Modern forecasting is computational physics rather than pattern recognition, and its accuracy has improved dramatically.
Observation
Satellites, radiosondes, aircraft and surface stations feeding data.
Which establishes the starting state.
Numerical models
Equations of atmospheric physics solved on a grid.
Which is enormously computationally demanding.
Ensemble forecasting
Running the model many times with slightly different starting conditions.
Which produces probability rather than a single answer.
Why accuracy falls with time
Small initial errors growing.
Which is a property of the atmosphere rather than of the models.
Why forecasts got so much better
Three things improved together: satellite observation covering the whole atmosphere, computing power allowing finer model grids, and better methods for combining observations with model states.
Which has produced roughly a day of additional accurate lead time per decade over recent decades.
A five-day forecast today is about as accurate as a three-day forecast was a generation ago, which is a substantial and largely unremarked achievement.
Probability of precipitation
A percentage describing likelihood at a location.
Which is frequently misread as intensity or duration.
Ensemble spread
Disagreement between runs indicating uncertainty.
Which is genuinely useful information.
Local effects
Terrain and coastlines below model resolution.
Reading a forecast well
Treat it as probability rather than prediction.
What the percentages actually mean
A thirty percent chance of rain means that in similar situations it rained about three times in ten at that location.
Which is neither a prediction that it will rain nor a statement about how much.
People routinely read it as intensity or as coverage of the area, and forecasters have limited ability to correct that in a small icon.
Warnings
Issued on impact as well as likelihood.
Which is why a warning can be issued for a modest event in a vulnerable area.
Long-range outlooks
Seasonal signals rather than day-by-day forecasts.
Which are frequently reported as though they were the latter.
App differences
Different apps using different models and post-processing.
Where to get the best information
National meteorological services, whose forecasts and warnings are authoritative and free.
Why forecasts still get things wrong
The atmosphere amplifies small differences, so tiny errors in the starting conditions become large errors in the forecast within days.
Which is a fundamental property rather than a limitation that better computers will remove.
Ensemble forecasting exists to quantify that uncertainty rather than to eliminate it, and a forecast presented without any indication of confidence is throwing away useful information.
Nowcasting
Very short-range forecasting from radar and satellite.
Which is what rain-in-the-next-hour features use.
Climate against weather
Long-term statistical behaviour against individual events.
Which are different questions with different methods.
Severe weather warnings
Worth acting on rather than waiting for certainty.
Free and authoritative
National meteorological services publishing forecasts, warnings and explanation.
Why it is worth knowing how these things work
Most of the systems that shape ordinary life are invisible by design. Nobody explains why a parcel took an odd route, why a film left a streaming service, why a wait extended, or why a price changed between two searches.
The absence of explanation is rarely deliberate concealment. It is that the people running these systems are solving their own problems, and the reasoning behind their decisions is obvious to them and completely opaque from outside.
Understanding the mechanism does not always change what you can do about it. It does remove a category of low-grade frustration that comes from assuming something is arbitrary, unfair or aimed at you personally when it is usually none of those things.
The pattern that recurs
Across almost all of these systems, the same three things turn out to be true. The behaviour that looks irrational from outside is optimising for something the observer cannot see. The cost that seems unexplained is usually concentrated in one specific stage of the process. And the information that would resolve the confusion is generally published somewhere and read by nobody.
That last point is the most useful. Regulators, operators and industry bodies publish an enormous amount of explanatory material that answers exactly the questions people complain about not being able to get answers to. It is dry, it is not promoted, and it is free.
A note on sources
Where regulation is involved, national regulators publish the actual rules and they are generally clearer than press coverage of them. Where an industry is involved, trade publications aimed at people working in it are considerably more informative than consumer coverage.
Practices described here vary substantially between countries, and anything with legal, financial or medical consequences warrants checking against the rules that apply where you are.
The pattern that recurs
Across almost all of these systems, the same three things turn out to be true. The behaviour that looks irrational from outside is optimising for something the observer cannot see. The cost that seems unexplained is usually concentrated in one specific stage of the process. And the information that would resolve the confusion is generally published somewhere and read by nobody.
That last point is the most useful. Regulators, operators and industry bodies publish an enormous amount of explanatory material that answers exactly the questions people complain about not being able to get answers to. It is dry, it is not promoted, and it is free.