Estimated fatalities from motor vehicle crashes topped 40,000 in 2017, whereas approximately 4.57 million people were injured seriously enough to require medical attention with a total cost to society of approximately $\$\,413.8$ billion. In addition to human and monetary loss, it has been recently estimated that traffic accidents cause 4.2 billion vehicle–hours of delay across metropolitan areas, and an equivalent to $\$\,2.8$ billion gallons in wasted fuel. Reducing the delay between the time accidents occur and their mitigation is therefore critical to avoid large-scale, cascading, and costly effects on human lives. In this talk, I will discuss the problem of spatiotemporal quickest change detection, where the goal is to detect statistical changes in spatiotemporal series, and how it relates to the problem of traffic accident nowcasting. Within this context, I will present a new approach that optimally detects the time of an accident and subsequently estimates its geographic location in near–real–time based on speed sensor readings. Finally, I will demonstrate the performance of the proposed approach on a large-scale real-world dataset comprising speed readings and accident reports from the I405 freeway in the Los Angeles County.