Probabilistic freeze forecasting in the Midwest and Florida

Weather forecasts have a profound impact on many economic areas, including agriculture. Extreme weather events, such as freezes, have been known to vastly alter the agricultural sector in terms of commodity pricing, forecast implementation, and crop damage prevention. Although preventative measures and insurance have countered some of the economic loss during these disasters, weather forecasts have become a primary player in determining the cost and effectiveness of these actions. In order to specifically focus this research, qualitative analysis was a primary for data in order to compare and correlate results. Freezes in Florida (to the orange crop) and in the Midwest (to the winter wheat crop) could be analyzed using a high resolution set of surface observations, as well as providing upper air data, pressure/temperature anomalies, soil moisture data, and climatology. Analyzed cases will also be tested for recent temperature anomalies and persistence. Some attributes often linked with freeze events include: low height/thickness values, light to calm winds, clear skies, low dewpoints, low soil moisture, snow cover, a minimum in daylight hours, and a low specific heat and thermal conductivity of the surface, a positive PNA, negative NAO, negative AO and neutral ENSO and Sudden Stratospheric Warming (SSW) events. The surface data (upper air data) will be analyzed for 19 (2) stations in Florida and 39 (4) stations in the Midwest for both freeze events and null cases (associated with cold air outbreaks that failed to meet a standard criteria for a “freeze”). Freeze lengths and severity, through this and past research, will be taken into account to decipher between freeze cases and null cases. The highest correlated variables will be put into an equation that will give freeze probabilities for future events through modeling efforts. Given a climatological based initial freeze probability, the parameters studied will increase or decrease that probability for the model. Timing and weighting of these various parameters may be different between the two locations, thus modeling will be performed separate between the two locations. Since significant freeze events take place at different times of the year for the two locations, the model will switch domains rather than run simultaneously. Weighting variability will also contribute towards a sensitivity ensemble part of the model to test different scenarios and aid in forecasting.