This paper introduces a novel technique to measure indoor ambient air temperature using the battery temperature sensor found on typical smartphones. We develop physics-based models to predict ambient air temperature that consider the many warming and cooling scenarios faced by phones and account for the excess heat generated by smartphone components such as the CPU, screen, network, and charging hardware. To accommodate never-beforeseen devices, we also develop a domain adaptation technique that leverages previously derived models, substantially reducing the overhead of learning accurate models for a new phone. We evaluate our models for a range of devices, operating scenarios, ambient temperatures, and phone cases, with mean errors generally less than 1.5% of ambient temperature. We also present a case study to demonstrate the utility of our approach for spatial and temporal monitoring of ambient temperature variations in an office building; while indoor conditions vary by as much as 13°F, mean error in measurement by our models is 1.4%.