Modelcasting has become a big issue in the meteorological world. Cell phone apps, websites, and even some forecasters simply look at what each individual run of the model shows them and makes that their forecast. This has led to large changes in some forecasts and led many of you to ask us why the forecast in your local area has gone from rain, to snow, back to rain and then back to snow 22 times in the past 5 days only for the sun to be shining the day of.
The truth is, just as our knowledge of what the weather will do is imperfect, a computer’s knowledge of what the weather will do is just as imperfect. A human forecaster can use logic to look at a map and realize that something doesn’t make sense. A computer has rigid equations and programs and follows them to the letter. It doesn’t matter to the computer if its showing the worst storm ever known to man, if that’s what the programs and equations say goes there, then the computer puts it there.
As trained forecasters, meteorologists are supposed to know better. We know the model has biases, that the model has errors. We are well aware that the model is just downright bad at forecasting certain storm evolutions. This brings me to newly formed Tropical Depression 8, probably better known to many of you as Investigation area 99L. A little while back in the comment section on an article about then Tropical Storm Franklin, I was asked about TD8 and how the Euro model was showing a stronger system than other models. I explained then that the GFS was very weak in forecasting the type of system that 99L was and continues to be, with even the 06Z run this morning continuing to show barely any development of TD8 as we’ll see down below.
Reality against the Machine
The image above is the model forecast for early this morning. The current NHC data shows that TD 8 has a minimum central pressure of 1011 mb while this model data shows a minimum pressure of 1014 mb. As you can see from the image, it also doesn’t show much convection.
This is where a meteorologist needs to know their models and how those models do and don’t work. We here at Firsthand have been looking at this system for Tropical development for days and first mentioned that the system might develop in this area in an article back on August 8th. We did this despite models like the GFS not showing any development because we know of the models weakness and added value to the model based on our own meteorological knowledge. This added value by the meteorologist is essential in any forecast. We head to the image below for another example.
This image, which is the model forecast for Monday afternoon, shows no major closed circulation. Based on the previous image, this image would even indicate that TD 8 is weakening. However, the forecast from the National Hurricane Center doesn’t call for weakening. The forecast expects that TD 8 will have been a Tropical Storm with winds between 50 and 60 mph. Do you see that Tropical Storm on this image? I certainly don’t. Modelcasting this storm would be a disaster if you happened to be on a cruise ship in that area.
Finally, this image shows what’s supposed to be a tropical system on Tuesday the 14th. The forecast during this time period calls for a strengthening Tropical Storm with the NHC forecast heading up to 65 mph and nearing hurricane strength. That type of forecast simply isn’t here on this model and it wouldn’t be on any app that uses this model for it’s forecast.
This same logic seen here with this forecast for TD 8 can be applied to many other scenarios. Whether its the development of a winter storm, a major outbreak of severe weather, or even something as simple as the morning temperature, The data that goes into and comes out of computer models allows us a much better idea of what the weather is going to do, but that data and the programming that goes into creating those models is only as good as our understanding of that data and programming.
Modelcasting with a computer can solve a lot more equations, but if the equation itself isn’t perfect then the answer to it isn’t going to be perfect either. The difference is that meteorologists who use the data as a tool, instead of just modelcasting and outright using the data, can understand what when the tool isn’t giving you the right answer.