Bayesian model-based clustering of ecological time series
Abstract Ecological time series provide opportunities for investigating temporal dynamics across different levels of biological organization and can reveal distinct cyclical frequency signatures. Time-series clustering allows the identification of shared fluctuation patterns among ecological systems and consequently help explaining these patterns. However, missing data, heterogeneous sampling, and uncertainty present major challenges for clustering ecological time series in practice. We present a novel model-based approach for clustering time series. Our framework uses frequency domain analysis (Fourier analysis) to identify temporal frequency signatures and assign time series to clusters. The model is implemented within a hierarchical Bayesian framework, enabling the identification of clusters while accounting for uncertainty and incorporating covariates that explain cluster membership. This approach enables ecologists to group communities or populations with similar fluctuation dynamics and investigate the potential drivers underlying differences among groups. We evaluated the proposed method using a simulation study, which showed that the framework can successfully recover distinct frequency profiles. We then applied the method to Finnish moth monitoring data collected between 1993 and 2022, including 162 moth species in a species-level analysis and 49 moth monitoring sites in a community-level analysis. The method identified distinct temporal frequency profiles among species and communities and revealed associations between fluctuation dynamics and ecological characteristics. Our framework provides a flexible tool for studying temporal ecological dynamics, identifying groups with similar fluctuation patterns , and investigating the factors associated with those patterns in long-term monitoring datasets.
Authors
- Arthur Vinícius Rodrigues (ORCID: https://orcid.org/0000-0003-2656-558X)
- Pasi Sihvonen (ORCID: https://orcid.org/0000-0003-2237-9325)
- Nourhan Shafik
- Jarno Vanhatalo (ORCID: https://orcid.org/0000-0002-6831-0211)
Institutions
- University of Helsinki (FI)
- Finnish Museum of Natural History (FI)
Publication Details
- Journal
- Research Square
- Published
- 2026-11-16
- DOI
- https://doi.org/10.21203/rs.3.rs-11525622/v1
- Primary Topic
- Time Series Analysis and Forecasting
- Type
- preprint