http://scienceofdoom.com/2011/06/02/water-vapor-trends/
and
http://scienceofdoom.com/2011/06/05/water-vapor-trends-part-two/

Part Two includes commentary on Paltridge et al 2009 as well as Dessler & Davis 2010.

Radisonde analyses show increasing water vapor in the Northern Hemisphere. Two important papers on the subject demonstrated this and decided there was not enough quality data for the Southern Hemisphere (Ross & Elliot 2001, Durre et al 2009).
Satellites show increasing water vapor over the oceans (radiosonde data over the oceans is very sparse).
Other reanalyses show increasing water vapor, e.g. ERA40

NCEP/NCAR was demonstrated by Trenberth & Smith in 2005 to be worse than ERA40 – via the dry mass of the atmosphere.

Paltridge, Arking and Pook don’t demonstrate what is wrong with Trenberth & Smith’s analysis, or what is wrong with the radiosonde trend analyses. They don’t demonstrate what is wrong with the satellite data.

Trenberth, Fasullo & Smith have already shown NCEP/NCAR water vapor trends in their 2005 paper.

So what have Paltridge, Arking and Pook added to the sum of knowledge?

It may be, for example, the tree rings of the last several decades were actually OK and that they did not match the temp records because the temp records are not OK. We have a huge number of papers based on bad data. IIRC, the Met is now laboriously recreating the work that Phil Jones did and “lost”. That seems like a good start but the process has to be transparent if it is to inspire confidence.

The results are shown in the 2 graphs below and the following table. In summary there is much ARn autocorrelation as was the case with RSS SSM/I and the trend is flat and cannot be distinguished from 0. The trend slope was 0.0000128; the standard error was 0.0000931; the p = 0.89 and the AR1 correlation = 0.57. The R code is listed below.

library(ncdf)
ncep500=open.ncdf(“NCEP.nc”)
ncep500
[1] “file NCEP.nc has 4 dimensions:”
[1] “lat Size: 73″
[1] “level Size: 1″
[1] “lon Size: 144″
[1] “time Size: 361″
[1] “————————”
[1] “file NCEP.nc has 1 variables:”
[1] “short shum[lon,lat,level,time] Longname:Monthly Mean of Specific Humidity Missval:32766″
nc500=get.var.ncdf(ncep500)
dim(nc500)
[1] 144 73 361
nc25SN=nc500[,mean(c(27:47)),]
dim(nc25SN)
[1] 144 361
SN25=colMeans(nc25SN)
x=1:361
lmSN25=lm(SN25~x)
summary(lmSN25)
acf(residuals(lmSN25))$acf[2]
[1] 0.5715495
plot(x,SN25,type=”b”, main=”Specific Humidity 25S to 25N from 01/1979 to 01/2009″,xlab = “Months Starting at 01/1979″,ylab=”Secific Humidity”,col=”dark red”)

For instance, here are the changes in RH of GISS Model E from 1960 to 2003, by latitude and height.

http://data.giss.nasa.gov/work/modelEt/lat_height/work/tmp.3_E3Af8aeM20_1_0112_1960_2003_1951_1980_-L3AaeoM20D_lin/mean.gif

Here is the average global change by height over the same period.

http://data.giss.nasa.gov/work/modelEt/lat_height/work/tmp.3_E3Af8aeM20_1_0112_1960_2003_1951_1980_-L3AaeoM20D_lin/map.gif

But specific humidity, q, should have increased like this big red blob (in ppmv).

http://data.giss.nasa.gov/work/modelEt/lat_height/work/tmp.4_E3Af8aeM20_1_0112_1960_2003_1951_1980_-L3AaeoM20D_lin/map.gif

You can play around with different years and different scenarios here. (increasing temp years versus decreasing temp years for example).

http://data.giss.nasa.gov/modelE/transient/Rc_pj.1.11.html

Or look at lots of different aspects of the model here.

http://data.giss.nasa.gov/modelE/transient/climsim.html

My post can be found here.

I wanted to do another analysis of the atmospheric water content, q, using the RSS SSM/I data set at another latitude in order to check the sensitivity of the first result (reported above) of the global zone from 50S to 50N to the latitude zone selection. To that end, I did another analysis for the global zone from 25S to 25N from July 1987 through January 2009 and present the results below. The R code is essentially the same as for the previous analysis and is not shown here.

There were only slight differences between the trends for the two zones analyzed. They both shown a highly auto correlation of the regression residuals even at higher orders and both time series show a step function at the 1997-1998 time period. Both zones also include zero in the CIs and thus cannot be shown to be statistically different than zero.

Trend slope = 0.00289; AR1 Corr = 0.707; SE = 0.000798; Adjusted CI = -0.00096 to 0.00673

The question remains about the trends found by Trenberth that were claimed to statistically significant. Trenberth used Version 5 of the RSS SSM/I data set for the 1988 to 2003 time period while I used the later Version 6 from mid 1987 through January 2009. Trenberth determined limits using a Monte Carlo procedure and I used the Nychka procedure from Santer et al. (2008) to adjust the CIs. Version 5 of the RSS data set is no longer available to the public and can be accessed only by special permission.

I had the same timeout issue. It occurs after going through the process of selecting the appropriate data and requesting the download. ERA-40 worked like a charm, but for some reason the ERA-Interim isn’t working.

Nice work with the RSS data.