Inference on spatiotemporal dynamics for coupled biological populations.

Area of Science:

• Ecology
• Epidemiology
• Computational Statistics

Background:

• Mathematical models in ecology and epidemiology require data consistency for reliable insights and policy.
• Metapopulation systems present statistical inference challenges due to nonlinear, stochastic interactions.
• Computational difficulties in inference can hinder understanding the link between models and data.

Purpose of the Study:

• To develop a statistically principled data analysis workflow for metapopulation systems using a novel inference algorithm.
• To address limitations of previous approaches in analyzing complex dynamic models.
• To critically evaluate epidemiological models and their policy implications using real-world data.

Main Methods:

• Utilized a recently developed algorithm for computationally tractable likelihood-based inference in high-dimensional, partially observed stochastic dynamic metapopulation models.
• Applied the algorithm to build a data analysis workflow for metapopulation systems.
• Conducted a case study using COVID-19 data to demonstrate the workflow's capabilities.

Main Results:

• The workflow successfully addressed limitations of prior methods for metapopulation system analysis.
• Identified and rectified weaknesses in an influential early-pandemic COVID-19 metapopulation model.
• Developed a new model with significantly improved statistical fit and parameter identifiability.

Conclusions:

• The developed workflow enables self-critical data analysis for metapopulation models.
• The findings suggest that the COVID-19 lockdown in China initiated on January 23, 2020, was more effective than initially estimated.
• This approach provides a robust framework for evidence-based policy in epidemiological contexts.