A Simple Statistical Model for Predicting Fine Scale Spatial Temperature Variability in Urban Settings

Given that mortality rates during a heat wave are a sensitive function of temperature, forecast maps of temperature anomalies within cities should be useful to the health community. An empirically based approach for predicting daily spatial variations in the Urban Heat Island has been developed for New York City. Our technique is derived from two data sets: high spatial resolution temperature data collected by multiple synchronized traverses of Manhattan by foot; and several months of high temporal resolution data collected at 10 fixed locations by instruments mounted on lamp posts. The high spatial resolution data is regressed against local characteristics such as vegetation, albedo and building height to produce a statistical model of relative temperature anomalies. The fixed instruments show local temporal variability attributed to convection, and spatial variability between instruments attributed to local surface characteristics. The magnitudes of both types of variability are regressed against weather conditions such as cloud cover, wind speed, lapse rate and humidity. When applied to the average spatial anomaly map, the amplitude of the temperature variations within the city each day can be predicted based on a weather forecast. A working model should be online by late fall of 2014, predicting temperature variations within the city 24 hours in advance. The technique should be easily portable to other cities.