The Weather Effect Decomposition tool decomposes observed demand based on observed weather, seasonal normal weather and other causal variables. The result is a breakdown of demand into the portion that would have occurred under normal weather circumstances and that which occurred due to actual weather deviations from seasonal norms.
This process is powered by the same TESLA Model used for forecasting. To decompose weather effects, we run the model at least twice over the desired horizon. The first run uses observed weather data and results in a demand “backcast.” The deviations of the backcast from the observed demand are the TESLA Model residuals. The second run uses seasonally normal weather. We combine the residuals from the first run with the output of the second run to craft Weather Adjusted Demand. This is our determination of what the observed load would have been had the weather behaved according to seasonal norms.
The difference between the Weather Adjusted Demand and the observed demand is the aggregate weather effect.
We then run the TESLA Model a number of subsequent times, each time substituting seasonally normal weather for one of the six primary weather variables: temperature, humidity, cloud cover, wind speed, wind direction, and precipitation. This allows us to attribute portions of the aggregate weather effect to each of those six weather variables.