and can we expect it to get any better?
yes
yes
Living in England, where the weather is always hard to predict, my impression is that the short term forecast, ie one to three days, is fairly reliable. That said they do use a lot of weasel words like ‘sunny with a chance of rain’ or ‘cloudy with sunny spells’. They are currently telling us to expect the tail end of Hurricane Gonzalo on Tuesday.
It has gotten vastly better, at least in the US. The validity of 7 day forecasts today are better than 3-day forecasts were 15 years ago.
They will continue to get better if we fund sufficient satellites and ground stations to gather more data on a tighter mesh. If we decide that’s too expensive, then we’ll have an ever decreasing source of raw data as ground equipment ages & satellites fail. Better computer models will offset some of the decline, but not all.
A lot of people’s dissatisfaction with weather forecasting is based on simply not understanding what’s being forecast. e.g. within a major metro area we can say with pretty good certainty whether ort not rain will fall somewhere within the area the day after tomorrow and about howmuch will fall. We can’t say exactly when or whether you personally will happen to be standing outside in the place it happens at the time it happens. And that latter thing is what folks desire: a personal forecast for their expected location and activity trajectory over the next few days.
The weather forecast on the BBC was different in the past:
OK, it’s seven years ago, but still: Are weathermen checked for forecast accuracy? Have advances in technology improved the accuracy of weather forecasting? - The Straight Dope
Thanks for that link
We’re getting better at providing this for the very short term, though. My smartphone can now tell me with pretty good accuracy when it will be raining (and how hard) during the next hour or so.
They should be more accurate as of a few weeks ago; NOAA came out with their High Resolution Rapid Refresh model that is both quite a bit more granular (3km blocks) and more often (every 15 minutes) than the prior model’s 13 km / 1 hour resolution.
You don’t need your smartphone for that - just go outside for a couple of minutes and you can make the same prediction with the same accuracy using your own senses. 
I think online weather forecasts are a bit better than that high-paid eye-candy “meteorologist” they have down at the local TV station. Weather dot com and Weather Underground seem to get it right, down to the hour, at least for the next 3 days or so. With everyone using the same source data, I wonder why there can be such a wide range of interpretation?
Haha
No
I’m always amazed at how well they do. The fact that they even come close on 10-day forecast, I find incredible!
I wonder if crowd sourcing of data from jillions of cell-phones could someday yield that tighter data mesh mentioned above.
They’d want to gather temp, humidity, pressure, wind speed and direction, and whether or not it’s precipitating.
Between phones in cars, phones in pockets, phones in purses, phones in air conditioned restaurants, phones laying in the sun, etc., it’s going to be hard to obtain calibrated consistent data from all of them. Hard enough just to calibrate all the devices, much less correct for the micro-environment they’re in.
So I doubt that application will go very far.
I don’t want to start an argument, but I’d like to ask a question?
If we can’t rely weather predictions up to about 3 days ahead.
How can we have climate models that can predict up to 5 or 10 years in the
future?
The advance climate models incorporate a lot more data than your typical three day predictions.
They could make incredibly accurate predictions for a 3 day report by imputing all available data, but gathering and imputing the data from all available sources and modeling it into a prediction would take longer than the three days. No one really wants to know what the weather will have been 2 weeks after it’s already passed.
3 Day forecasts are cliffs notes of the data they have readily available.
With multi-year climate models they have the luxury of time to compute them. They also don’t need to predict hour by hour reports, they are looking for what the climate expectation will be in a given year.
Weather forecasting has gotten much better but I’m not sure how well the communication of it has improved. There is an incredible amount of science, thinking and understanding about a very complex and fluid situation that is usually displayed to the end user as a little icon that says “30% chance of rain.” Then the person looks at it 3 hours later and it says 50% chance of rain, then it never rains so they think the weatherman is nuts.
If you are really interested in weather, you can read the Area Forecast Discussion from your closest NWS office to see the pros talking about many of the variables that are going into the forecast.
Because the climat model is not trying to predict whether it will rain in the morning on October 22, 2019. It’s trying to predict more or less how hot October 2019 will be overall, and about how much rain there will be the range from September 2019 through November 2019.
Well, people are certainly working on it, and they address this question often, here it is addressed in a guest blog post from one of the groups: http://blogs.scientificamerican.com/guest-blog/2013/10/11/weathersignal-big-data-meets-forecasting/
And there are others:
http://opensignal.com/reports/battery-temperature-weather/
Most of them are simply operating as an experiment, to go ahead collect the data and see if there are correlations or useful information that can be gleaned from it. And it appears that there is indeed some use for it.
As for ultra-personal forecasting, I too use an app that uses my location to tell me something like “Thunderstorm, 5mi SE, rain starting in 10 minutes” – and it’s often crazily accurate. As mentioned above, it is just for the very short-term, but in the near future almost anything is possible… Considering Google Now knows my routines and movements well enough to give me travel times to the places I usually go, on the days I usually do it, hours before I even think about leaving…there’s no reason that couldn’t combine with ultra-local forecasting to tell me that my favorite bike trail has a higher chance of rain this weekend than another one, or that overnight temps at the campground I’m registered at will be in the low 40s, etc.
Concur. Several times when I was deciding to run an errand by bicycle or car, the sky was gray as a battleship and it was surely going to downpour any minute, but the “minutecast” said it wasn’t going to rain for 30 minutes (or however long the errand would take), so I took a chance and went by bike. Every time but once, it didn’t rain, and that time it started drizzling as I arrived home. 