Real-time prediction of the lake breeze on the western shore of Lake Michigan

A forecast verification study of the occurrence and inland penetration of t he lake breeze on the western shore of Lake Michigan was conducted. A real- time version of The Pennsylvania State University-National Center for Atmos pheric Research fifth-generation Mesoscale Model (MM5) with 5-km grid spaci ng was evaluated for a set of 71 (68) dates at 12-24-h (36-48-h) forecast r ange from summer 1998 and spring 1999. Contingency measures showed skill in predicting lake-breeze development at both day 1 and day 2 [Kuipers skill score (KSS) of 0.80 and 0.74. respectively]. These skill levels exceeded ca pabilities demonstrated by a simple lake-breeze index computed for the iden tical set of cases and also surpassed the published performance of the oper ational 29-km Eta Model for another site. Skill in the prediction of the in land penetration of the lake breeze (relative to a baseline defined by the sample average distance of the lake-breeze front from the shoreline) peaked at 2100 UTC at both day 1 and day 2 (KSS of 0.28 and 0.22, respectively) b ut fell thereafter due to a westward forecast bias. The origins of this bia s are tied in part to errors in the static model specification of the lake surface temperature, resulting from both errors in initialization and the o bserved diurnal cycle. Skill is also constrained by the sensitivity of the lake breeze to the prevailing synoptic flow; modest errors in forecasts of coast-normal winds can lead to substantial errors in the forecast position of the lake-breeze front by 0000 UTC, it is suggested that future research should focus on coupling a meteorological model to a dynamic lake model, im proved initialization of lake water temperatures, and further refinements o f the planetary boundary layer physics to improve near-surface winds. Such a model may be needed to allow reliable extension of forecasts out to times cales beyond 48 h.