Wednesday, 9 January 2013: 2:00 PM
Room 18A (Austin Convention Center)
One of the more challenging aspects of weather forecasting is predicting convective initiation. Just knowing whether an event will occur in 1-2 hours is challenging enough, let alone predicting its timing and placement. It is to this end that we employ a statistical tool called a random forest. Not only can the random forest produce a probabilistic forecast of whatever weather phenomenon it was trained to predict, it also produces a ranking of predictor importance that can serve as a template for simpler models. For this study, a random forest was trained to predict small-scale convective initiation along the Gulf of Mexico coast using SATellite Convection Analysis and Tracking (SATCAST) observations and model forecasts. Each CI object was tagged with satellite and model attributes, and the random forest learned how to best combine the attributes into a CI forecast. It then ranked the importance of each predictor. This new random forest model is then used to augment the SATCAST algorithm and to provide a basis for probabilistic SATCAST forecasts.
![](images/flag.png)
![](images/blue_ribbon.png)