Follow

A few days ago @1wheel made a nice version of the chart that also shows the individual temperatures as points + linear trend lines
bl.ocks.org/1wheel/89ec11ab72a

And today I forked Adams chart to use Loess regression instead of the linear model, because I thought temperature isn't increasing linearly (but maybe exponentially?)

bl.ocks.org/gka/1d716deada5f05

the result is sort of interesting. for some months, like July and August the temp. increase really has picked up the pace in the late 1960s. But for October the increases seem to be a lot more "linear". Might be just some Loess artifacts.

@gka I’m a fan of showing as much data as possible but I’m leaning towards Lisa’s version. The simple arrows pointing up convey the situation bluntly. In fact, I’d be curious to see them on a common baseline for once. I find myself looking for the moths that increased the most. (I can convert anything to a bar chart - I’m fun at parties)

@gka I generally like LOESS, any "methodology" for picking the smoothing etc. parameters? Was it chosen to give a visually smooth yet non-linear result? Also, wondering if curves would differ if the month beginnings and ends were augmented by a few adjacent data points from the prior and subsequent months for continuity

@gka Btw. one interesting visual effect of this version, or even the Adam's linear version (both more so than with @lisacrost's vertical arrows) is that the temp increases in the early months coincide with the monthly temp increase, appearing as if they were linked, while the right-hand side of the chart has almost perpendicular offshoots. Despite the common metric, it makes me ponder if there's a way to visually uncouple, eg. years by Z axis, or X/Z rotating around mid-month Y? 😂

@monfera the "methodology" is starting with bandwidth/smoothing of zero (full noise) and gradually increasing it while observing the patterns that emerge. it's really useful as interactive method, as the parameter depends on the data. At least this is what Cleveland suggested in his LOESS paper in 1979: pdfs.semanticscholar.org/414e/

Sign in to participate in the conversation
vis.social

vis.social is an open social platform for creative people, especially anyone in data, visualization, creative coding, and related arts and research.