You can see that the trend has leveled off since around 2005, but that it clearly rose between 1998 and 2005. This isn't too surprising, nor does it mean that global warming has stopped since 2005. You can find several similar 20year time periods over the last 40 years where atmospheric temperatures appear to have leveled off or even fall for short periods of time despite the overall warming trend. For instance, here's a look at the first full 20 years of the UAH satellite record:
You can do this with GISS or any of the other temperature datasets. All it shows is that by carefully choosing short chunks out of a larger dataset, you can make the data show anything, from warming to cooling–even a trend that looks more like a mustache on an oldtime Hollywood movie villain than a trend. In science, carefully picking your start and end points to get the answer you want is called cherrypicking. And if you have to resort to cherrypicking to support your argument, it means that your argument was invalid to start with.
How does this apply to the "no warming since 1998" claim? When skeptics talk about "no warming," they're usually talking about linear regression trends, not loess regression trends. Linear regression gives trends that look like this:
Linear regression gives the average trend over the entire time period, ignoring the shortterm fluctuations in the data that loess regression picks up. The linear regression trend shown in the above graph showed an average temperature rise of +0.010199ºC +/ 0.001984 per year between 1979 and 1999. And yes, that rise is statistically significant, with a pvalue of 0.0000005566. For those who haven't taken a statistics course, that pvalue means that there's a 1 in 1,796,622.35 chance that the trend is due to random chance. Normally, we say that a pvalue of 0.05 (a 1 in 20 chance) is statistically significant, with a smaller pvalue obviously better. Above 0.05, we chalk up the results to random chance and say that it's not statistically significant.
The main limitation with linear regression is that you need quite a bit of data before any trend will be statistically significant. Global temperatures fluctuate due to El Niño/La Niña, volcanic eruptions, etc. You need enough data so that regression can pick out the trend despite those fluctuations. Santer et al. (2011) showed that you need at least 17 years worth of data to reliably pick out linear regression trends in global temperature data. Get much below that and you run the risk of simply not having enough data. So with that background, let's look at the linear regression trends for different starting points:
Start year

Linear trend (ºC per year)

pvalue

2000

0.011098

0.0005069

1999  0.014848  0.0000003344 
1998  0.005549  0.05915 
1997  0.009302  0.0006517 
1996  0.012098  0.000001458 
1995  0.012356  0.00000006008 
1994  0.014259  0.00000000001891 
1993  0.017354  <0.00000000000000022 
No comments:
Post a Comment