#15 John C.,
What are the overlaps for the AMSUs in UAH? I’m not too clear on this. AFAIK, data from NOAA-15 AMSU starts around 1998, and AQUA late 2003 or early 2004, but I’m unclear on the details. What were the overlaps involved in these two transitions?
In the case of RSS, I suppose it’s NOAA-15 AMSU alone for the last several years.
Steve,
As implied above, I think the first AMSU (NOAA-15) switchover corresponds to circa 1998 “breakpoint”, while the more recent switch to AQUA AMSU would be circa 2004. I imagine John C. will set us straight as to the exact details.
Re: Deep Climate (#16),
For AQUA, ice data starts in mid 2002 so I presume the other measures started then too. The UAH switchover to AQUA was probably later as cross calibration would be necessary.
Re: Jim (#17),
Those circulation cells arise in the models as well. The relative efficiency of heat transfer is a question, though, as for example IIRC, Jerry Browning has stated the jet streams in the models bear little relation in position and velocity distribution to the real world.
Steve McIntyre: “This is a point of view that seems at odds with the tight coupling of surface and troposphere in the models and is obviously a very big and important question.”
If ENSO heating get carried to the Arctic via Hadley/mid-latitue/polar cells, wouldn’t that imply a generality that heat in the LT is rather efficiently carried away and would account for the lack of the “hot spot?” It would also account for polar melting, would it not? (I am just trying to learn more about climate – I’m not even a talented amateur.)
This is OT, but related to AMSU data and, I suspect, to tropical tropospheric behavior.
Here’s a current plot of unadjusted Channel 5 (“600 mb”) global tropospheric temperature anomaly data for the past eighteen months, available from here
There is current blogosphere interest in this because the July, 2009 global anomaly peak is happening at the same time as the annual (boreal summer) peak, creating a “record” absolute temperature.
I added thin green lines to indicate where the typical peaks and valleys align. I added a red dot where I think the anomaly will be in two or three weeks, if past patterns hold.
The plot indicates that the global troposphere experiences some kind of “heat kick” on a roughly regualr basis. The heat then dissipates in rather short order, returning the troposphere to its more-or-less initial state.
My guess is that this, if it is real, is a function of tropical thunderstorm activity, probably Indo-Pacific in origin. Or, maybe somehow it relates to occasional bursts of energy moving from the tropics to the extratropics. Or, maybe it is an illusion created by the measurement system or by my layman’s handling of the data.
If it is real then I’d sure like to understand the physical behavior that underlies the pattern. Opinions welcomed.