GRAND RAPIDS, Mich. (WOOD) — This is such a cool viewer question, and I’m so glad someone asked it. Meteorologists use forecast models every day. There are dozens of forecast models, but we usually rely on about five daily here in America. One of the most accurate forecast models is the High Resolution Rapid Refresh model or the HRRR.

Back in December 2020, the HRRR underwent an upgrade. During the upgrade, a small error was made. The elevation of the Great Lakes bodies of water was mistakenly set to sea level. Lake Michigan’s elevation is 577 feet.

This error was rectified on July 19, 2022, but went unnoticed for more than two years!

What is the difference between models?

The HRRR is one of the most accurate models available to forecasters in the United States. It pulls more than two million data points at the surface and 50 data points every hour to make a fresh forecast. It has been specifically designed to handle thunderstorms, wildfire smoke, and lake-effect snow well.

The biggest catch to the HRRR model is that it can only forecast out to 36 hours four times a day.

Here is a comparison of the HRRR model on the left with the European model on the right. Notice how much more detail and accuracy there is in the HRRR output.

The European model is able to forecast out to 90 to 240 hours, which is one of its clear advantages over the HRRR.

Forecast models are all “built” a bit differently. The HRRR has the best spatial and temporal resolution. This is why it is typically quite good at showing weather trends in the short term. But because it can only forecast about a day and a half, other models like the European and the GFS are essential.

Processing forecasts take a lot of computer power. As our supercomputers get better, our models will all be improved. Each new model upgrade usually folds in a myriad of new weather research to help the forecasts perform even more accurately. That’s one reason meteorology has become much more reliable in the last two decades alone.

For short-term forecasts, the HRRR is unmatched. But if you want to look “far out” in time. It will come at the cost of resolution. For example, the GFS model plots forecast charts for all levels of the atmosphere a full 384 hours out, but it has a grid resolution of 8 miles instead of less than two.

Correction: GFS has a resolution closer to 8 miles but public data is exported as a larger resolution. For details on model upgrades in America through the years click here.

Did the error affect our forecasts?

The error of making the Great Lakes at sea level that programmers made was determined in post to have had fairly negligible effects. In short, it didn’t matter all that much.

We are able to determine that by looking at all the weather events over the Great Lakes that the HRRR forecast for from December 2020 through July 19, 2022, and check to see how well the model forecasted an event before it happened. This was done for big weather days, slow weather days and during all four seasons.

Surprisingly, the elevation snafu did not cause enough errors to have been noticeable. (This is courtesy of our local NWS office during the last meeting of the local American Meteorological Association and National Weather Association).

I had a director here at WOOD TV ask me if that made me trust the model less. Surely something so big as a 600ft difference in the terrain around the entire Great Lakes should make a difference, right?

Here is my take: No. It does not make me trust the HRRR model less.

I’m going to get into the weeds a bit with my answer why.

A slight change in the initial conditions having negligible effects is not all that rare in the forecasting world. In fact, every day, many of our forecast models are put into “ensemble” mode. In an ensemble, a single model is run many different times using slightly different initial conditions. A forecaster can then look at the ensemble and determine how solid a forecast is.

If a single change can completely derail the forecast, then the forecast must not have a very good certainty level. But if a ton of things are changing and all the forecasts continue to perform similarly, then it increases our confidence in what the model has to say.

Ensembles are used daily, especially with the GFS and European models. The elevation error with the HRRR having little impact actually increases my faith in its performance. It shows that even with a big error like an almost 600 feet sheer drop from land to water in its code, it still was able to accurately predict a whole host of weather in the Great Lakes. If anything, it makes me, as a meteorologist, trust the HRRR even more.