Data gaps and outliers distort critical-slowing-down-based resilience indicators

Area of Science:

• Ecology
• Complex Systems Science
• Data Science

Background:

• Natural systems face increasing threats from anthropogenic pressures, necessitating accurate resilience assessment.
• Data-driven resilience indicators, like variance and autocorrelation, detect critical slowing down, signaling potential regime shifts.
• The reliability of these indicators is compromised by common data issues such as missing values and outliers.

Purpose of the Study:

• To develop a mathematical framework to understand the statistical dependency between variance- and autocorrelation-based resilience indicators.
• To rigorously characterize the impact of data issues on the interpretation of resilience indicators.
• To provide a foundation for improving data preprocessing and accuracy assessments for resilience studies.

Main Methods:

• Developed a general mathematical framework to analyze the statistical relationship between resilience indicators.
• Utilized synthetic and empirical time series data to test the framework.
• Quantified the effects of missing values and outliers on indicator agreement and bias.

Main Results:

• The agreement between variance- and autocorrelation-based resilience indicators is fundamentally influenced by the initial data point of a time series.
• Missing data significantly reduces the agreement between resilience indicators.
• Outliers introduce systematic biases, leading to an overestimation of resilience when using temporal autocorrelation.

Conclusions:

• The study provides a rigorous understanding of how data quality affects resilience indicator interpretation.
• Findings highlight the need for careful data preprocessing to ensure accurate assessment of system resilience.
• This work offers a foundation for improving the reliability of resilience indicators across various scientific disciplines.