Recent years have witnessed the explosion of spatiotemporal big data (e.g. GPS trajectories, vehicle engine measurements, remote sensing imagery, and geotagged tweets) which has a potential to transform our societies. Terabytes of earth observation data are collected every day from thousands of places across the world. Modern vehicles are increasingly equipped with rich sensors that measure hundreds of engine variables (e.g., emissions, fuel consumption, speed, etc) annotated with timestamps and location data for every second of the vehicle’s trip. According to reports by McKinsey and Cisco, leveraging such data is potentially worth hundreds of billions of dollars annually in fuel savings. Spatiotemporal big data are also enabling many modern technologies such as on-demand transportation (e.g. Uber, Lyft). Today, the on-demand economy attracts millions of consumers annually and over $50 billion in spending. Even more growth is expected with the emergence of self-driving cars. However, spatiotemporal big data are of volume, velocity, variety, and veracity that exceed the capability of common spatiotemporal data analytic techniques. My thesis investigates spatiotemporal big data analytics that address the volume and velocity challenges of spatiotemporal big data in the context of novel applications in transportation and engine science, future mobility, and the on-demand economy. The thesis proposes scalable algorithms for mining “Non-compliant Window Co-occurrence Patterns”, which allow the discovery of correlations in spatiotemporal big data with a large number of variables. Novel upper bounds were introduced for a statistical interest measure of association to efficiently prune uninteresting candidate patterns. Case studies with real world engine data demonstrated the ability of the proposed approaches to discover patterns which are of interest to engine scientists. To address the high velocity challenge, the thesis explored online optimization heuristics for matching supply and demand in an on-demand spatial service broker. The proposed algorithms maximize the matching size while also maintaining a balanced provider utilization to ensure robustness against variations in the supply-demand ratio and that providers do not drop out. Proposed algorithms were shown to outperform related work on multiple performance measures. In addition, the thesis proposed a scalable matching and scheduling algorithm for an on-demand pickup and delivery broker for moving consumers with multiple candidate delivery locations and time intervals. Extensive evaluation showed that the proposed approach yields significant computational savings without sacrificing the solution quality.