No it doesn't imply that. Many error ranges follow a normal distribution of probability where the liklihood decreases the further from the midpoint.

Even if the probabilies were equally distributed across the error range (and I don't believe they would be - as if all of a sudden one value is possible and 0.1 below it is totally impossible) you used for co2 rise in 2006 they would be:

1.33-1.73 20%
1.73-2.13 20%
2.13-2.53 20%
2.53-2.93 20%
2.93-3.33 20%

You used the very highest figure of 3.33 which is in the < 20%. That's what I mean by using a liberal error rate - you have taken the most extreme amount of error possible for the example. I understand you were showing it was possible for co2 rise in a certain year to be more than human co2 emissions, but it certainly isn't anyway near as probable as the co2 rise being less than the co2 emissions.

As I have said before a particular year is subject to more uncertainty than a range of multiple years. So there is a small probability that cannot be written off that co2 emissions do not meet the co2 rise in a particular year. It's only when using multiple years that you reduce the uncertainty and see human co2 emissions are greater than overall co2 rise.

At least 95% of the possible configurations are compatible with co2 rise in 2006 being less than human emissions in 2006. Only 5% are compatible with co2 rise being more than human emissions.

Gt co2 rise error range for 2006 10.4 - 26
Gt co2 emissions error range for 2006 22 - 32

You require the co2 rise to be more than 22 in order for co2 rise to be more than co2 emissions. That's just 25% of this error range you can accept (ie only 22-26 out of 10.4-26)

You require the co2 emissions to be no more than 26, otherwise co2 emissions will be more than the co2 rise. That's just 40% of the error range you can accept (ie only 22-26 out of 22 - 32)

Altogether you require both of these to be true. 25% x 40% = 10%

But also half of those 10% of configurations will still result in co2 emissions being more than co2 rise (eg a 23gt co2 rise and 25gt co2 emissions), so in fact you can only accept 5% of possible configurations.

Not to mention that this assumes an flat distribution of probability across the error ranges, when the actual case is probably a normal distribution which would lower your proportion even more.

So the probability that the human emissions in 2006 were higher than the co2 rise in 2006 is more than 95%

I said above that
"a particular year is subject to more uncertainty than a range of multiple years. So there is a small probability that cannot be written off that co2 emissions do not meet the co2 rise in a particular year."
And here you can see that this small probability less than 5% for 2006

That figure gives an indication of how certain I think the recent co2 rise (100ppm) being attributable to humans is. Equally I could say I am also 99.99% sure that common descent of species is true, but no statistical data there either. The 99.99% is to emphasise that I see it as more certain than 80% or 90%.

You do require them to be off. Using the estimated figures will show that the co2 rise is much lower than co2 emissions for 2006. You require the estimates to be off, ie for the co2 emission estimate to be overestimated and the co2 rise estimate to be underestimated. You require both estimates to be off, not just one of them, and they must be off in a certain direction and by quite a lot too.

Using a longer period than a single year minimizes the proportion of the error range for the co2 rise. For example using the 2006 figure you gave of 2.33ppm+-1ppm, that error range means it could be as much as 43% off.

But for a ten year period the error range is just 20ppm +- 2ppm, meaning the actual figure can only be as much as 10% off the estimate of 20ppm. So clearly it is better to use data with less error range no?

And when using data with less possible error what emerges is that co2 emissions are clearly over the co2 rise.

Nope, we don't need to extrapolate - we have the data for the last 10 years so why are you so reluctant to use that? What's wrong with taking the rise over the past 10 years, and emissions over the past 10 years?

Well lets take 20 years then, a co2 rise of about 33ppm +-2ppm (notice the proportion of the error range has decreased again, so now it is only going to be about 6% off, that is the key to why using longer periods causes the fact to emerge that humans are putting more co2 into the atmosphere than co2 concentrations are rising)

co2 rise error range: 241Gt - 272.6Gt

co2 emissions for last 20 years: 485Gt (from http://cdiac.ornl.gov/ftp/ndp030/global.1751_2003.ems 2004,2005 and 2006 taken as the same as 2003)

So the co2 emission estimate is 212Gt more than the upper end of the co2 rise error range. You need co2 emissions to be 44% overestimated in order to stand any chance of co2 rise being more than co2 emissions in that time period.

So no I didn't have to stop at 10 years.

Yes, both. By definition if you take co2 measurements every hour you are going to be taking them every 10 years too...

Look it's simple. We took measurements in 2006 with +-1ppm error, and took measurements in 1997 with +-1ppm error. So the co2 rise in 10 years is calculated as the measured level in 2006 minus the level in 1997, with the error rate being +-2ppm (the combination of both end year errors)

Yes it's simply a matter of taking the measured co2 concentration in 1997 and taking this from the measured co2 concentration in 2006. That will be the rise over the 10 years.
Global Monitoring Division
http://www.cmdl.noaa.gov/projects/we...co2_mm_mlo.dat

http://cdiac.ornl.gov/ftp/ndp030/global.1751_2003.ems gives you the years 1997-2003, then I took 2003 value as the 2004-2006 figure. That actually works in your favor given that co2 emission rates have risen in that time period (Growth rate of carbon dioxide emissions doubles since 1990s "From 2000 to 2005, the growth rate of carbon dioxide emissions was more than 2.5 per cent per year, whereas in the 1990s it was less than one per cent per year"), but I do not have the actual figures for every one of those years.

If you look at the monthly data you can see the natural background noise, that does cause uncertainty in estimating the co2 level for a given year:
http://www.cmdl.noaa.gov/projects/we...co2_mm_mlo.dat

When the 90% confidence interval is between 3.1 and 3.5, how much confidence do you actually believe there is for a value below 0.1? It's very likely you are outside even the 99.9% confidence interval there.

Well you are implicitly doing this by ignoring the land use change emissions. I am also arguing that it cannot be partitioned without using the atmosphere.

Yes. The second part of the table is giving the ratios for the anthropogenic co2 emissions go into. The second part is dependent on the first. The first is not dependent on the second.

Just because it is possible for something not to be true doesn't mean it cannot be scientific fact.

Something might be possible, yet too improbable to entertain. Effectively a low enough probability event is treated as effectively impossible in science. Otherwise nothing could be a fact and be given in absolute and definitive statements.