Source-linked AI summary
Event coincidence analysis for quantifying statistical interrelationships between event time series: on the role of flood events as possible triggers of epidemic outbreaks
Jonathan F. Donges, Carl-Friedrich Schleussner, Jonatan F. Siegmund, Reik V. Donner
TL;DR
The paper addresses the need for systematic tools to quantify interrelationships between sparse event series and develops event coincidence analysis for that purpose. The framework measures strength, directionality, and lag, tests null hypotheses, and yields evidence that floods may have triggered epidemic outbreaks globally in the same countries, while the application remains illustrative and data limitations require caution.
Problem
Systematic tools are needed to quantify impacts and statistical interrelationships between event series, including series containing few events.
Method
Event coincidence analysis quantifies strength, directionality, and time lag using precursor and trigger coincidences, with significance tests based on temporal point-process null hypotheses.
Results
Flood events may have acted as possible drivers of epidemic outbreaks in the same country, with more than 50% of outbreaks preceded by flooding within 12 months and about 20% of floods followed by epidemics.
Takeaways & Limitations
Event coincidence analysis can support climate-impact and adaptation research on relationships between extreme events and epidemic outbreaks.
Takeaways & Limitations
The application is illustrative and requires additional independent databases and analyses accounting for reporting biases and other systematic effects.
Abstract
from arXiv · showhide
Studying event time series is a powerful approach for analyzing the dynamics of complex dynamical systems in many fields of science. In this paper, we describe the method of event coincidence analysis to provide a framework for quantifying the strength, directionality and time lag of statistical interrelationships between event series. Event coincidence analysis allows to formulate and test null hypotheses on the origin of the observed interrelationships including tests based on Poisson processes or, more generally, stochastic point processes with a prescribed inter-event time distribution and other higher-order properties. Applying the framework to country-level observational data yields evidence that flood events have acted as triggers of epidemic outbreaks globally since the 1950s. Facing projected future changes in the statistics of climatic extreme events, statistical techniques such as event coincidence analysis will be relevant for investigating the impacts of anthropogenic climate change on human societies and ecosystems worldwide.
1 Introduction
The paper introduces event coincidence analysis to quantify and test directed, lagged interrelationships between event series, addressing limited systematic tools for sparse event data. It illustrates the framework by examining whether floods precede epidemic outbreaks.
- Climate extremes are projected to increase, but systematic assessments of their impacts on ecosystems and society remain scarce beyond individual cases.
- Event series represent ordered event timings without using associated amplitudes, spanning applications from photon arrivals to neuronal spikes.
- Existing event-series methods often emphasize exploratory association, motivating more detailed tests of statistical interrelationships between event series.
- Event coincidence analysis quantifies the strength, directionality, and time lag of statistical relations between event series while testing hypotheses about their nature.
- The framework is applied to observational flood and epidemic data, yielding evidence that floods acted as epidemic drivers in the same country globally.
2 Methods
Event coincidence analysis quantifies directional and lagged interrelationships between pairs or groups of event series by counting events within specified coincidence windows. It supports significance testing under point-process null models, including Poisson and burst-generating processes.
- Event-series setup: Event series are ordered event timings analyzed as binary, unmarked point processes without using event amplitudes.The method considers event sets A and B over a common interval, with event rates defined by event counts divided by interval length.
- Coincidence definitions: The method tests whether B-events precede A-events within a tolerance interval, optionally after applying a nonnegative time lag.Instantaneous coincidences use a temporal tolerance ∆T, while lagged coincidences compare time-shifted A-events with B-events.
- Directional rates: Precursor and trigger coincidence rates provide directional measures by normalizing coincidences against A-events and B-events, respectively.Multiple events within one coincidence window are counted only once; the lag parameter allows lagged relationships to be represented.
- Aggregated rates: Aggregated coincidence rates combine normalized precursor or trigger coincidences across multiple paired event series, including country-level flood and epidemic data.The construction counts coincidences within each pair before aggregating across the set, while retaining the single-pair rule that multiple events in a window count once.
- Null models and testing: Statistical significance is assessed with null models generated by stochastic point processes, including Poisson processes and heavy-tailed inter-event distributions that produce event bursts.Analytical results are used where available; otherwise, Monte Carlo simulation estimates coincidence-rate statistics or p-values, including for the flood–epidemic application.
3 Application: extreme flood events as possible drivers of epidemics
Using country-level EmDAT data from 1950–2009, the study applies event coincidence analysis to test whether floods precede epidemic outbreaks. The results indicate statistically significant short- and long-range coincidence patterns, while emphasizing geographic, seasonal, and data-related limitations.
- Data and setup: The analysis covers 3,468 flood events and 1,152 epidemic outbreaks worldwide from 1950–2009 using monthly country-level data.Floods are treated as B-events and epidemics as A-events.
- Method: Event coincidence analysis tests both epidemic risk enhancement from preceding floods and possible flood-triggered outbreaks, using Poisson-process Monte Carlo significance tests.The framework evaluates precursor and trigger coincidences, with 95% and 99% significance levels derived from surrogate event series.
- Short-term results: 7% of floods were followed by an epidemic outbreak within the next month, a rate significant at the 99% level.Short-window results were robust across other small coincidence intervals.
- Long-term results: More than 50% of epidemic outbreaks followed a flood within 12 months, while about 20% of floods were followed by an outbreak within that window.Significant coincidence clusters occurred around monthly and annual timescales, including intervals between 9 and 13 months.
- Spatial results and limitations: Country-wise maps suggest substantial trigger coincidence rates in parts of South America, Southeast Asia, India, and Sub-Saharan Africa, but lack significance information and have country-resolution limitations.Large-country results may not align with the geographic regions and watersheds where floods occur.
- Interpretation and caveats: Long-term coincidence rates may reflect seasonal clustering of floods and epidemics, including causally unrelated events occurring in successive rainy seasons.The authors suggest that indirect effects may contribute to longer-lag associations, but state that seasonal confounding should be controlled in future studies.
4 Conclusions
Event coincidence analysis quantifies the strength, directionality, and lag of statistical interrelationships between event series while supporting significance tests under varied point-process null hypotheses. Its application provides evidence linking flood events with epidemic outbreaks and motivates methodological extensions for richer event data.
- Event coincidence analysis quantifies coincidence strength, directionality, and lag between event series.It distinguishes precursor from trigger coincidences and uses coincidence rates to measure strength.
- The method supports significance tests under null hypotheses based on Poisson and stochastic point processes with prescribed inter-event distributions.
- Globally aggregated country-level observations provide evidence that flood events may have acted as possible drivers of epidemic outbreaks in the past.The authors identify this potential causal relationship as a subject for further climate-impact and adaptation research.
- Future developments include richer null-hypothesis analysis, event amplitudes through marked point processes, and more explicit spatial information.
- Partial or conditional extensions could measure interrelationships between two event series conditional on additional event series.