Presenter: Sabrina Reis – Mathematics and Computer Science
Faculty Mentor(s): Weng-Keen Wong
Session: (In-Person) Oral Panel—Fuel, Fire, Grass and Compost
The ability to detect anomalous data is a critical component of any useful statistical analysis, but the process for identifying anomalies can prove time-consuming and arduous. To address these challenges, researchers often delegate data processing to an algorithm, which analyzes data with more speed, efficiency, and accuracy than manual calculations, enabling earlier detection of anomalies. The property of early detection is especially critical when monitoring spatio-temporal events such as wildfires. The critical impact of these events necessitate data sources that provide current and complete information. This need is often met by networks of sensors–for instance, air quality sensors–that collect real-time, localized data. When processed with an anomaly detection algorithm, the comprehensive data collected by sensor networks can reveal aberrations indicative of a spatio- temporal event. To explore how anomaly detection algorithms can facilitate early detection of events of interest using sensor data, we gathered historical data from open-source Purple Air sensors to build case studies of past wildfires. We then applied various types of changepoint detection algorithms to the data in hopes of identifying changes in the distribution of data that indicated a wildfire had broken out. The toolkit of detection methods produced by the project offer a cost-effective and portable way of enhancing our ability to monitor the formation and spread of wildfires.