Why did not they go with a simple approach:
* Device with a low power thermal sensor, cellular connection and a battery that sends a heart beat with the current temperature once every minute or so. Maybe even BLE mesh network of such sensors. Simple patterns like a sudden increase in the are or sensors going offline can indicate something is happening.
* Does it really need a live video? Time-lapse would require less traffic and energy. Many camera's seems to be attached to the existing poles or lines and potentially limiting coverage area.
* I think a modern approach for such projects should be: put streams from sensors of different types online + an archive of data and have it as an ongoing kaggle-like competition for the best approach with the condition that resulting model should be made available to CalFire for their own use.