5 ms·

I feel like this app gives the illusion of higher-fidelity data than it can really provide. It gives a temperature reading for any pixel I select, but appears t

I feel like this app gives the illusion of higher-fidelity data than it can really provide. It gives a temperature reading for any pixel I select, but appears to be interpolating from a relatively limited number of real-world data points. I live in the flatlands near some mountains. I can look at one of the websites that show data from personal weather stations to see that the temperature is 15 degrees cooler up on the mountains. But windy shows the temp there being the same as here. I figure it’s just averaging the various locations around (but not on) the mountains. Better data exists, but they don’t seem to have it.

cconcepts 5y ago
I agree, was recently using it in a remote area for reasonably important decision making. The high level of detail fooled me into thinking it had a high level of accuracy. Turns out being very specifically wrong is more convincing than just being generally wrong. EDIT: I should add that windy may be the best forecasting app there is. I was more commenting on the fact that the specificity made me follow it blindly when I should have been treating it like any other forecast.
rhodysurf 5y ago
I work in this field and this is exactly true
- LeoPanthera 5y ago
  Can you elaborate?
nkurz 5y ago
> It gives a temperature reading for any pixel I select, but appears to be interpolating from a relatively limited number of real-world data points. I think you are misinterpreting what Windy is showing. It's not trying to show the current temperature at the selected location, it's showing the current forecast for the temperature at that location according to a particular model. On the lower right, you can choose from different forecast models, and the "temperature" at the location you select will probably change. If you choose a finer grained model than the default (there are different third party models available for different locales) you might find one that does a better job of dealing with mountains. And on the lower left, you can change the "time" slider to show the forecast for the next few days according to the chosen model. So yes, there is interpolation happening, but it's interpolation from a model that is already based on an artificial grid, and not based on interpolating from the actual temperatures as recorded on the ground. What I find useful about Windy is the ability to compare the output of different weather models, and thus get a idea of the uncertainty of the forecast. To access this, choose a location, and then click "Compare" on the bottom bar. If all 5 models available to me agree, it's probably a good guess. If one of them is radically different, one should be slightly skeptical. If they all disagree, probably best to assume that the forecast is inherently unpredictable.
deepsun 5y ago
The truth is, all the weather in the world is interpolated. Most people don't know that new _forecasts_ are usually made only twice a day (at 00:00Z and 12:00Z) (more frequent _forecast updates_ are actually just changing times in forecasted ranges). That's why pilots don't like Windy actually. We prefer to read source readings and forecasts and guess-interpolate manually, no need in extra layer of complexity. E.g. if I want to check visibility at my small airport, where no weather forecasts are made, I check a few larger airports around, and also learn today's "big picture" weather state (fronts? inversions?), to estimate what should I expect at my local airport. It's like debugging a website software having raw DB data vs. screenshot.

cconcepts 5y ago

I agree, was recently using it in a remote area for reasonably important decision making. The high level of detail fooled me into thinking it had a high level of accuracy. Turns out being very specifically wrong is more convincing than just being generally wrong. EDIT: I should add that windy may be the best forecasting app there is. I was more commenting on the fact that the specificity made me follow it blindly when I should have been treating it like any other forecast.

rhodysurf 5y ago

I work in this field and this is exactly true

LeoPanthera 5y ago
Can you elaborate?

nkurz 5y ago

> It gives a temperature reading for any pixel I select, but appears to be interpolating from a relatively limited number of real-world data points. I think you are misinterpreting what Windy is showing. It's not trying to show the current temperature at the selected location, it's showing the current forecast for the temperature at that location according to a particular model. On the lower right, you can choose from different forecast models, and the "temperature" at the location you select will probably change. If you choose a finer grained model than the default (there are different third party models available for different locales) you might find one that does a better job of dealing with mountains. And on the lower left, you can change the "time" slider to show the forecast for the next few days according to the chosen model. So yes, there is interpolation happening, but it's interpolation from a model that is already based on an artificial grid, and not based on interpolating from the actual temperatures as recorded on the ground. What I find useful about Windy is the ability to compare the output of different weather models, and thus get a idea of the uncertainty of the forecast. To access this, choose a location, and then click "Compare" on the bottom bar. If all 5 models available to me agree, it's probably a good guess. If one of them is radically different, one should be slightly skeptical. If they all disagree, probably best to assume that the forecast is inherently unpredictable.

deepsun 5y ago

The truth is, all the weather in the world is interpolated. Most people don't know that new _forecasts_ are usually made only twice a day (at 00:00Z and 12:00Z) (more frequent _forecast updates_ are actually just changing times in forecasted ranges). That's why pilots don't like Windy actually. We prefer to read source readings and forecasts and guess-interpolate manually, no need in extra layer of complexity. E.g. if I want to check visibility at my small airport, where no weather forecasts are made, I check a few larger airports around, and also learn today's "big picture" weather state (fronts? inversions?), to estimate what should I expect at my local airport. It's like debugging a website software having raw DB data vs. screenshot.