Add weather observations

[JackKelly]

Detailed Description

Add weather observation data such as temperature, wind speed & direction, rainfall, humidity, and irradiance from sources such as:

• [ ] #80
• [ ] Ground-based meteorological station data (both "professional" met stations and amateur weather stations)
• [ ] Weather balloon data
• [ ] Weather observations from aircraft
• [ ] Weather observations from renewable energy sites
• [ ] Cloud-penetrating radar (both ground-mounted radar and satellite radar)
• [ ] #200

Some data sources are listed here: https://docs.google.com/document/d/1tk9cF4O539TzaMaUufn9Ay4f6qKKEyoNKmzP03kbSDo/edit#

Context

Ultimately, we're trying to teach ML models to forecast the weather better than numerical weather models (over time horizons of a few hours). To do that, our ML models need as much information as possible about the recent state of the atmosphere.

All of these weather observations are sparse in time and/or space. But that's fine! One of the awesome things about Perceiver IO is that it can handle sparse observations.

This definitely isn't something to prioritise for 2021, though!

[jacobbieker]

Somewhat related is the rainfall radar: #80

[JackKelly]

Good point! After your suggestion, I've added #80 to the tick-list above.