Currently, a statistic (mean / std / sum) of each feature is taken from a circle of n radius (10km, 20km) around the lat/lon of the power plant.
This doesn't account for wind, and has no (endogenous) method of considering background noise (i.e. car pollution) besides also including those as features.
What if we take observations at 2 (3?) points... Using daily weather data can't we take an observation 10km upwind (i.e. a sample of background noise) and 10km downwind (i.e. where pollution from the powerplant should have been blown, and difference them.
Advanced version: rather than just using a circle or a point upwind/downwind, what about an 'ice cream cone' stretching back to the powerplant.
Currently, a statistic (mean / std / sum) of each feature is taken from a circle of n radius (10km, 20km) around the lat/lon of the power plant.
This doesn't account for wind, and has no (endogenous) method of considering background noise (i.e. car pollution) besides also including those as features.
What if we take observations at 2 (3?) points... Using daily weather data can't we take an observation 10km upwind (i.e. a sample of background noise) and 10km downwind (i.e. where pollution from the powerplant should have been blown, and difference them.
Advanced version: rather than just using a circle or a point upwind/downwind, what about an 'ice cream cone' stretching back to the powerplant.