Weather Nerd Corner 5 - Models

Originally sent via email October 18, 2019

Let's talk a little about models. I'm not going to go in to a lot of detail about the meteorology behind how they make a forecast. Instead I want to explain the process that happens each model run. There are many more steps than I am listing here, this is just a highly simplified overview.

Models do not just ingest data from observations every few hours, run some calculations, spit out a forecast, and then rinse and repeat. One of the main reasons for this is the huge lack of data, especially over the oceans. Data can also have errors. The first step in the model cycle is to produce a short range forecast, not the published one, but a first guess you could call it. The fundamental assumption of the model process is that the short range forecast is accurate. In other words we're assuming that we have a good model! This short range forecast or guess helps fill in the gaps where we don't have observations. This also helps keep some history from previous model runs so that the model isn't starting off fresh, it helps to maintain some run-to-run consistency.

Next, real weather observations are used to correct the short range forecast! This produces what is called an analysis, it is what the weather is thought to be like at time zero or the start of the model run. This is also called model initialization. From there the rest of the forecast is calculated out however far the model is programmed to do so and the results are published. The final step is that the next short range forecast is made to start off the next model cycle.

One thing to note about initialization is this can be a good way to judge how well a given model run may do. Meteorologists can look at the analysis/initialization and then compare it to what actually occurred at that same time by looking at radar or satellite data for example. If the model didn't do a good job initializing a system then you may not want to trust the forecast it produces for 24 hours from now.