Forecasts combinations have shown improved prediction accuracy in many
situations than individual forecasts. In this work, we have used five linear
regression methods (multiple least squares regression and ridge regression with
or without variable selection, principle component regression) to conduct
temperature forecasts combinations based on forecasts made by four websites. It
is found that these methods can outperform most of the websites. In certain
cases, the forecast combination methods have been found to be capable of
outperforming any single website prediction or the average of these predictions.