Key propagation pathways of extreme precipitation events revealed by climate networks

Extreme precipitation events have substantial societal and economic impacts through stresses on infrastructure, outbreaks of waterborne diseases, and triggering of floods and landslides. Their magnitude and frequency are increasing under global warming, raising concern and motivating research. While advances in machine learning have improved short-term prediction of precipitation extremes, the spatiotemporal propagation patterns of EPEs and their underlying mechanisms remain insufficiently understood, particularly from a global comparative perspective beyond local case studies. This study establishes and evaluates a global directed network of EPEs over land using event synchronization to identify preferred propagation pathways. It investigates physical drivers, focusing on travelling Rossby waves and topographic influences, and assesses the potential predictability of EPEs along identified pathways.

Prior work has applied climate networks and event synchronization (ES) to EPEs and teleconnections. Boers and colleagues combined ES with climate networks to forecast over 60% of EPEs in the central Andes, attributing synchronization and propagation to travelling Rossby waves. Similar approaches revealed global teleconnection patterns of extreme precipitation and regional synchronization in the Ganga River basin, the United States, and East Asia. Studies highlight roles of regional weather systems (mesoscale convective systems, tropical cyclones), low-level jets, atmospheric rivers, and monsoons in transporting moisture and enabling EPE propagation. However, most studies focus on specific regions or weather systems, lacking a unified global analysis and comparison of propagation patterns across continents. This work addresses that gap by constructing a global land EPE network to extract preferred propagation pathways and their predictability.

Data: Daily precipitation from NOAA CPC Global Unified Gauge-Based Analysis (0.5° grid, 1980–2020) is used to derive EPEs, with extremes defined per grid as days ≥ the 95th percentile of wet days (≥1 mm/day). Consecutive extreme days are counted as one event, timestamped at the first day. Atmospheric fields (850 hPa geopotential height, zonal and meridional winds) come from NCEP–NCAR Reanalysis 1 (2.5° grid) for compositing. Topography uses ETOPO1 (1 arc-minute). Antarctica is excluded due to sparse stations. Event synchronization (ES): ES quantifies co-occurrence of extreme events between two event time series with dynamic delay tolerance up to a maximum delay τmax. For each pair of grid cells i and j, counts of events at i shortly after j (and vice versa) are computed under τmax, yielding directed synchronization strength qij in [−1,1], positive when events at i typically precede j. Typical dynamic delays are 6–8 days in this study; networks are tested for τmax = 3, 6, 10 days. Climate network construction: Nodes are land grid cells; directed edges are retained where ES is significant at p < 0.01 against a null model (1000 randomized surrogate pairs with uniformly distributed event times; threshold taken as the 99th percentile of the null). Edges are weighted by ES. A short-distance subnetwork is defined by links with geographic distance less than 2500 km, interpreted as regional weather-system influences; longer links are associated with teleconnections. Robustness is assessed by comparing edge-distance distributions across τmax values, which follow consistent power-law behavior, indicating stable network structure. Network diagnostics to extract pathways: Network divergence ΔS at each node (outdegree minus indegree) identifies sources (positive) and sinks (negative). Weighted mean azimuth of outward links from source regions indicates preferred directional propagation. Outward strength S(R) from a selected source region R to each grid cell is computed as the average ES from nodes in R; coherent spatial bands of high outward strength delineate propagation pathways from sources to sinks. Representative sequences of grid cells along these bands are chosen to characterize pathway timing using Hovmöller-type event counts. Atmospheric mechanism analysis: Composite anomalies of 850 hPa geopotential height and winds are calculated from 2 days before to 3 days after EPE onset at the pathway’s first grid cell to diagnose synoptic evolution, focusing on propagating low/high-pressure anomalies and low-level jets. Predictability metric: For each pathway, potential predictability is estimated as the fraction of EPEs at each downstream grid cell that occur within 3 days after an EPE at the first grid cell, providing an empirical probability along the pathway. This assesses timing coherence but does not consider event intensity.

• The global short-distance EPE climate network is robust across maximum delay settings (3, 6, 10 days); edge-distance distributions follow a consistent power-law. Synchronization effects diminish with increasing geographic distance.
• A pronounced West-to-East source–sink pattern emerges over North America, Europe, and Australia: western coastal or near-coastal regions typically act as sources (positive network divergence), while eastern regions act as sinks (negative divergence). This aligns with moisture transport by low-level jets and atmospheric rivers and the eastward phase propagation of Rossby waves. Northern Hemisphere divergence magnitudes are larger, reflecting higher land-node density.
• Sixteen preferred EPE propagation pathways are identified globally, mostly constrained by topography (e.g., mountain ranges and plateaus) and associated with mesoscale convective systems in the tropics/subtropics: • North America: From the southern U.S. source (region C), outward strength aligns along the Appalachian Mountains; EPEs propagate northeastward with typical inter-grid timing ≤2 days. Composites show eastward-moving low-pressure anomalies and extension of the Great Plains Low-Level Jet transporting Gulf moisture, plus orographic lifting along the Appalachians. Additional pathways include southwestern Canada to Hudson Bay Coastal Plain and southeastern Canada along the Laurentian Mountains to the Labrador Peninsula. • Australia: Three source regions near the northwest, central, and southern coasts produce pathways across the Western Plateau, Great Artesian Basin, and along the coast and the Great Dividing Range. Moisture contributions involve atmospheric rivers from the Indian Ocean and an Australian Low-Level Jet; composites show continued eastward movement of low-pressure anomalies. • Europe: Northern Europe pathway along the leeward Norrland Plateau (east of the Scandinavian Mountains) with weaker direct atmospheric river–extreme precipitation linkage due to rain shadow; variability tied to North Atlantic SSTs and anticyclonic circulation. Southern Europe shows two pathways associated with the Iberian Peninsula to the Alps and the Carpathians. Eastward Rossby wave activity underlies propagation. • South America, Asia, and Africa: Pathways reflect interplay of moisture transport (e.g., Amazon moisture, cold air intrusions toward the Andes) and topographic lifting; Rossby wave activity features prominently.
• Potential predictability along pathways is substantial: across the 16 pathways, the mean probability that a downstream grid cell experiences an EPE within 3 days after the source-cell EPE is 0.45. In North America, grid 2 steps along three pathways reach up to 0.68 on average, and final downstream cells still exceed 0.39. On average, more than 32% of EPEs conform to the identified pathway chains, reaching as high as 65.81% in some regions (e.g., South America region A, Europe regions B and C, Australia region C). Probabilities decline with distance from the source and depend on the specific box/grid selection.
• Seasonal occurrence of pathway “chain events” concentrates in summer–autumn over North America, Europe, and Asia, and in winter–spring in the Southern Hemisphere.
• Mechanistically, travelling Rossby waves embedded in westerly jets and topographic lifting consistently explain propagation patterns; low-level jets and atmospheric rivers provide moisture pathways enabling event sequences.

The study directly addresses the open question of whether EPEs exhibit preferred and predictable propagation pathways at the global scale. By constructing a robust, directed ES-based climate network and analyzing network divergence, outward strength, and weighted azimuths, the authors identify coherent source-to-sink pathways on all inhabited continents. The findings highlight how Rossby wave propagation and topographic lifting structure EPE chains and how moisture conveyers (low-level jets, atmospheric rivers) condition their development. The quantified empirical probabilities demonstrate meaningful potential predictability windows (order of days) along pathways, supporting early warning of hazards like floods and landslides. While spatial configurations of pathways appear only weakly influenced by ENSO, teleconnections such as ENSO and the MJO can modulate the predictability along pathways by altering background circulation and moisture availability. The work underscores the utility of climate network approaches for understanding and exploiting spatiotemporal connectivity among extremes for forecasting and risk management.

This work establishes a global, event-synchronization-based climate network of EPEs and reveals 16 preferred propagation pathways shaped by travelling Rossby waves, moisture transport, and topography. It demonstrates robust synchronization patterns over land, continental-scale source–sink structures, and substantial potential predictability on multi-day horizons along identified pathways. The approach offers prior knowledge to augment ensemble and machine-learning forecasts and provides a benchmark for evaluating and improving EPE representation in climate models using observed ordinal patterns. Future research should systematically quantify how teleconnections (e.g., ENSO, MJO) modulate pathway predictability, incorporate event intensity into predictability metrics, optimize pathway grid selection, and extend the framework to other extremes such as heatwaves, cold surges, and air pollution episodes.

• Potential predictability is based solely on event occurrence within a 3-day window relative to the source grid and does not account for EPE intensity or false-alarm/skill metrics, so it does not reflect realized forecast skill.
• Predictability depends on the specific selection and placement of pathway grid cells (spatial boxes); probabilities can change with alternative choices.
• Analysis focuses on short-distance links (<2500 km) to emphasize regional weather systems; long-distance teleconnections are not fully exploited in pathway extraction.
• Data limitations include exclusion of Antarctica due to sparse stations, reliance on gauge-based gridded precipitation with inherent interpolation uncertainties, and coarse resolution of reanalysis fields (2.5°) for composite diagnostics.
• Although robustness to τmax variations is examined, other parameter choices (e.g., percentile thresholds for EPE definition, significance thresholds) may influence network structure.
• Teleconnection influences on pathway predictability (e.g., ENSO, MJO) are noted but not fully quantified within the main results (some findings not shown).