This month’s Splitblog topic comes from our Head of Development: Bartosz. Those who know him know that he is a human Swiss Army knife, and as such, he is not only well-versed in meteorology but also a talented writer. That’s why he naturally took the opportunity to write the text on his chosen topic himself. But read for yourselves!
Hey Siri, how’s the weather?
Seemingly a simple question. But also one that appears not so easy to answer. And upon closer inspection, it only becomes apparent what is truly needed to determine whether it’s worth taking an umbrella or not.
First and foremost, we need to collect data. And a lot of it. Across the entire planet, there are countless measuring stations that record temperature, air pressure & humidity, and wind direction and speed. Additionally, we use data from weather balloons, airplanes, ships, and satellites.
To determine the weather from this, we also need a weather model. This is not a physical model, but rather a mathematical description of our weather through a multitude of equations and parameters into which we feed our measurement data to see how the weather will develop.
Parameters? What kind of parameters again? Well… that’s a lot of data and a lot of equations, and to calculate that, we previously needed extremely powerful supercomputers, and even they couldn’t complete this task quickly enough, because: what good is the best forecast if it’s only ready after the event has already occurred?
So we start by simplifying things, for example: what happens near the ground? Or at the transition from water to land? The complicated processes are significantly simplified, making them a little less precise, but also calculable within a foreseeable timeframe.
Enough about weather services! After all, this is about AI. Or is there a connection?
Yes, there is. For some time now, there have been very exciting AI approaches that promise weather forecasts without being quite as computationally and time-intensive. Last year, Google’s AI research division DeepMind released the GraphCast model, which, trained on historical data, requires only a fraction of the computational power of numerical (i.e., the previously described) weather models, and can thus, for example, deliver a 10-day forecast in under a minute. And GraphCast appears to be not only faster but also – at least in part – more precise than numerical weather models, and for example, it predicted earlier than numerical models where Hurricane Lee would likely make landfall.
And so it is hardly surprising that all major weather services are now experimenting with AI approaches, including the German Weather Service (DWD), which even announced a breakthrough in AI-supported weather forecasting just a few days ago in a press release with their newly developed model AI-Var.
Given the pace at which the entire field of artificial intelligence is currently developing, it remains very exciting to see what the future holds for us – also in meteorological terms. And until then, perhaps we can be a little more lenient regarding our meteorologists and their forecasts, as we have seen that it is not at all uncomplicated.
Sources:https://www.dwd.de/DE/presse/pressemitteilungen/DE/2024/20240617_pm_ki_news.html
https://www.spektrum.de/news/graphcast-neues-ki-modell-soll-genauere-wettervorhersage-liefern/2198859
