Nonlinear dynamics of river runoff elucidated by horizontal visibility graphs

Area of Science:

• Complex Systems Science
• Hydrology
• Network Science

Background:

• Horizontal Visibility Graphs (HVGs) construct networks from time series, reflecting nonlinear dynamics.
• HVGs can differentiate between chaotic and stochastic time series.
• Previous studies show conflicting results on river runoff data characteristics.

Purpose of the Study:

• Investigate the sensitivity of HVG methodology to real-world data properties and preprocessing.
• Analyze the impact of time series length, ties, and deseasonalization on HVG analysis of Brazilian river runoff data.
• Reconcile contradictory findings in previous HVG studies of river runoff.

Main Methods:

• Applied HVG methodology to approximately 150 Brazilian river runoff time series (daily resolution, ~65 years average length).
• Examined the effects of data preprocessing steps, specifically deseasonalization, on network properties.
• Conducted empirical analysis and extensive simulations to understand methodological impacts.

Main Results:

• Minor differences in data preprocessing, such as deseasonalization, can reconcile apparently contradictory HVG results.
• Deseasonalization significantly alters long-term correlations and runoff dynamics.
• HVG analysis, after accounting for preprocessing, reveals complex behavior in river runoff, a mix of short-term correlated noise and long-tailed behavior.
• River dam construction tends to increase short-term correlations in runoff series.

Conclusions:

• Methodological choices and data preprocessing significantly impact the interpretation of river runoff dynamics using HVGs.
• HVG analysis is sensitive to preprocessing, requiring careful consideration for robust real-world time series interpretation.
• The study highlights the general applicability and challenges of HVGs for analyzing complex real-world time series.