Weather forecast is best told as a narrative and reducing it to a sports score like temperature / rainfall will always be problematic for many regions.
I’m in SF and I have not looked at a “TV/app style” forecast in years. Instead every morning I read:
I also like the pure NWS data, and will take a moment to shill Deep Weather on iOS, which provides a nice, easy interface to the NWS products for an area.
I've been looking for something like this but unfortunately I am halfway between two stations and with the mountains here makes enough of a difference that neither one is accurate.
The `<pre>` on the NWS site was really messy on mobile, this is nice and neat. Feel free to fork it and use your own, or this link should work for your NWS site too!
Does anyone know of any weather sites which show the actual outcomes? It is very easy to find forecasts. It is quite difficult to find what the weather actually was on a given day at a location, but at least it is possible. I have never seen a site that shows the predicted weather for a day alongside the actual weather for that day. I presume professional meteorologists know where to look for this data, but for the rest of us, it feels like it is hidden.
I know no one wants to advertise their mistakes, but showing a history of your accuracy which was transparently updated would build more credibility with the public than anything else. Perhaps someone is scraping this info and posting it?
Somewhat related: Before launching our new long range (5 day) turbulence forecast tool (for nervous flyers), we spent time experimenting to see how far out we could reliably predict turbulence. Once we had something promising, we started posting daily snapshots—each showing the forecast from five days earlier alongside what actually happened. The model updates continuously, and we’ve kept the process public to demonstrate how well it works, not just claim it does.
This NOAA National Blend of Models (NBM) viewer lets you enable observations and then roll back the forecast initialization date to see forecasts and observations at the same time.
It’s experimental, and selecting weather stations is a little clunky, but it has some really cool info that’s hard to find other places.
I have wondered this for years. Several times I looked for a dataset of historical forecasts. It seems wild that with all the data available out there, there's no go-to source that has past forecasts together with the actual recorded data for each day. That said, these posts do mention some sources:
One challenge is the forecast is constantly updated, as often as several times in a single day. So which forecast would you use for the predicted weather?
I'm curious if anyone has a good solution for this, how to accurately display cases where a storm might hit either Monday or Tuesday, say 50% chance each. If you just say there's a 50% chance of rain on both days, it looks like there's a significant chance (I guess 25%) that it will rain both days, when the real likelihood of both days raining might be far lower.
- The chance of precipitation line tends to gradually go up and then back down over several hours.
- The expected mm of precipitation bars tend to be more actionable: the chance of precipitation may be high, but if the expected amount is low you are not likely to experience much, if any rain; much less need an umbrella.
Tangent, but I encourage everyone to actually learn how to predict the weather. Every region has its own peculiarities, and even on a small scale geography may have quite significant impact, so it is natural to focus on where you live. Moreover, you quickly find out that this skill is not about looking at the sky in a single moment and taking a snapshot of it, but rather about recognizing some trends and changes, more prolonged and more subtle as your knowledge grows.
You will find out how urban areas behave differently than rural ones, how plains differ from hills and mountains, how forest differs from meadows, where the small cumulus clouds are being formed and where they disappear, how wind shifts before the storm, how animals change their behavior etc. Some of this stuff can be widely applicable, but some will be specific to your area. And you do not even need to live in some remote areas to enjoy it.
I aggregate weather forecasts for the Event Horizon Telescope Collaboration, which is the collaboration behind those black hole images you might have seen.
We want to pick the best nights during an observing window based on the weather in 12 locations around the world. These are mostly locations on the peaks of mountains, where it's hard to do a good forecast because the ground is rugged on length scales that are smaller than the grid in the numerical computation.
Forecast accuracy has improved in recent years, but I'm really looking forward to AI forecasts. They appear to fix some systematic errors that are still present in traditional forecast simulations.
Weather forecasts are based on models and extrapolation based on a lot of factors. I'm not too certain but I think that the smallest weather model cell size is about two miles squared or perhaps cubed - I don't know how the models work.
Weather forecast distribution is by "news". News is deployed at mostly fixed intervals.
Anecdote: In Yeovil, Somerset, UK. The topography here (its quite hilly here in a county with some fairly famous "levels") means that it can be pissing down with rain on one side of the town and be dry on the other side and so on.
Weather forecasting is a quite hard problem. The really hard problem is delivering it.
I lived in a place where there is a lot of rain. But the rain can have many faces and especially in summer all weather sites would forecast rain and use a rain symbol for days when the data clearly says there is a high change of minimal rain and a lot of sun otherwise. Which is argueably a good "sunny" day for that area.
My solution was to reduce days to average temperature and sun hours instead of using precip to visualize a full day. And then focus on a "24 hour display" with 24 little icons where the rain instantly seemed a lot less.
Readit News
I’m in SF and I have not looked at a “TV/app style” forecast in years. Instead every morning I read:
https://forecast.weather.gov/product.php?format=CI&glossary=...
During the day I look periodically at:
https://fog.today
NWSNOW for Android, includes the vaunted "Forecast Discussion". Side load, No ads, no tracking, pure NWS and NOAA data.
https://www.nwsnow.net/
The `<pre>` on the NWS site was really messy on mobile, this is nice and neat. Feel free to fork it and use your own, or this link should work for your NWS site too!
https://greg_dryke--0816a61c279c11f09d52569c3dd06744.web.val...
(No promises on possible future changes if you do start using it, but it's pretty stable.)
I know no one wants to advertise their mistakes, but showing a history of your accuracy which was transparently updated would build more credibility with the public than anything else. Perhaps someone is scraping this info and posting it?
https://www.turbulenceforecast.com/drift
It’s experimental, and selecting weather stations is a little clunky, but it has some really cool info that’s hard to find other places.
https://apps.gsl.noaa.gov/nbmviewer/?col=2&hgt=1&obs=true&fo...
Edit:
This GEFS plume viewer is cool, too.
https://www.emc.ncep.noaa.gov/users/meg/gefs_plumes/index.ht...
You can check a box to plot observations and then pick an older “cycle”.
Both are limited to the US.
https://opendata.stackexchange.com/questions/2009/historical...
https://www.reddit.com/r/meteorology/comments/1crs6lu/seekin...
(These aren't "sites" that display the past data, just datasets you can download.)
Deleted Comment
On https://weather-sense.leftium.com, you can see the daily forecasts change on a regular basis.
Even a site that displays 'just' recent weather specifics (sans how the forecast was) would be quite helpful.
e.g., the hourly precipitation for the last week or month.
Am NOT talking about generalized monthly climate data, fwiw.
Anyone?
https://mesonet.agron.iastate.edu/request/download.phtml
- The chance of precipitation line tends to gradually go up and then back down over several hours.
- The expected mm of precipitation bars tend to be more actionable: the chance of precipitation may be high, but if the expected amount is low you are not likely to experience much, if any rain; much less need an umbrella.
- Sample screen shot with precipitation in forecast: https://github.com/user-attachments/assets/e003badb-4832-430...
---
https://openweathermap.org even gives minutely rain forecasts for the next hour.
You will find out how urban areas behave differently than rural ones, how plains differ from hills and mountains, how forest differs from meadows, where the small cumulus clouds are being formed and where they disappear, how wind shifts before the storm, how animals change their behavior etc. Some of this stuff can be widely applicable, but some will be specific to your area. And you do not even need to live in some remote areas to enjoy it.
We want to pick the best nights during an observing window based on the weather in 12 locations around the world. These are mostly locations on the peaks of mountains, where it's hard to do a good forecast because the ground is rugged on length scales that are smaller than the grid in the numerical computation.
Forecast accuracy has improved in recent years, but I'm really looking forward to AI forecasts. They appear to fix some systematic errors that are still present in traditional forecast simulations.
Weather forecast distribution is by "news". News is deployed at mostly fixed intervals.
Anecdote: In Yeovil, Somerset, UK. The topography here (its quite hilly here in a county with some fairly famous "levels") means that it can be pissing down with rain on one side of the town and be dry on the other side and so on.
Weather forecasting is a quite hard problem. The really hard problem is delivering it.
My solution was to reduce days to average temperature and sun hours instead of using precip to visualize a full day. And then focus on a "24 hour display" with 24 little icons where the rain instantly seemed a lot less.
Basically focusing on the good things.
Linked is that definition explained:
https://www.weather.gov/media/pah/WeatherEducation/pop.pdf