When available, direct observations are the most accurate way to characterize atmospheric variables. However, the necessary observing network does not exist and would be impractical to build; therefore, models are used to fill in the temporal and spatial gaps. Models that synthesize weather data for use in power system analysis should ideally capture the physical and dynamical relationships between weather variables and produce weather states that are physically plausible, evolve realistically in time and space, and produce distributions of conditions like those that are observed in reality.