Couple things involved here: 1. There's actually not much to train the data on. Say you have 1 weather station (that is maintained and calibrated) in each major

Couple things involved here: 1. There's actually not much to train the data on. Say you have 1 weather station (that is maintained and calibrated) in each major or minor city. The GFS[1] has a resolution of ~18 miles. This is fine over densely populated areas, but oceans, forests, deserts? No data 2. Additionally you don't need to train a model, since most of the math is pretty well defined. All you need to do is initialize with latest data and let it crank. Where most of the development happens is in learning more about how the physical planet it set up, then adjusting your model to incorporate that newly known aspect 3. The weather forecasting market is actually rather robust, especially for an industry largely propped up by government grants. This probably isn't a great candidate for 10x disruption here and may not pass the "toothbrush test" either. [1] https://www.ncdc.noaa.gov/data-access/model-data/model-datasets/global-forcast-system-gfs https://www.ncdc.noaa.gov/data-access/model-data/model-datas...