Think of weather as a mixture of ingredients such as temperature, wind speed in direction, pressure, etc. These ingredients interact in certain ways according to the physical laws of the universe; the result is weather. If we knew all the ingredients at all times and all the ways they interact, we would be able to predict the weather perfectly. Obviously, we do not always know all the ingredients at all times, nor do we understand all the ways they interact. However, we know enough to translate some of the interactions into mathematical equations, such that when we input the ingredients for a specific time, we can get the idea of the weather. This is known as a computer model.
To create weather models, meteorologists divide the atmosphere into a three dimensional grid full of different coordinates. They enter “ingredients” for each of the coordinates, then use the mathematical equations to forecast weather for an area. These models are known as “dynamical,” which means that there are multiple inputs that evolve over time. A global dynamical model of the weather will take several hours to produce forecasts on the world’s fastest supercomputers.
There are many different existing computer models that meteorologists use. Some are global; some forecast for a particular area. Each model may have a different set of inputs or use slightly different equations. The best way to get the most reliable forecast using models is to use many models and combine their forecasts to create a “consensus” forecast.
Computer models are imperfect for three big reasons. First of all, as mentioned above, we don’t know all the equations for how “ingredients” interact. Second of all, also mentioned above, we don’t always have all the “ingredients”. While various pieces of weather equipment can give us a lot of data, we still don’t have information about all the weather everywhere, especially over oceans.
Lastly, resolution is a problem. Meteorologists split the atmosphere into a three dimensional grid full of coordinates, but these coordinates are often around 40 kilometers (25 miles) apart on a global scale. This means the model is less likely to predict small things such as thunder storms.