We take the forecasts each model made one day ahead and compare them, hour by hour, with what the nearest suitable weather station actually measured: temperature, wind speed and daily precipitation. For each of these we work out every model’s average error, then compare it with the average of all the models. The model with the lowest combined score ranks first. The station is picked for being close and at a similar height to your point, and the forecasts are worked out for the height of the station, so it is a fair test.