Umsted, L., Liu, J., Trujillo, P., Burrell, E., Baracaldo Lancheros, L., Ruiz, T., ... & Dodge, S. (2023, June). Understanding and Modeling Human Mobility Response to California Wildfires. In In Proceedings of 29th ACM SIGKDD Conference on Knowledge Disco

• Research Objective:
  • To address changes in human movement patterns using self-supervised learning
• Study Area:
  • Lake Fire area in Los Angeles County in 2020
• Data:
  • Geospatial coordinates of the fire burn edges attained from FRAP
  • Human mobility data from SafeGraph and MapBox
• Methods:
  • Self-supervised learning
    • Spatio-temporal density-based clustering to group POIs
    • Binary segmentation to detect change points in the aggregated activity series for each cluster
    • Train a logistic regression model to estimate the probability of impact
• Results:
  • The overall accuracy of the model was 75%
  • Locations within 20 km of the fire edge and in the direction of the burn are most likely to be impacted

• Research Significance:
  This study provides a predictive model to understand and predict human mobility responses to wildfires. Also, this model generates an impactedness label based on spatiotemporal clustering.