Tips

WxChallenge is a challenging competition, and forecasting skill largely comes with experience. However, there are some tips that will give you a head start:

• Know the geography of the area you are forecasting for and exactly where the weather station is. Geography heavily influences the weather. For example, a station located at the top of a hill will likely be cooler and than one in a valley during the day, but often also cools down less at night since valley locations tend to be more sheltered from the wind, allowing better radiational cooling, and cold dense air flows down the valley. Also, the weather station used for a city is usually at an airport and may not be in the city center. This is especially important when the geography is hilly. For example, the station for Binghamton, NY is at Binghamton Regional Airport, which is located 8 miles northwest, and more importantly, 750 feet higher than downtown Binghamton. With a maximum lapse rate of 5°F/1000 ft, this means the station can be up to 4°F cooler than downtown.
  

• Before the forecasting for a city starts, evaulate the performance of models for the past several days and get to know their biases. For example, if the models underestimated the high temperature on a sunny, dry day a few days ago, then they will likely underestimate the high temperature again on a future day with similar conditions. But this "analog method" of forecasting only works if the conditions between the two days are similar. Models could underestimate the high temperature on a sunny, dry day but could overestimate it on a cloudy, rainy day. Similarly, if the conditions are only half similar (like a partly cloudy day with a different wind direction and different air mass), use this method with caution. Weather.us contains a great archive of model output for about the past year. The Southeast Regional Climate Center links to daily weather statistics data based on the local weather office serving the station. If you don’t know which local weather office serves the station, enter the city’s name into the National Weather Service's home page.

• Use the models and your experience wisely. Consider the USL model into your forecast. It gives you the specific parameters the WxChallenge is looking for and is generally the most accurate model. However, it does have its flaws (such as overdoing the precipitation). The nowcasting models HRRR and RAP go out to 18 and 21 hours respectively and are generally quite good. However, when there is snow cover, they have a cold bias. Also, they don't cover the entire forecast period. Beyond 24 hours, GFS, ECMWF, and NAM 3-km are the best (beware of NAM 3-km's high precipitation bias though). For the best forecast, use the model output as tools and not as the actual forecast. Use basic meteorology principles and your intuition to decide which model to lean toward (if any) for which variables. Sometimes, you might want to take risks, such as forecasting warmer than all the models. However, only do so if you are confident, or you could bust your forecast hard.
  

• Make your forecast 30 minutes to 2 hours before the deadline, if possible. Making the forecast late ensures you have the latest available data and models, giving an edge to your forecast. For example, the GFS 18z run becomes available about 2 hours before the deadline. However, don't make it right before the deadline, to ensure margin of error for unexpected things (like computer crashes) and that you won't be rushed.
  

• Don't get caught off guard by untimely highs and lows! Don't assume that the temperatures will follow the "typical" diurnal cycle. Check the entire forecast period (06z-06z the next day). With fluctuating weather systems, it is not uncommon for the high to occur at 06z (when the forecast period begins) or the low to occur at 06z the next day (when the forecast period ends) or at other "atypical" times.
  

• Use caution when using modeled wind predictions. The WxChallenge wind variable you're forecasting for is the maximum 2-min sustained wind speed, which is inevitably higher than modeled winds because models (and even the National Weather Service) only show the average wind, which does not take into account the minute to minute spikes in wind speeds. For example, if the highest modeled sustained wind speed during the forecast period is 13 kt, the maximum 2-min sustained wind speed is probably between 18 and 22 kt. This is where the USL model really comes handy, because it actually gives you the predicted maximum 2-min sustained wind speed.
  

• After making your forecast, check to see how your forecast went. This feedback is crucial in improving your forecasting. If your forecast busted, try to figure out why it did. For example, maybe it was warmer than you expected because it was sunnier than you expected. You can then avoid repeating some of the same mistakes again. However, precipitation is trickier and more random -- busts are inevitable and are often luck-based, especially with convection.