Rain totals, how far apart

The shaded field is the gap between the wettest and driest of GFS, ECMWF, ECMWF AIFS, GEM — four models and three centres, named because a roster is a fact about a chart: two of these are ECMWF's, and where those two agree they agree partly for family reasons. Each model's total is the difference between two of its own accumulations from the same run, so this is what falls inside the window rather than since the run began. Uncorrected model output on every side.

Four windows, and the reason is measured rather than tidy. Over six to twenty-four hours these models disagree by a FACTOR — a median 16 to 20 times — and hardly at all by an amount, so a map of the amount is flat almost everywhere and blown out on a few cells; that short window is what the coverage map counts instead. By three to four days the ratio has fallen to four or five times and the gap itself spans a quarter inch to three, and at six days the ratio holds there while the gap keeps widening. The scale changes between the one-day window and the rest for the same reason — a quarter inch is most of a day's rain and a rounding error over four — so the key is worth a second look after stepping off 24 h. The three longer windows share one, which is what lets the argument be seen widening from three days to six.

The contours and dots are GFS alone, chosen with the chips above the map — one model's own answer inside the argument. Deliberately not an average of the four: these models agree closely on how much water falls over the map as a whole and disagree about where it lands, so an average would draw rain in a place no model forecast.

A dot is a local maximum standing at least 12 mm high over these 24 hours and clearly above its own surroundings, far enough inside the frame for that comparison to be even-handed; up to ten are kept and the heaviest few are named under the map.

A dot is a local maximum standing at least 25 mm high over these 72 hours and clearly above its own surroundings, far enough inside the frame for that comparison to be even-handed; up to ten are kept and the heaviest few are named under the map.

A dot is a local maximum standing at least 25 mm high over these 96 hours and clearly above its own surroundings, far enough inside the frame for that comparison to be even-handed; up to ten are kept and the heaviest few are named under the map.

A dot is a local maximum standing at least 25 mm high over these 144 hours and clearly above its own surroundings, far enough inside the frame for that comparison to be even-handed; up to ten are kept and the heaviest few are named under the map.

Liquid-equivalent, which is what the models publish: frozen precipitation appears here as the water it would melt to. Every window ends at 18Z, the hour the other two precip and temperature maps share, so charts of one moment compare — and each one starts at the same time, so they are nested rather than unrelated slices. The six-day window comes from a second download of each model that lands a few minutes after the first; on the rare run where one model's has not, that window says so rather than drawing the other three.

Nothing here is smoothed. The shading, the contours and the widest gap are all read from the models' own grids at their own resolution, because averaging a cell into its neighbours fills in exactly the extremes this map exists to show — and a number printed over a colour has to agree with it.

Shaded ground is a disagreement rather than an error — none of these models is the truth here, and the map cannot say which is right. For who has actually been closest, the model leaderboards are the place.