There's a combination of physics models and statistics. And they're constantly updated with current measurements. They should get more accurate as the physics are modeled and measurements get more detailed. Of course we fired a lot of the people that do the modeling and shut down a bunch things doing the measurements.

No, because forecasts take what is currently happening and making a guess based on a probability about what can happen.

Anything beyond 3 days is unreliable

The models used aren’t just statistical models that predict the odds of the next weather pattern, temp, events based on historical data anymore, they’re more like actual world weather models like a video game where they render many different models all slightly different and they pick the most reliable, and the nearest outliers to assess the most likely weather. Climate trends changing as more and more energy is injected into the global model from climate change is easily coded into such models. It’s still chaos because climate and weather are affected by so many different variables all at once, that’s why having a slew of different models is useful and necessary. We are still unaware of all the variables, but have enough to be pretty accurate weeks out, but the further out you get, the less accurate the predictions

People have researched this.

IIRC weather forecasts have indeed gotten less reliable over the past decades, and there is correlation to climate change, but the effect is much smaller than one might think.

I think potholer54 did a video about it, long ago.

Numerical weather prediction models are not based on statistics, but on our physical understanding of the atmosphere. And physics is not going to change. On the contrary, our physical understanding of the atmosphere is what led us to understand climate change in the first place.

The further into the future you are trying to predict the weather, the more the reliability drops. The last time I heard a meteorologist talk about it, it was 5 days into the future for a specific location. Anything more than that was as reliable as reading tea leaves. You can say it'll rain in Asia 10 days from now but you cannot say with any certainty it'll rain in this village.

The climate which we ruined will provide us with punishment. Warmer oceans will give us stronger tropical storms. But they've gotten pretty good at detecting and outlining their likely paths. That sort of predicting is likely to improve with more storms and more data. Another suitable punishment is torrential train causing flash flooding. In the places with good satellite cover, they can see that and warn people with an hour's lead time or thereabouts. The time span is short but the prediction accuracy is good.

We as consumers are not good judges of the weatherpeople. We look at our phones and form an opinion. But we don't investigate where the data came from, which LLM may have added its hallucinations, and what other corners may have been cut to get us this information, which we don't pay for. So we see it's pissing down outside, the forecast was for blue skies, ergo the weatherpeople are getting worse at their job. When it is more likely the quality of the free data declined or somebody asked ChatGPT and Claude.