Efficient Likelihood-Based Temporal Changepoint Detection in Spatio-Temporal Processes.

Area of Science:

• Environmental Science
• Data Science
• Statistical Modeling

Background:

• Analyzing complex spatio-temporal data is vital for understanding dynamic environmental phenomena.
• Traditional time series methods are insufficient for detecting changes in spatio-temporal processes.
• Existing approaches often rely on unrealistic independence assumptions across changepoints.

Purpose of the Study:

• To introduce a novel likelihood-based methodology for detecting abrupt changes in spatio-temporal processes.
• To address the limitations of traditional methods and unrealistic assumptions in change-point detection.
• To provide a scalable solution for analyzing evolving spatial data.

Main Methods:

• Developed a likelihood-based methodology for spatio-temporal change-point detection.
• Utilized a family of covariance models allowing for temporal nonstationarity.
• Employed a Markov approximation to reduce computational complexity in likelihood calculations.

Main Results:

• Applied the method to two years of daily wind speed data from Irish weather stations.
• Successfully identified a significant changepoint on July 24, 2021.
• The detected changepoint correlated with a major shift in observed weather patterns.

Conclusions:

• The proposed methodology is effective for detecting abrupt changes in spatio-temporal data.
• The method demonstrates utility in environmental and climatic studies, particularly with wind speed data.
• Offers a scalable approach for analyzing dynamic spatial patterns over time.