The Cherry-Picking Machine

By ordinary least squares — the straight line that minimizes the summed squared vertical distances to the annual points. It's the same fitting method used for the trend overlays on klymot.com's station charts. The slope is reported in °C per decade; the same fit expressed per year or per century would just rescale the number without changing anything else.

It's the 95% confidence interval on the slope: the range of slopes that would be statistically compatible with the scatter of the points around the fitted line. A wide band means the years in the window vary a lot around the line, so the slope is poorly pinned down; a narrow band means the opposite.

One subtlety: consecutive years are not independent — a warm year tends to follow a warm year. The interval here accounts for that by shrinking the effective number of data points using the window's own year-to-year correlation (an AR(1) "effective sample size" correction, standard in the climate literature — see Data sources). Without it, the band would be too narrow.

Each annual value is the average of the station's twelve monthly mean temperatures for that year, from the NOAA GHCN monthly dataset (see Data sources). Years missing any month are left out entirely, so a year with no winter readings can't masquerade as a warm year. Gaps this creates are visible as breaks in the chart line, and a window must contain at least 80% of its years to produce a trend.

The lab uses long-running stations with substantial historical records that collectively span a broad range of surrounding built-up land. They are presented in a random order to avoid suggesting that any one station is the "right" place to start.

Station histories contain changes that have nothing to do with climate: the station moves, the instrument is replaced, the time of day the thermometer is read changes. NOAA's pairwise homogenization algorithm compares each station with its neighbours to detect such changes and estimates an offset for each one; the adjusted series is the reported series with those offsets applied. The unadjusted series is the reported series without them.

You can inspect what the adjustments look like for any individual station on klymot.com (Adjustments tab in the station view).

The shorter lengths (15 and 20 years) match the windows most often quoted in public arguments about recent trends. The longer ones are there because of a specific, genuinely contested question: several studies have reported an apparent oscillation of roughly 60–70 years in parts of the climate record — most prominently in North Atlantic sea-surface temperatures — while others argue the apparent cycle is an artifact of records too short to contain even three full repeats. The Data sources section below cites one paper on each side; this lab endorses neither.

The lengths are chosen so you can probe the idea yourself. If a record did contain a roughly 60-year cycle, a 30-year window could sit entirely inside a rising or falling phase, a 60-year window would tend to span one full swing, and a 90-year window would always straddle more than one. One caution when comparing: the spread of window trends shrinks as windows get longer for any noisy record, cycle or no cycle — so the comparison to make is between the tallies at different lengths, not the mere fact that longer windows disagree less.

The full record is itself a window — the longest one this station allows. Fitting one straight line through two or three centuries answers a different question than fitting one through a decade, and neither is automatically the question you care about. This lab only shows what each choice of window produces; deciding which window matches the question being asked is yours to make — and worth asking of any trend you're quoted.