You may hear us raving or complaining about a computer model or models during our weathercasts. So what are they?
Computer models help us forecast the weather. We take all available information from surface observation stations, weather balloons, information from hurricane hunters. The new thing added to this list is satellite observations and estimates of clouds and precipitation or remote sensing. A lot of that though is still experimental at this time.
Once the data is collected it is then put into complex mathematical equations. This is where calculus comes in to play. By using math we can predict the future (we don't use fortune tellers or dart boards). Pretty neat, eh? Computers then crunch the numbers and spit out intial conditions and a forecast. One issue with computer models is that we don't and won't have observations for every point on earth. So, we have to estimate the conditions between the actual observations. Sometimes computer models get the initial conditions wrong. When this happens it can throw off the entire forecast by that model. Most of the time though, the initial conditions are pretty good. Each model uses different physics or conditions (thus different math equations), so this is where we may tell you "this model is saying blizzard and this one is saying flurries."
One way to try to find a solution to the forecast is to use "Ensemble Forecasting". This is simply when a meteorologist looks at several different models, or even different forms of the same model. If one were to look at all of these models at once, it may look like spaghetti on a map. The closer the lines are to each other the more confidence one has in the forecast. The more separated or spaghetti-like, the less confidence one has in the forecast.
One example was the first weekend of August. A few days before the weekend, the main models were indicating a mini heat wave, forecasting highs in the upper 80s and lower 90s. The other forms of those models were saying lower 70s for highs. That's quite a difference if you ask me. I ended up going 8-10 degrees cooler than the main models as a result, splitting the difference between the models. By using this approach we can actually forecast with more accuracy.
If you would like to know more about ensemble forecasting click here. It may be a bit technical in spots, but if you scroll through the website, you'll see that spaghetti map I was talking about above. As always if you have any comments or questions email us at email@example.com.