I use ML in weather research too. That's not the problem. The problem is Dark Sky used more simple radar prediction methods, like taking existing rainfall, using wind speed and 'nowcasting' the rain into the future.
There are severe limits to this technology. It cannot detect the initiation of new precipitation. They never publicly explained what they were doing with their barometer data, which they claimed to collect for forecasting. But in such an active area of basic science research, they did not contribute data or algorithms or papers or any discussion at all. I believe it is because they did not overcome the challenges of using the noisy phone data to produce useful weather forecasts. But they lead people to believe they do. I am suspicious.
Dark Sky didn't use ML AFAIK when they started, it was straight statistics. Maybe they did change to add ML, I don't know, but for my knowledge of the space, they always used core math and statistics while ignoring physical science and not participating in the research community, or any community really.
I would love to know more about how Dark Sky worked. My gut feeling is that it is largely a scam, because they never published any of their own reanalysis data, or verification data for 1-minute forecast accuracy, etc.
They were so closed-off in an industry that is traditionally hugely open, and they made large claims without backing any of it up publicly.
Aha, I want a lot from weather apps. I am a weather app developer, for what that's worth. I expect a lot because the current state of weather data communication and accuracy is a disaster. There is a lot of work to do to improve forecasts, and when a company steps up to say they will do it, I expect a lot out of them.
I also take major issue with the claim that they provided reasonable forecasts. I think their forecast accuracy was extremely low in many cases and fares very poorly when put against competitors: https://theoutline.com/post/3826/crap-weather-apps-meteorolg...
Edit: "The scam" is that they exaggerated accuracy claims without backing it up.
There are severe limits to this technology. It cannot detect the initiation of new precipitation. They never publicly explained what they were doing with their barometer data, which they claimed to collect for forecasting. But in such an active area of basic science research, they did not contribute data or algorithms or papers or any discussion at all. I believe it is because they did not overcome the challenges of using the noisy phone data to produce useful weather forecasts. But they lead people to believe they do. I am suspicious.
Dark Sky didn't use ML AFAIK when they started, it was straight statistics. Maybe they did change to add ML, I don't know, but for my knowledge of the space, they always used core math and statistics while ignoring physical science and not participating in the research community, or any community really.
I would love to know more about how Dark Sky worked. My gut feeling is that it is largely a scam, because they never published any of their own reanalysis data, or verification data for 1-minute forecast accuracy, etc.
They were so closed-off in an industry that is traditionally hugely open, and they made large claims without backing any of it up publicly.