The urban heat island (UHI) of Charlotte (North Carolina) - a rapidly expanding subtropical city - was studied through statistical analysis and short-term predictive modeling. Our study used hourly surface observations taken from 12 weather stations over a 5-year period. One station was identified as the least urban and used as the rural reference. Two stations were deemed the most urban and were used for UHI analysis and development of a UHI predictive. Observations from the two urban stations were combined to account for dissimilar parameters observed by each station.Previous work has shown that the daily UHI maximum (4-10 K) often occurs during nocturnal hours when optimal weather conditions (i.e., clear skies, light winds, low humidity, and strong static stability) are present. This study examines whether such nocturnal UHI maximum can be predicted from daytime weather parameters observed 6-9 hours prior. Using daytime weather parameters, two statistical model types (generalized linear models and bootstrap random forests) were evaluated for predicting the nocturnal UHI magnitude. The bootstrap random forest models were found to out-perform the generalized linear models. Stratification of the data by season and day of the week further improved random forest models.