The Spread

spreadwx.comRainfall forecast: how much the models disagree — The SpreadHow far apart GFS, Euro (ECMWF), AIFS and GEM are on how much rain falls over the next one to six days — with one model's own totals drawn inside the argument.

Rain is coming to #losangeles and #sandiego! But how much? There's a good amount of disagreement now through Sunday - more than 2 inches delta in San Diego. Either way, everyone is going to get some. #cawx spreadwx.com/maps/precip/...

spreadwx.comHow much rain the weather models forecast in Phoenix — next 14 daysEach model's rain for Phoenix, Arizona, day by day and over the next six days: GFS, Euro, ICON, GEM, UKMO and the AI models, the wettest named, no average.

Lots of rain coming to Phoenix over the next 6 days, today through Sunday: AIFS forecasts 3.29 in, AIGFS 1.39 in, 1.90 in apart. The wettest day is Fri Oct 30 with UKMO at 2.52 in. This is an interesting setup where the AI models are furthest apart in rain totals. spreadwx.com/location/pho...

spreadwx.comTemperature spread in forecast weather models — The SpreadHow far apart GFS, Euro (ECMWF), AIFS and GEM are on the coming days' temperature, at every point on the grid they share.

Sunday afternoon's temperatures: GFS, ECMWF, AIFS and GEM are furthest apart near Caldwell, Idaho, 20.6°F between the warmest and coolest of the those four models. Unshaded ground on the map is where all four land within 2°F of each other. spreadwx.com/maps/tempera...

spreadwx.comWhere the weather models disagree in Denver — next 14 daysThe range GFS, Euro, ICON, GEM, UKMO and the AI models span on Denver, Colorado's daily highs, day by day, with the roster named per lead.

Why do seven models disagree by 13° on Denver's high for Monday? It's the position of a cold front: AIFS has it about 130 miles east-southeast near Lamar, Denver on the cold side (48°). GEM's is about 40 miles northeast near Greeley, Denver on the warm side (61°). spreadwx.com/location/den...

spreadwx.comHow much rain the weather models forecast in Miami — next 14 daysEach model's rain for Miami, Florida, day by day and over the next six days: GFS, Euro, ICON, GEM, UKMO and the AI models, the wettest named, no average.

Bad timing in Miami wrt the king tide: potentially heavy rain over six days, Thu Oct 22 to Tue Oct 27: UKMO forecasts 7.36 in, GEM 0.46 in, 6.90 in apart across seven models. The wettest day is Mon Oct 26, where seven of seven forecast rain, UKMO wettest at 3.96 in. spreadwx.com/location/mia...

spreadwx.comHow much rain the weather models forecast in Los Angeles — next 14 daysEach model's rain for Los Angeles, California, day by day and over the next six days: GFS, Euro, ICON, GEM, UKMO and the AI models, the wettest named, no average.

On the first big Pacific storm of the season: rain in Los Angeles over six days, Wed Oct 21 to Mon Oct 26: ECMWF forecasts 1.29 in, UKMO 0.45 in, 0.84 in apart. Could be the start of a very wet winter. spreadwx.com/location/los...

How much rain will fall in Phoenix? Models have a big spread over the next 6 days, from ICON 1.44 inches to GFS 0.24 inches. The wettest day is Thursday, where 6 of 7 models forecast rain. spreadwx.com/location/pho...

spreadwx.comHow much rain the weather models forecast in Boston — next 14 daysEach model's rain for Boston, Massachusetts, day by day and over the next six days: GFS, Euro, ICON, GEM, UKMO and the AI models, the wettest named, no average.

Will rain from the upcoming nor'easter reach Boston? Over six days, Mon Oct 19 to Sat Oct 24: AIGFS forecasts 0.39 in, AIFS 0.06 in; most models are dry. spreadwx.com/location/bos...

spreadwx.comWhere the weather models disagree in Dallas — next 14 daysThe range GFS, Euro, ICON, GEM, UKMO and the AI models span on Dallas, Texas's daily highs, day by day, with the roster named per lead.

12 days out, the models are furthest apart in Dallas on Fri Oct 30: four of them put the high anywhere from 55° to 89°, AIFS coolest and GFS warmest. The typical spread there at that lead is 14°. spreadwx.com/location/dal...

spreadwx.comWhere the weather models disagree in Minneapolis — next 14 daysThe range GFS, Euro, ICON, GEM, UKMO and the AI models span on Minneapolis, Minnesota's daily highs, day by day, with the roster named per lead.

Ten days out, the models are furthest apart in Minneapolis on Tue Oct 27: four of them put the high anywhere from 42° to 72°, AIGFS coolest and GFS warmest. The typical spread there at that lead is 12°. spreadwx.com/location/min...

spreadwx.comWhere the weather models disagree in Philadelphia — next 14 daysThe range GFS, Euro, ICON, GEM, UKMO and the AI models span on Philadelphia, Pennsylvania's daily highs, day by day, with the roster named per lead.

Monday's high in Philadelphia, three days out: seven weather models put it anywhere from 51° to 65°, in three camps · median moved 64° → 61° since yesterday. GFS is coolest at 51°, ICON warmest at 65°. spreadwx.com/location/phi...

Models disagree by different amounts on different days. Does the disagreement tell you anything about how wrong the forecast ends up? 258 verified days, 21 cities, ERA5 truth. Seven models: ECMWF, GFS, UKMO, ICON, GEM, and the two AI runs, AIFS and AIGFS.

The edges do have regulars. AIFS was the coolest of the seven on 71 days and the warmest on none; UKMO warmest on 47, coolest on 6. Bias runs AIFS −1.5°F to UKMO +1.1°, ECMWF nearest zero — and the median beat all seven, 1.78° against ECMWF's 1.83.

11 Aug

Caveat that carries the weight: 20 August days, 21 US cities, one lead. That's this sample's bias, not a skill ranking, and season or region should flip the sign.

Sort the days by how much those seven disagreed one day out. The third that agreed most: the consensus missed by 1.27°F on average. The third that argued most: 2.38°. Same direction at 3, 5 and 7 days — 1.35x to 1.69x.

11 Aug

Before believing it: r = 0.37. Real, and loose. But it isn't one maverick — drop any single model and it holds, 1.48x without AIFS through 2.07x without UKMO. Spread is informative, not decisive.

A forecast page overwrites its old answer. Here is one date with the argument kept. Seattle, Friday Aug 7: fourteen days out, four models put the high anywhere from 71 to 96°F. It verified at 86°.

The final consensus called 82°. ERA5 says the day came in at 86° — inside the range the models still spanned, above the middle of it. ECMWF landed closest, off 0.6°. AIFS ran 6° cold.

spreadwx.comWas the forecast right for Seattle on Fri Aug 7?14 days out the models said 71–96; it hit 86. ECMWF led (off 0.6°F); AIFS ran 6°F cold.

One date is not a league table — a model can win a day it deserved to lose. Every lead, every model, and what verified: spreadwx.com/model-accuracy/seattle-washington/2026-08-07

A forecast page overwrites its old answer. Here is one date with the argument kept. Seattle, Friday Aug 7: fourteen days out, four models put the high anywhere from 71 to 96°F. It verified at 86°.

On the way there the spread doubled twice — 5°→11°, then 8°→18° — and the camps merged three times, changing their mind twice. Sunday the cooler camp won. Monday night the hotter camp. Thursday the middle one.

112 archived snapshots later — every one different from the last. Individual model values moved. Two more models came into the window. The cross-model answer did not change.

Seattle, Mon Aug 10. Last material change: Saturday night, 6° cooler at eight days out. 131 snapshots since — median 76.1°F then, 76.5° now. Spread 7° → 10° as ICON and UKMO entered range. Nothing crossed the bar.

"Nothing material changed" is an analytical conclusion, not an absence of notification. A product that only speaks when it can alarm you is selecting for volatility. I'd rather mine be able to say the forecast held, and show the runs behind it. #wawx

Weather models don't simply produce a spread. They form scenarios, and you can watch one lose its supporters.

"Confidence increased" hides what happened. One possible future lost its supporters. NBM and my own blend never join the camps; a consensus product would manufacture the agreement. How would you tell real convergence from models sharing upstream assumptions?

Weather models don't simply produce a spread. They form scenarios, and you can watch one lose its supporters.

The winners: AIGFS, GEM, ICON, UKMO. I name only the winning roster on purpose. Camp membership belongs to an event, not to a date: the Aug 3 merge named a different set, and splicing the two would invent a story neither run told.

Weather models don't simply produce a spread. They form scenarios, and you can watch one lose its supporters.

New Orleans, Sat Aug 8. Aug 3: the models merged on one answer. Wednesday evening: they broke into three camps again. This morning they re-merged: a lone 99°F solution and an 87–88° pair gave way to the 91–94° group. Consensus 88–94°, spread 12° → 5°.

What counts as a real change? A forecast should say when its own story changes, and stay quiet when it hasn't. Our bar isn't a round number: max(4°F, 1.5 × our blend's recent scatter at that lead). It is tighter for tomorrow than for ten days out, because our errors are.