date: Fri Feb 15 12:16:35 2008
from: Phil Jones
subject: Re: Additional calculations
to: santer1@llnl.gov
Ben,
Apologies for not getting back sooner. Doing these difference assessments is
very useful, particularly as you've got interesting results.
I'm attaching a paper I reviewed (Tom Smith/Dick Reynolds et al) of their
new version that has been accepted by J. Climate. My purpose is not
for you to get their latest version, which is newer than ERSST-V3, but to
point out that any differences you had from CCSP are not due to buoys. You'll
see from the paper that buoys could be important, but they aren't confident
enough yet about making adjustments for them. Buoys would only
affect SST after 2000 - if accepted they would raise SSTs. This could be
contentious when it gets into a paper.
A second point, I reckon HadSST2 is the latest HC SST dataset and not
HADISST1. HadCRUT3 is a merged version of HadSST2 and CRUTEM3.
Also attaching the HadSST2 paper from Nick Rayner in 2006.
You can download HadSST2 from the site [1]http://hadobs.metoffice.com/
[2]http://www.cru.uea.ac.uk/cru/data/temperature/ It is also on our site.
All a useful addition to the paper. It is the surface - LT differences that get talked
about.
Cheers
Phil
Cheers
Phil
At 02:34 14/02/2008, you wrote:
Dear folks,
Sorry about the delay in sending you the next version of our manuscript. I decided that
I needed to perform some additional calculations. I was concerned that we had not
addressed the issue of "differential warming" of the surface and troposphere - an issue
which Douglass et al. HAD considered.
Our work thus far shows that there are no fundamental inconsistencies between simulated
and observed temperature trends in individual tropospheric layers (T2 and T2LT). But we
had not performed our "paired trends" test for trends in the surface-minus-T2LT
difference time series. This is a much tougher test to pass: differencing strongly damps
the correlated variability in each "pair" of surface and T2LT time series. Because of
this noise reduction, the standard error of the linear trend in the difference series is
typically substantially smaller than the size of the standard error in an individual
surface or T2LT time series. This makes it easier to reject the null hypothesis of "no
significant difference between simulated and observed trends".
In the CCSP Report, the behavior of the trends in the surface-minus-T2LT difference
series led us to note that:
"Comparing trend differences between the surface and the troposphere exposes potential
discrepancies between models and observations in the tropics".
So it seemed wise to re-examine this "differential warming" issue. I felt that if we
ignored it, Douglass et al. would have grounds for criticizing our response.
I've now done the "paired trends" test with the trends in the surface-minus-T2LT
difference series. The results are quite interesting. They are at variance with the
above-quoted finding of the CCSP Report. The new results I will describe show that the
"potential discrepancies" in the tropics have largely been resolved.
Here's what I did. I used three different observational estimates of tropical SST
changes. These were from NOAA-ERSST-v2, NOAA-ERSST-v3, and HadISST1. It's my
understanding that NOAA-ERSST-v3 and HadISST1 are the most recent SST products of NCDC
and the Hadley Centre. I'm also using T2LT data from RSS v3.0 and UAH v5.2. Here are the
tropical (20N-20S) trends in these five datasets over the 252-month period from January
1979 to December 1999, together with their 1-sigma adjusted standard errors (in
brackets):
UAH v5.2 0.060 (+/-0.137)
RSS v3.0 0.166 (+/-0.130)
HADISST1 0.108 (+/-0.133)
NOAA-ERSST-v2 0.100 (+/-0.131)
NOAA-ERSST-v3 0.077 (+/-0.121)
(all trends in degrees C/decade).
The trends in the three SST datasets are (by definition) calculated from anomaly data
that have been spatially-averaged over tropical oceans. The trends in T2LT are
calculated from anomaly data that have been spatially averaged over land and ocean. It
is physically reasonable to do the differencing over different domains, since the
temperature field throughout the tropical troposphere is more or less on the moist
adiabatic lapse rate set by convection over the warmest waters.
These observational trend estimates are somewhat different from those available to us at
the time of the CCSP Report. This holds for both T2LT and SST. For T2LT, the RSS trend
used in the CCSP Report and in the Santer et al. (2005) Science paper was roughly 0.13
degrees C/decade. As you can see from the Table given above, it is now ca. 0.17 degrees
C/decade. Carl tells me that this change is largely due to a change in how he and Frank
adjust for inter-satellite biases. This adjustment now has a latitudinal dependence,
which it did not have previously.
The tropical SST trends used in the CCSP Report were estimated from earlier versions of
the Hadley Centre and NOAA SST data, and were of order 0.12 degrees C/decade. The values
estimated from more recent datasets are lower - and markedly lower in the case of
NOAA-ERSST-v3 (0.077 degrees C/decade). The reasons for this downward shift in the
estimated warming of tropical SSTs are unclear. As Carl pointed out in an email that he
sent me earlier today:
"One important difference is that post 1985, NOAA-ERSST-v3 directly ingests "bias
adjusted" SST data from AVHRR, a big change from v2,
which didn't use any satellite data (directly). AVHRR is strongly affected in the
tropics by the Pinatubo eruption in 1991. If the "bias adjustment" doesn't completely
account for this, the trends could be changed".
Another possibility is treatment of biases in the buoy data. It would be nice if Dick
Reynolds could advise us as to the most likely explanation for the different warming
rates inferred from NOAA-ERSST-v2 and v3.
Bottom line: The most recent estimates of tropical SST changes over 1979 to 1999 are
smaller than we reported in the CCSP Report, while the T2LT trend (at least in RSS) is
larger. The trend in the observed difference series, NOAA-ERSST-v3 Ts minus RSS T2LT, is
now -0.089 degrees C/decade, which is very good agreement with the multi-model ensemble
trend in the Ts minus T2LT difference series (-0.085 degrees C/decade). Ironically, if
Douglass et al. had applied their flawed "consistency test" to the multi-model ensemble
mean trend and the trend in the NOAA-ERSST-v3 Ts minus RSS T2LT difference series, they
would not have been able to conclude that models and observations are inconsistent!
Here are the observed trends in the tropical Ts minus T2LT difference series in the six
different pairs of Ts and T2LT datasets, together with the number of "Hits" (rejections
of the null hypothesis of no significant difference in trends) and the percentage
rejection rate (based on 49 tests in each case)
"Pair" Trend 1-sigma C.I. Hits Rej.Rate
HadISST1 Ts minus RSS T2LT -0.0577 (+/-0.0347) 1 (2.04%)
NOAA-ERSST-v2 Ts minus RSS T2LT -0.0660 (+/-0.0382) 1 (2.04%)
NOAA-ERSST-v3 Ts minus RSS T2LT -0.0890 (+/-0.0350) 0 (0.00%)
HadISST1 Ts minus UAH T2LT +0.0488 (+/-0.0371) 28 (57.14%)
NOAA-ERSST-v2 Ts minus UAH T2LT +0.0405 (+/-0.0403) 25 (51.02%)
NOAA-ERSST-v3 Ts minus UAH T2LT +0.0175 (+/-0.0370) 15 (30.60%)
Multi-model ensemble mean -0.0846
Things to note:
1) For all "pairs" involving RSS T2LT data, the multi-model ensemble mean trend is well
within even the 1-sigma statistical uncertainty of the observed trend.
2) For all "pairs" involving RSS T2LT data, there are very few statistically-significant
differences between the observed and model-simulated "differential warming" of the
tropical surface and lower troposphere.
3) For all "pairs" involving UAH T2LT data, there are statistically-significant
differences between the observed and model-simulated "differential warming" of the
tropical surface and lower troposphere. Even in these cases, however, rejection of the
null hypothesis is not universal: rejection rates range from 30% to 57%. Clearly, not
all models are inconsistent with the observational estimate of "differential warming"
inferred from UAH data.
These results contradict the "model inconsistent with data" claims of Douglass et al.
The attached Figure is analogous to the Figure we currently show in the paper for T2LT
trends. Now, however, results are for trends in the surface-minus-T2LT difference
series. Rather than showing all six "pairs" of observational results in the top panel,
I've chosen to show two pairs only in order to avoid unnecessarily complicating the
Figure. I propose, however, that we provide results from all six pairs in a Table.
As is visually obvious from the Figure, trends in 46 of the 49 simulated
surface-minus-T2LT difference series pairs are within the 2-sigma confidence intervals
of the NOAA-ERSST-v3 Ts minus RSS T2LT trend (the light grey bar). And as is obvious
from Panel B, even the Douglass et al. "sigma{SE}" encompasses the difference series
trend from the NOAA-ERSST-v3 Ts/RSS T2LT pair.
I think we should show these results in our paper.
The bottom line: Use of newer T2LT datasets (RSS) and Ts datasets (NOAA-ERSST-v3,
HADISST1) largely removes the discrepancy between tropical surface and tropospheric
warming rates. We need to explain why the observational estimates of tropical SST
changes are now smaller than they were at the time of the CCSP Report. We will need some
help from Dick Reynolds with this.
With best regards,
Ben
----------------------------------------------------------------------------
Benjamin D. Santer
Program for Climate Model Diagnosis and Intercomparison
Lawrence Livermore National Laboratory
P.O. Box 808, Mail Stop L-103
Livermore, CA 94550, U.S.A.
Tel: (925) 422-2486
FAX: (925) 422-7675
email: santer1@llnl.gov
----------------------------------------------------------------------------
Prof. Phil Jones
Climatic Research Unit Telephone +44 (0) 1603 592090
School of Environmental Sciences Fax +44 (0) 1603 507784
University of East Anglia
Norwich Email p.jones@uea.ac.uk
NR4 7TJ
UK
----------------------------------------------------------------------------