Private forecasting companies have traditionally worked with available numerical weather prediction model data from governments around the world (such as the United States and the European Union) to make weather forecasts for their clients. They have typically used methods such as experience, local knowledge, and customization for clients to try to make their forecasts more accurate than competitors’ forecasts or tailor them to their clients’ needs.
Some start-up companies, however, are now attempting to contribute to forecast accuracy in more direct ways – obtaining private observations themselves and using internally-developed algorithms that theoretically should help forecast accuracy by giving them a leg up on the competition.
Saildrone, for example, uses drones that looks like sailboats and are powered by wind and the sun, dispersed across the ocean, to collect data where little data currently exists. Salient Predictions is attempting to improve seasonal forecasts by using machine learning. Other startups explore using private data points or private satellites in an attempt to improve their forecasts.
While forecasts from these companies aren’t yet significantly better than public forecasts and those from current numerical weather prediction models, their methods to get them there are theoretically on the right track for better predictions of extreme weather in particular. Perhaps in the near future, they will be able to generally predict events like this January’s train of atmospheric rivers in California months in advance – or at least, for example, be able to get a sense of abnormally wet weather coming in the middle of a decades-long drought.
For this information and more, see Pranshu Verma’s article in the Washington Post.
ForecastWatch evaluates accuracy of temperature, precipitation, and wind forecasts. Our data offers the ability to request customizable collections – by location, time period, and forecast parameters, such as wind and temperature – and have them conveniently delivered as easy-to-use CSV files. Contact us for more information.
March 6, 2023 at 1:58 pm
I am skeptical that any of the private forecasts are any better than deterministic GFS, on average. It is much easier to degrade a physics-based model with “secret sauce” than it is to improve it, on average. You can tune for certain locations, stations, and time horizons to impress potential clients. But you can’t beat science, on average.
I’ve seen some pretty egregious verification errors in one well-know private forecast. But the terms of service prevent you from stating these errors publicly to protect the brand. Verification is critical. Keep up the good work.
Comments are closed.