That's because they carefully picked 1997 as the starting year. What does the trendline look like if you pick other starting years?
More generally, why do we assume that a linear trendline is the appropriate way to analyze this data? Has anyone actually done an analysis that doesn't start from that assumption?
Thanks for the links. I at least see why the second graph had to pick 1880 as the starting point, since that's as far back as the instrumental record it used went. And at least it considered the possibility of a non-linear fit.
However, the curve fitting is still subject to error, because you don't have the trend before 1880, at least not in this dataset. What if the temps around 1880 were anomalously warm compared to, say, 1800? (Which we have reason to believe they were.) Then the actual trendline from 1800 or so might still be roughly linear.
But more importantly, 1880 is still an arbitrary starting point; any starting point is arbitrary unless you know you have the entire dataset, which we obviously don't. If the actual trend is, say, a sine wave with a period of roughly 800 to 1000 years, with the last peak being around 1000 - 1200 AD and the last trough being around 1600 - 1700 AD, what we're seeing now could just be the approach to the next peak.
I note that the article links to a GRL paper, which I'll have to read; its abstract doesn't make it quite clear what is actual data and what is extrapolated from models.
Top graph is slope of a linear trend line for the 15 years prior to each indicated year. Bottom graph is the underlying temperature data.
For 30 of the 15-year blocks the trend was negative; for 88 the trend was positive. The last year in which there was a linear cooling trend for the 15 years prior was 1977.
http://blogs-images.forbes.com/petergleick/files/2012/02/Glo...