Possibly consider other providers due to questionable Open-Meteo accuracy?

[ferdnyc]

Description

This is something of a mixed bug report / feature request, because I want to first document the reason for the feature request. (And just to state it up front: Complaints about data accuracy are not in any way SimpleWeather bugs, obviously. But "using Open-Meteo as provider" may be a bug, if they have accuracy issues.)

My location is one of the boros of New York City, so it's well served by most weather providers. Here's how various sites report temperature data and forecasts for the next few hours.

Time	Weather.com	Weather Underground	Open-Meteo
current	14°F	14°F	4°F
6am	13°F	13°F
7am	12°F	12°F	-1°F
8am	13°F	12°F
9am	14°F	13°F	4°F
10am	15°F	15°F
11am	17°F	17°F	17°F
----	-----------	-------------------	-----------
1pm	21°F	20°F	21°F
3pm	22°F	22°F	23°F
5pm	22°F	21°F	20°F

(My original plan was also to include OpenWeatherMap, which I have an API key for from the previous extension, but their new weather site uses WebGL and either the page crashes trying to initialize WebGL, or it succeeds and practically hangs the browser with multi-second input lags, in Chrome on my Fedora 43 box with almost-nonexistent GPU support. Their old site loads, but it only shows hourly temps in a graph with no precise data points. Suffice it to say, it appears closer to Weather.com/WUnderground than to Open-Meteo's numbers.)

I don't know why Open-Meteo's current-conditions and next-several-hours forecast temps differ so wildly from other sources, when their numbers going farther out tend towards agreeing with them. But this isn't the first time I've noticed my computer reporting a current temperature and immediate forecast that's completely out of sync by several degrees with my phone and every weather website I can find.

As I said, I personally already have an API key for OpenWeatherMap, and their free offerings are so generous that I never needed to pay them a cent to use their forecast data in the old extension. The forecast is limited to 3 days, instead of 7, but I prefer less data that's more accurate, all things considered.

[ferdnyc]

Actually, when I say Open-Meteo data tends towards agreeing with other providers' numbers, that doesn't really hold either. Here are the 24-hour forecast numbers the extension is currently showing, vs. Weather.com's hourlies:

Time	Open-Meteo	Weather.com
6am	2°F	13°F
8am	-1°F	13°F
10am	13°F	15°F
12pm	19°F	19°F
2pm	22°F	22°F
4pm	22°F	23°F
6pm	14°F	21°F
8pm	8°F	19°F
10pm	3°F	17°F
12am	1°F	15°F
2am	0°F	13°F
4am	-2°F	12°F
6am	-4°F	11°F
8am	-8°F	10°F

Open-Meteo seems to have real difficulties with nighttime temperature forecasting / reporting.

[romanlefler]

Thank you for the comment and data you've neatly collected. I thought it probably just used the NWS for the U.S. but it appears it doesn't, which is probably where the inaccuracy comes from.

I originally used it because the no API key + open-source makes it super easy, but it looks like that's coming with some sacrifices. I'll prioritize adding another more accurate provider when I have time.

[ferdnyc]

I thought it probably just used the NWS for the U.S. but it appears it doesn't, which is probably where the inaccuracy comes from.

*nod* I get the impression they're trying to be an aggregator a la DarkSky (RIP). But clearly they don't have DarkSky's algorithms — which improved accuracy by combining multiple data sources for the same region —­ and what they do have seems a bit buggy.

It's also possible that tweaking the request could select certain data sources and improve the accuracy, but if the default is anything other than "most accurate data available" that's still a worrying sign.

[ferdnyc]

Hah! Well, I think I've just discovered the source of the inaccuracies.

Weather.com, right now

GNOME Weather, right now

The Norwegian Meteorological Institute website, right now

(In °C because there doesn't seem to be any way to change it.)

So, it appears Norway is skewing the numbers for both GNOME Weather and Open-Meteo.

[ferdnyc]

Huh! Well, this is promising.

I edited the extension code to change one line in openmeteo.js

--- openmeteo.js.orig	2026-01-31 14:50:26.453191058 -0500
+++ openmeteo.js	2026-01-31 14:50:38.749203079 -0500
@@ -18,7 +18,7 @@
 import { Condition, gettextCondit } from "../weather.js";
 import { getGIconName, Icons } from "../icons.js";
 import { getTimezoneName } from "../utils.js";
-const ENDPOINT = "https://api.open-meteo.com/v1/forecast";
+const ENDPOINT = "https://api.open-meteo.com/v1/gfs";
 export class OpenMeteo {
     #soup;
     #config;

...And now my conditions/forecast numbers are no longer insane.

/v1/gfs is the Open-Meteo endpoint that represents NOAA forecast data, specifically and exclusively, and therefore eliminates the influence of Norway's patently insane forecast models for the US.

So, this both:

1. Proves that Open-Meteo's aggregated forecast models can be completely broken by outlier non-local data (so, they have a weighting problem), and the issue is definitely on their end
2. Means that Open-Meteo data can still be useful, but it may be necessary to allow the user to select which of their data sources to pull from (or do so automatically by location, somehow).

[ferdnyc]

...And now my conditions/forecast numbers are no longer insane.

Hmm. Well... maybe I spoke too soon.

The numbers are no longer insane, but they're not quite "right" either.

Like, with the change to the /v1/gfs endpoint, the extension is now reporting the current temperature as 21°F. That's certainly an improvement over the 10°F (!!) it was reporting before the change, but it still disagrees quite a bit with Weather.com's current temperature of 27°F. And the hourly temps it's showing are similarly out of whack, though again significantly less so than before.

[romanlefler]

I added a hidden feature to switch weather models, but I see what you mean I think I'll just have to add another provider. Interestingly in other American locations it's basically correct, but I see in New York there seems to be a larger inaccuracy (maybe because it's coastal...?). Regardless changing to use NOAA didn't seem to make a difference, so I'll just have to add another provider.