The Internet of Things (IoT) paradigm is seeing a number of inexpensive devices connected to the internet. These devices are generating large amounts of data and are resulting in new classes of applications. With the ubiquity of cellular data channels, mobile devices are also generating large amounts of spatio-temporal data. We propose to enhance three areas of interest in spatio-temporal data analysis - imputation, causal inference and causal estimation from indirect measurements. The proposed approach is to investigate various machine learning models to extract information from spatio-temporal data. We propose to evaluate these models using fine-grained pollution measurement data from metropolitan areas.