There was an interesting post on Climate Audit about a paper covering trends in atmospheric humidity. The substance of the post was discussing why a paper by Gareth Paltridge and others was rejected by the journal of climate. But I was quite interested to learn that there was long term trend data on humidity.
As most people who follow the issue understand a doubling of CO2 by itself should increase atmospheric temperature about 1 degree C at equilibrium. The big question is what feedback is caused by this temperature increase. The largest feedback is caused by an increase in water vapor as the temperature increases. The theory is that relative humidity (r) should remain constant, which would mean that as temperature increases specific humidity (q) would increase. This means that the total water vapor of the atmosphere would increase causing an increased forcing.
Over time I have seen a few papers which have confirmed that relative humidity is indeed staying constant. But these papers have had a lot of caveats and have covered limited areas and time frames. So I was interested in the paper by Dr. Paltridge which used a re-analysis data set called NCEP covering a 35 year period from 1973 to 2007.
I don't have access to the paper but the summary is that in this data set r decreased over the period in some relevant regions. This would be counter to theory, and interesting.
Ryan Maue, who is a student at Florida State made several comments and here on CA pointing out issues with the Paltridge paper. Essentially the objection is that prior to 1979 the data is based on a small radiosonde network, and that the subsequent data is based on a combination of satellites and radiosondes. He felt that at the minimum the results should be compared to other re-analysis data sets. He chose ECMWF as the best. He even commented that he would take a look himself.
So I thought I would go ahead and see if I could figure out how to download the ECMWF data and do a quick analysis. It turns out to be quite possible.
The ECMWF data covers parts of 46 years, but only 44 complete years from 1958 to 2001. So I downloaded the r and q data for all grid locations for those years. The results have something for everyone I suppose.
For completeness I started out looking at the trends for the entire period and the entire globe. In this case q is only negative at the very highest altitudes above 100hPa. Below 500hPa q is positive. R on the other hand is negative above 400hPa, and positive below 925hPa with the altitudes in between not being significant. Again according to theory the theory the trend in r is supposed to be zero and the trend in q is supposed to be positive.
But I get the impression that the global figure over this time period is not the most interesting. As Mr. Maue points out most of the radiosondes are in a small band in the Northern Hemisphere. So the trends that cover that area are called out both by Dr. Paltridge and By Mr. Maue in a subsequent post.
Looking at the NH results for the entire period q is significantly negative all the way down to 700mb. It only becomes significantly positive at 925hPa. In a result I don't completely understand r is negative above 400hPa and insignificant below that. I would have thought that in a warming atmosphere that if q was negative r would have to be negative as well. In any event the negative q trend over the NH which is where the majority of the real measurements would have been made in this time period seems to be different than theory and in line with the results from NCEP. Note that I tried two definitions of the Northern mid latitudes with no change in results.
In the SH the trends are largely positive for both q and insignificant for r which would be in line with theory. And this is true for the entire mid latitude and tropic region, which has negative r only for the altitudes above 400hPa.
In summary then over the 44 year period the area that shows negative q seems to be the mid latitudes of the Northern Hemisphere. Since the areas outside of the measurement regions are computed using climate models I'm not sure of the relationship of the "real" data to areas where there were no radiosondes.
I took a separate look at the "post satellite" period. Unfortunately this is a very short time in this data set since it ends in 2001 unlike the NCEP data which goes through 2007. The trends were not particularly significant over this period.
The R code for this analysis can be found here.
The data for this post is from ERA-40 and was graciously supplied by the ECMWF data server.
Am I getting this right: the hemisphere showing the most warming (that's the Northern) is also the one which is showing the largest divergence to climate theory ?
ReplyDeleteIt also seems strange that, again in the Northern hemisphere figures, there does seem to be a trend in r as well as q, in the same direction at most altitudes.
The best data does not seem to support particularly well some rather critical assumptions in the models.
You need to be very careful using reanalysis, and using humidity data even more so. Be aware that the reanalysis quality in general is significantly worse pre the satellite era (1979) which is why the first reanalysis started then. Just running the model output through a stats package is meaningless, sorry.
ReplyDeleteHi Belette,
ReplyDeleteYes I am aware of the issues they were covered in the CA post that I linked to.
I don't know what you mean about the "first" reanalysis.
On the other hand why are these people spending all this time and money on doing a reanalysis if it doesn't mean anything?
Anyway I'm just reporting on the results of analyzing the data. You can take it any way you want.
> Yes I am aware of the issues they were covered in the CA post that I linked to.
ReplyDeleteBut you show no awareness of them in *this* post. I'm not going to trawl CA to find out what caveats you know and don't! And you cannot reasonably expect me to do so.
> I don't know what you mean about the "first" reanalysis.
You're using ERA-40 (or possibly ERA-interim, if you say which I missed it) (and the corresponding NCEP/NCAR version). There was an earlier ERA-15. See http://ecmwf.int/research/era/do/get/Reanalysis_ECMWF. Using data without even knowing what you are using is not good (again, sorry).
> On the other hand why are these people spending all this time and money on doing a reanalysis if it doesn't mean anything?
Of course the reanalysis means something. Come on, away with the straw men! But different variables have different degrees of reliability. Surface pressure is well observed, for example, and liable to be reliable early on. Upper atmosphere humidity is very poorly observed early on, and liable to be extremely unreliable then. See http://ecmwf.int/research/era/ERA-40/Data_Services/section3.html#ha for example for some stuff, I'm sure there is more.
> Anyway I'm just reporting on the results of analyzing the data. You can take it any way you want.
No, not good enough - or at least, not if you want to be taken seriously.
Belette,
ReplyDeleteI actually included several links to the original posts which discuss these issues, thought it would be repetitive to include them here.
Ryan Maue who works in this field indicated that the ECMWF data was the best to work with as per one of the linked posts. The purpose of this post was to compare the results to the Paltridge paper which used the NCEP data.
I'll think about updating the post to try to make this background clearer.
BTW, obviously I'm using ERA-40 because of the time frame it covers.
ReplyDeleteWell, as a person directly involved in the actual measurements my remark is that this analysis is a wasted effort. Humidity profiles are of very poor quality so this is GIGO.
ReplyDeleteOver 44 years the sensor technology developed from chemically dried human hair or flattened and dried lamb intestines to rather more sophisticated constructions of thin films.
Correspondingly the mesurement accuracy improved significantly in terms of time constants and dependence on undesirable factors (such as temperature regimes during the observation, sensor soaking when passing through liquid clouds, sensor freezing, etc.) and also the rather complex computation algorithms evolution.
The primary user requirement did not include high absolute accuracy, either. Relative accuracy did help define the cloud base and top locally, and provided the main motivation. The weather prediction models deliver outline humidity fields as a by-product. This is fortunate, as the humidity field measurement is extremely difficult due to large variations in very small scale (both vertically and horizontally).
Sorry to say, but that is the situation. No amount of post-processing can deliver the accuracy of humidity measurement required by reliable climate analysis. You have to trust the rather indirect methods used by current climate analysis models as they are more reliable. If the model and the measurement disagree, in this case the model clearly wins.
The primary purpose of this post was to compare ECMWF to NCEP which I stated above.
ReplyDeleteHaving said that these are both climate reanalysis models, so I'm not sure which product you think would be preferable.
Finally as stated in the article they are based on radiosondes in the early periods, and satellites in the later periods. Were you involved in either?