Objective Guidance for 0–24-Hour and 24–48-Hour Mesoscale Forecasts of Lake-Effect Snow Using CART

Abstract

Lake-effect-snow-possible (LESP) days were identified for each of 29 climatological stations in the lee of Lake Huron and Georgian Bay for the November-to-April winters of 1984?1988, using 0?24-h and 24?48-h wind and temperature forecasts from the Canadian Meteorological Center operational spectral numerical weather prediction model. Observed 24-h snow amounts on LESP days were separated into five ordered categories. A symmetric distribution centered on category 3 (>5?12.5 cm) was found for the aggregate of the 29 stations, but the distributions for individual stations peaked on category 1 (0-trace) or category 2 (>trace-5 cm). This suggests lake-effect snow occurrence, and amounts are likely to be overforecast for specific sites and small areas in forecasts issued for many public forecast regions, whose boundaries tend to be large. A recently developed nonparametric classification procedure known as ?Classification and Regression Trees (CART)? was used to find decision trees that classify the categorical snowfalls with threshold values of predictors in binary decision nodes. Predictors were designed from meteorological parameters known to be important in lake-effect snow formation, and were calculated from 0?24-h and 24?48-h forecast data from the NWP model on a 63-km interpolation grid. Verification with independent forecast data showed the CART forecasts to perform relatively well, considering the difficulty of the five-category mesoscale forecast problem. The best success was achieved with forecasts of snow not exceeding 5 cm. The success of forecasts for snow amounts greater than this for specific sites and small areas can be substantially increased when groups of forecasts for small areas are used. Predictors related to 1000-mb divergence were ranked among the most important predictors more frequently than any other types of predictors at nearly every station, followed by those related to air-water temperature difference. Some typical cases of 1000-mb divergence and accompanying snowfall patterns on LESP days are discussed.