Elevated urban temperatures are a significant concern across the globe due to their negative health effects and increased energy use. Understanding the spatial variation in urban air temperatures can lead to informed mitigation and planning efforts. Air temperatures for multiple urban areas in the state of Iowa, USA, at three times of the day, were collected using customized sensors mounted on vehicles driven through a variety of landscapes in each urban area. Geographic information systems technology was used to process high-resolution landscape datasets and derive variables that summarize the urban landscape surrounding each temperature measurement point. Five different statistical models: standard regression, trend surface, geostatistical, time series, and random forest, were fitted to nighttime data in the Waterloo–Cedar Falls urban area. We demonstrate that the best method for predicting Waterloo–Cedar Falls nighttime data is to use Waterloo–Cedar Falls data collected at a different time of day. However, when data are not available in the same city for which predicted air temperatures are needed, we explore which substitute city’s data best forecast the target city’s air temperature, via four cross-validation strategies. We find that, when predicting evening and nighttime air temperatures for the Iowa urban areas, choosing the closest-in-population-size substitute city provides the best predicted air temperatures.