Why are weather apps so bad at telling you the current weather?
I understand that weather on TV can’t be hyperlocally accurate. But a weather app on my phone has my exact GPS coordinates. Why can’t it tell me exactly when a rain cloud will be passing over my location?
It’s gotten to the point where I just use precipitation maps to figure out my rain chances for the day.
The hourly forecast is mostly useless because it’s not a chance % but a % of the area that will be raining.
Weather apps don't do real time analytics, but show you the forecast some nearby weather station has calculated. Whether that's based on current data or a couple hours ago depends on the exact provider they use. And hardly anyone of those are done by actual humans, it's aggregated statistics.
If you look at precipitation maps, you are doing that forecast by yourself based on cloud movements and local knowledge, something no machine-generated forecast can do as good.
Plus, there's usually one weather station covering a large area, so hyperaccurate predictions would have to be made just for you - which simply costs to much.
Nearby is so highly dependent on where exactly those are located, and what they're connected to (some are handled by local volunteers that have hardware that reports periodically as opposed to being operated by an agency directly). Various apps don't all connect to the same data sources.
Official reporting locations may not actually be close to you and weather can be highly localized. A mile can make a massive difference in weather in some regions, and the official recording location for the city is 10 miles away.
Sounds like the system is just stuck on old tech. If I can tell that rain will reach my area from a precipitation radar map then I’m fairly certain an ML based system can do this too.
I am a pilot and flight instructor. I took a full year of meteorology at ERAU. Most weather forecast products presented to the general public are completely worthless, and we really should run "AccuWeather" out of business.
The temperature number they show you for what it is "now" was probably taken by the AWOS at the closest municipal airport to your location. If you've ever noticed like a car GPS reporting weather from three towns over and not the one you're in right now, it's because the town you're in right now doesn't have a nearby airport or it doesn't have an AWOS.
10-day forecasts are generated by tarot cards and have no basis in reality. Actual aviation weather forecasts seldom reach out beyond 24 hours.
Your best bet for getting a complete understanding of the weather is to start by looking at the GOES satellite imagery, ie actually look at the Earth from geosynchronous orbit and look at the clouds. You know those maps with the blue lines with triangles on them and red lines with half circles and big blue Hs and big red Ls? Those are called Prognostic charts, those map where high and low pressure systems and fronts are. Look at one of those and compare them to the satellite images. Then look at Doppler radar imagery, which shows where precipitation is. Look at what it's been doing for the last few hours, and then you can get a good sense of generally what the weather is going to do in the next few hours. Beyond that, not even God knows because the prick hasn't decided yet.
It's not that surprising to me. Chaos theory is not well known/understood and we are accustomed to seeing weather reports from childhood, it's not odd to expect them to be accurate.
My main beef with them is they take data from the public National Weather Service and package it for profit, and they're backing proposed legislation that would prevent the NWS from distributing their weather data directly to the public.
Other than that, most of what AccuWeather does is make the pretty weather maps for the TV stations, and most of the way they make them pretty is by making them less accurate, complete and precise. To a pilot, for whom weather information is a critical safety tool, an AccuWeather product is a bit like a traffic light covered in a half inch thick layer of Spongebob stickers.
The NWS is already trying to measure and predict the giant ball of chaos that is the Earth's atmosphere with probably fewer tools than they'd like, and AccuWeather takes that data and makes it worse so that it's prettier for TV.
Especially because in some places, with more or less frequency, it can be raining somewhere and not raining a mile from there, and randomly just interspersed around randomly. I hear Florida is exactly like that and it makes them extremely hostile towards meteorologists.
MinuteCast from AccuWeather does exactly this. It looks at your location, looks at radar data for storm systems approaching your location, and estimates when precipitation will start at your location and how intense it will be. It's generally pretty accurate, with some limitations. It seems to be pretty good for consistent rainstorms but it can get tripped up by pop-up thunderstorms, where the radar track can go suddenly from no rain to downpour. It doesn't make predictions more then 2-3 hours out because past that timeframe it's not easy to predict if weather will continue on its current track or change direction. Even with the limitations, I use it all the time. Mostly to tell if I should take the dogs out right away, or if I should wait an hour or two.
Darksky could do it back in the day more or less. you'd get messages that it would rain in about 15 minutes and stop in the next 30.
Thing is, precep maps don't work everywhere. You're probably in a location like me where a thick front rolling through will almost always bring rain. If you get into warmer tropical climates, rainclouds will just poof out of nowhere and drop rain on your ass while other crazy fronts will pass over with nothing but some dark clouds.
Weather prediction at point locations is extremely challenging to get right because we simply can't observe and make predictions for every single square inch of the earth. Many weather models are run on grids with boxes about the size of a few kilometers at the smallest scale, which means that any physical process in the atmosphere that is the size of that box or smaller won't be represented well by the model.
Specifically on your point about clouds passing over your location, cloud and precipitation formation is even more challenging. Clouds and precipitation form due to atmospheric processes ranging from hundreds of kilometers all the way down to micrometers, which practically means the weather models are making an educated guess (albeit a very good one that is informed by scientific research) about when and where clouds will form. And when a model does predict a cloud, it will cover an entire grid box.
Finally, I saw you made a comment about how machine learning should improve forecasts, and in fact it does! But the weather community is still working on data driven models (as opposed to models that solve physical atmospheric equations), and most of them are run by private companies so their output is not free. As these data driven models get better, it may be possible that they will be able to make predictions at scales less than a kilometer.
Unless I misunderstood what you said, that's not it either. 50% chance of rain means exactly that: according to their forecast models, there is a 50% change it will rain. Snopes did a writeup of this.
The statement “there’s a 40% chance of rain at any given point at any given time in the forecast area/period” is an average over both area and time.
Many different actual distributions of rain could result in that average, including a 100% chance of it raining 100% of the time in 40% of the are or a 40% chance of it raining in 100% of the time in 100% of the area, and a 100% chance of it raining 40% of the time in 100% of the area. Real distributions are typically messier than that.
Look at a weather map with animated radar overlay. You will often see precipitation approaching and can predict how soon based on its speed and heading.
I know any in my case. I'm in a black hole of info. The nearest weather station is over 30 miles away and it's not remotely accurate for my location because of terrain changes between here and there. There isn't even a weather station in my overly large and oddly shaped county.
We went so far as to contact the people in charge of weather stations to see if we could set one up in our yard and while there was some initial excitement from them they eventually ghosted us.
Why not buy a cheap weather station for kids that come with Bluetooth or server? There must be one with a web page on it ready to go for the less technically inclined.
I am the only nerd across multiple generations.
My nephew is a Hikikomori and not traveling to see me.
I'd rather have a real system that connects to the network and fills the need for accurate weather in my town.
NOAA sources are usually good in the US, weather.gov for a quick map and search by zip code or city, but they follow the same % system mentioned as most do
The couple apps/widgets I've tried haven't been good for working with VPN unless I want to know the weather halfway around the world.
For one thing, there's two competing weather services providing the data to countless apps in the US and one of them has more money to throw around than the other.
The weather channel has better weather predictions overall than Apple's own weather app, as rated by Forecastadvisor.com, but is not as accurate as Accuweather is although it's used in more apps.
Weather is about tracking and predictions. It's never going to be completely 100% correct. But taking a hodgepodge of information from several prediction services means you're more likely to be less accurate overall despite what people may think.
If all the private company weather services were only getting their info from the NOAA we wouldn't have such varying results most of the time. Which is basically my point. The results vary because they don't just use the NOAA's data and predictions. The second one is actually the US Armed Forces.
I went to Amsterdam over the weekend. The weather apps said it was gonna rain, did it fuck. I brought my puffer jacket and was almost dying from the heat until I got to the hotel room. Never had to wear it during the trip.
I just looked at my weather app (Today Weather), and the report of the current temperature varies by 6°F depending on which data set I choose. I go with the National Weather Service (federal govt).
As far as predicting precipitation, the radar seems pretty good for the next few hours. I press play on it, and it tells me what it thinks the radar will look like for the next 7 hours in 15 mins increments.
I think that predicting the weather beyond a day gets pretty difficult because the weather is too chaotic. The best they can do is to gather data from local weather stations and see how often it rained when they had these same data in the past. So basically, it's like saying, "When the temp has been 85°F, humidity 62%, wind from WNW at 5 kts, pressure at 1016 mBar, and the date was July 29, it rained 6 times out of 10."
I notice the rain predictions are quite a bit more accurate in the cooler months. You can see a weather front traveling west to east as it comes across the country (I'm in US) and rains can last all day. During the hotter parts in summer rain clouds appear out of nowhere usually in the afternoon and rains are heavy but brief. This happened here in NE Ohio just an hour ago. There was no forecast for rain that I was aware of but suddenly we got doused for twenty minutes.
Weather Underground allows you to filter down to your zip code. Not accurate down to your exact vicinity, but better than the weather forecast for your entire city.
Some airports are still reporting temps once an hour, so the info you get from an app can be old and irrelevant.
Also some apps use a forecast made days before, they are too cheap/busy to update it.
Because we have tons of ground-level sensors, but not a lot in the upper layers of the atmosphere, I think?
Why is this important?
Weather processes are usually modelled as a set of differential equations, and you want to know the border conditions in order to solve them and obtain the state of the entire atmosphere. The atmosphere has two boundaries: the lower, which is the planet's surface, and the upper, which is where the atmosphere ends.
And since we don't seem to have a lot of data from the upper layers, it reduces the quality of all predictions.
Not really. Using % of forecast area as % chance of rain inherently gives equal weight to your position being anywhere within that area. Even if you limit the forecast area to the 5m or whatever it is radius that smartphone GPS is typically accurate to which a weather app could theoretically do, simply using % of the forecast area covered as % chance of rain inherently gives equal probability of you being literally anywhere within that 5m radius. It would obviously still be more accurate, but those numbers wouldn't be the same thing.
Not really. Using % of forecast area as % chance of rain inherently gives equal weight to your position being anywhere within that area.
Yes, unless your location is a statistical outlier the two are the same.
If you happen to know you are on the lee side of a mountain that might change thinks but for most people they are one and the same.
In Australia BOM's Australian Digital Forecast Database uses a 3x3 KM grid for Victoria and Tasmania or a 6x6 KM grid for the rest of the country.
I'm in a 6x6 area but thats fine for daily forcasts.They also offer forecasts for 3 hour windows for the next 72 hours which is great for medium term planning but to be honest its the rain radar that I use the most. They offer a rain radar that has a 90 minute history and a 90 minute forecast that has sufficient resolution that I can time my breaks at work to stay dry.
Since people are sharing their weather apps, I use Breezy Weather! Multiple sources, lots of info, FOSS, what's not to like. I tried multiple sources untill I found one that was the most accurate for me.