This 10 part series will detail forecasting tricks that can be used to try to outforecast MOS. Outforecasting MOS is an important skill for a forecaster. MOS stands for Model Output Statistics and they are used as a guide for temperature prediction and precipitation prediction by forecasters. Model consensus is the average of the high temperatures, low temperatures or precipitation amount predicted by several forecast models.


Forecast models have a reputation for being wrong and that they can be outforecasted by an experienced forecaster. I have to say though these models are getting so good that it is getting more and more difficult to beat them especially in the tomorrow forecast. It used to be the NGM was the primary source for MOS. With several good models now it is getting more difficult to outforecast a consensus of these models.

Model consensus is the average forecast of several models. If 4 models forecast a low temperature tomorrow of 25, 29, 23 and 24 then the model consensus is (25+29+23+24) / 4 = 25 F.

If the model consensus is converging to a similar value than that is an indication the models have a very good handle on the forecast. If 4 models forecast a high temperature tomorrow of 50, 50, 50 and 51 then the models are all very near the consensus forecast. Could all the models be wrong? Yes, but you better have a good reason for why they would all be wrong.

You may run across a situation where 4 models are split on agreement. For example the predicted low tonight from the models could be 36, 38, 45, and 46. The model consensus would be (36+38+45+46) / 4 = 41 F. As you forecaster you should also determine why the agreement is split. Perhaps it is a cloud cover issue, there is a significant temperature gradient nearby or it is a precipitation issue (evaporative cooling). If you are fairly certain which of the splits will happen then go for that consensus. For example, if you are thinking the 2 cooler values are more likely, then the consensus for those will be (36+38)/2 = 37 F.

If the forecast values are wide ranging then it is more difficult to trust the consensus. As a forecaster you will need to spend extra time with this forecast to determine your prediction. For example, the model highs tomorrow could be 55, 61, 64 and 68. Look at the model panels and determine why there is so much of a spread. As yourself which high temperature is most likely. You may conclude a consensus is good or you may want to go with a higher or lower value depending on the situation. You may also conclude you need to take your own value and not the value given by any model.

There are cases where it is not a good idea to take consensus. The 8 previous forecast trick essays explain various situations in which it would not be the best idea to take model consensus. However, on many forecasts the taking of model consensus is a good idea. The situations in which taking model consensus is a good idea is when:
1) there is a stagnant air mass over the forecast area
2) all the models have a common consensus
3) there are no clouds
4) the weather pattern is unchanging
5) the weather is typical to what should occur that time of the year

The model output will tend to be more varied when the weather is changing. It is not a good idea to go blindly with model consensus when:
1) a front is moving through the forecast area
2) differential advection is occurring
3) all the models do not have a common forecasted value
4) there is a good chance of clouds and precipitation
5) the weather is unusual for that time of the year
6) there is a mesoscale process the models are not picking up well

Keep all this information in mind when putting together your temperature forecast.