Using a Model to Forecast and Why its so Difficult

Forecasting for weather is a tricky business. First of all, its not an exact science and probably never will be. There are just too many variables in the atmosphere. Some can be seen by the human eye but most cant. What we don't see, forecasting models try to see for us. However there are so many unknown variables in the atmosphere that it can be nearly impossible to paint a correct picture, but we do our best. Forecasting models take as many variables as it can see and paints a picture of what the current conditions look like for the entire United States. Then it tries drawing a forecast from by using them. But it doesn't run only once. Many of them will run a single forecast with the same variables several times and will get a different result every time. Then it takes the sum of all of those forecasts that it produced and averages them out to create one linear forecast.

Hopefully you can see how this would be a problem. Taking all of the accumulated forecasts and averaging them together removes all of the outliers like the extreme highs and extreme lows which can be correct from time to time.

For example, say for instance that there is a storm system that's headed toward Sioux Falls. This imaginary forecast model that we just completed runs 12 times before averaging out its forecast. Well, more then likely, all 12 model runs will come up with a different amount of rainfall that we are going to receive. So here are the twelve numbers; 0", .01", .02", .04", .05", .08", .16", .17", .19", .22", .29", and .86". So if you average out these 12 numbers, you get .17". So this model will tell us that Sioux Falls is going to get .17" inches of rain.

Now, what I hope you are thinking now is that there is a very large difference in those numbers. The model tried to average it out to give you the best possible result. However, what if the outlying model run was correct and we get .86"? Well then this is where meteorologists often make their mistakes. That's why we give probabilities on the chance of it raining. Because if we were to see all of those numbers and see that all of the model runs thought it would rain with the exception of one, then we might give an 80% chance of rainfall.

Well, we are having the same issue with the upcoming week. Some models are saying that we are going to see quite a bit of rainfall while others are saying that we wont see anything. If you look at the individual forecasting runs that were produced by each model (called ensemble members) they all look very different. This is what one such model looks like;

Each box represents an ensemble member. All are for the same period and give total amount of rainfall. Notice that many are very different. Then when the model itself sums up the numbers to output an average forecast for the world to see, it will give you something like this;I have shown something like this many times before. It just simply shows what the average amount of rain will fall by combining all of the ensemble members. This is where the meteorologist will come in and combine his or her knowledge to determine whether or not to trust this model and believe that its giving you an accurate forecast. So at this point, it looks like we will see some scattered showers by Sunday and Monday but we kept the probability of rain at just 30% because many of these models don't agree with each other. So stay tuned because model accuracy tends to get better as the storm gets closer so we will keep you informed with the latest!

~KDLT Meteorologist Cody Matz