EQ research shows: Removing “bad” weather stations substantially improves the quality and stability of power consumption forecast systems.
In general, you need to find the correct weather stations if you want to succeed in modelling the weather-driven parts of the power market. Using a sophisticated model, but skimping on included stations, could even make matters worse. Even if you want to report something as simple as the average temperature in an area, this is a golden rule. If the weather stations are ill-chosen, the spatial-averaged temperature will be far off whatever you want it to represent. And the inherent error will be passed on to subsequent models that use temperature, or any weather parameters, as input.