Global droughts connected by linkages between drought hubs

Droughts are among the costliest natural disasters and operate over large spatial and temporal scales, with impacts on water, agriculture, energy, and socio-economics. Recent concern has focused on spatially compounding, mega-scale droughts that can lead to global food price shocks, trade disruptions, political instability, and migration. Risks of synchronous drought conditions across major crop regions have risen in recent decades and may be amplified by projected increases in dry spell length and drought frequency under climate change. Understanding the global synchronization (teleconnection) structure of drought events is therefore critical for assessing the likelihood of spatially compounding droughts and potential simultaneous breadbasket failures. Prior studies have examined teleconnections and large-scale circulation drivers of synchronous droughts, often focusing on selected climate modes and using methods such as EOFs that are limited in capturing higher-order statistical interrelationships. To overcome these limitations, this study applies Complex Network (CN) analysis with Event Synchronization to quantify all statistically significant links of synchrony in global drought onsets without a priori assumptions about specific drivers. Research questions: (a) How are drought events regionally and globally linked? (b) How does the spatial scale of drought synchronization vary globally? (c) How does planetary-scale synchronization occur among drought events, and what physical processes connect synchronous occurrences?

The paper reviews work linking teleconnections (e.g., ENSO, PDO, AMO, NAO) and large-scale circulation to drought synchrony at regional-to-continental scales. Traditional approaches (EOFs, coupled pattern analysis) have been useful for identifying dominant modes but are limited in exploring higher-order statistical relationships and capturing complex teleconnection structures among extreme events. Complex Network (CN) concepts have recently been introduced in climate studies to analyze extremes (e.g., extreme precipitation, heatwaves, typhoon tracks) and their spatial structures and directionality. Prior CN work on drought focused on propagation within the continental United States, revealing regional differences in travel distances of droughts. However, CN methods had not been applied to global drought synchronization. This study positions CN and Event Synchronization as higher-order, nonparametric tools to characterize global synchrony in drought onsets, enabling detection of hubs, bottlenecks, and core–periphery structures not easily revealed by EOF-based methods.

Data: Self-calibrated Palmer Drought Severity Index (ScPDSI) at 0.5° × 0.5° from CRU for 1901–2018, with PET computed via Penman–Monteith. Sea surface temperature (SST) from NOAA ERSSTv5. Precipitation from GPCC. Supplementary sensitivity uses ERA-20C soil moisture (to 2010). Drought event definition: Threshold-level method. Moderate drought onset is when ScPDSI drops below −2; event continues while ScPDSI remains < −2. Events characterized by onset timing and duration. For duration-based analyses, short-term events are ≤6 months and long-term events >6 months. Event Synchronization (ES): Construct binary time series of drought onsets for each grid (1 for months with onset, 0 otherwise). ES counts temporally coincident onsets between grid pairs allowing dynamic delay with maximum lag τmax = 3 months (also tested 6 months). For locations i and j with onset times tl and tm, ES computes c(ij) and c(ji) using windowed coincidence criteria and defines synchronization strength Qij = [c(ij)+c(ji)]/(si sj). Result is a symmetric synchronization matrix for all grid pairs (60,787 nodes). Network construction: Build an undirected synchronization network by retaining only the strongest 0.01% of nonzero Qij values. The threshold θ is the 99.95th quantile; here θ = 0.51 (p-value 0.0001 under a null of independent uniform events). Adjacency Aθij = 1 if Qij > θ; else 0. Network metrics:

• Degree centrality (DC): number of links of a node; reflects count of locations synchronized in drought onset within τmax.
• Betweenness centrality (BC): fraction of all-pairs shortest paths passing through a node; indicates control over information (e.g., moisture deficit) flow in the network.
• Clustering coefficient (CC): density of interconnections among a node’s neighbors; higher CC implies strong local coherence and reduced likelihood of long-range synchrony.
• Mean synchronization distance (MSD): weighted mean great-circle distance to connected neighbors using Qij as weights; indicates spatial scale of synchrony (teleconnection reach). Regional metrics:
• Connection density (CD): percentage of a specified region’s area with significant links to a given node.
• Average synchronization density (ASD): average Q from a region to a node; indicates link strength. Homophily/rich-club: Quantified using assortativity coefficient (AC; degree assortativity) and rich-club metric as functions of degree threshold. Robustness and sensitivity analyses:
• Temporal robustness: compare 1901–1958 vs 1959–2015 DC patterns.
• Indicator sensitivity: compare ScPDSI-based network with soil moisture drought defined as monthly soil moisture below 15th percentile (ERA-20C, 1959–2010).
• Lag sensitivity: compare τmax = 3 vs 6 months.
• Duration dependence: separate networks for short- (≤6 months) and long-term (>6 months) events; compute DC ratio = log(DC_long/DC_short). Physical composites: For selected hubs (e.g., Altiplano/Bolivian High region) and teleconnected regions (e.g., South Africa), identify months when both (AND) or at least one (OR) regions are in drought; compute deseasonalized composites of SST and precipitation anomalies to diagnose drivers (e.g., ENSO).

• Global linkage heterogeneity and hubs: Degree centrality (DC) varies from >100 to >10,000, indicating highly heterogeneous connectivity. Degree distributions exhibit heavy tails beyond DC > 5000 across all continents, evidencing a small number of highly connected drought hubs (e.g., Southern Europe, Northeast Brazil, Australia, Northwest USA).
• Core–periphery and topology: Strong spatial correlation between DC and BC (≈0.83) shows hubs also have high betweenness, forming a core–periphery structure wherein highly connected core nodes are surrounded by sparsely connected periphery. Hubs often have low CC, implying they connect across regions rather than only locally; exceptions include Sahel and South Africa where high CC suggests spatially contiguous droughts.
• Spatial scale (MSD): Mean synchronization distance ranges from about 2000 to 10,000 km, indicating regional to circumglobal linkages. MSD tends to be larger in the Southern Hemisphere and along coasts. Multiple synchronization scales are associated with hubs, reflecting diverse teleconnection influences.
• Geographic patterns: High DC, BC, and MSD over western/northwestern USA (linked to North Pacific High, storm tracks, ENSO), Sahel and southern Africa (Botswana High, interhemispheric gradients, tropical ocean warming), European dipole (NAO influence), Central/Southwest Asia and Tibetan/Southeast Asia (Indian Ocean/ENSO, Rossby wave responses), and northern and southeastern Australia (Bilybara High, subtropical ridge, ENSO).
• Topographic modulation: Elevated terrains (e.g., Andes, Tibetan Plateau, Cordilleras) coincide with higher DC and MSD, consistent with topography’s influence on moisture transport and event propagation.
• Robustness: Drought hubs and broad spatial patterns are consistent across 1901–1958 vs 1959–2015, with increased synchronization in 1959–2015 over Sahel, Eastern Asia, and Southern Europe. Patterns from ScPDSI broadly align with soil moisture drought networks (1959–2010), though some regional differences appear (eastern USA, SE Asia, SE Africa). Increasing τmax from 3 to 6 months raises linkage magnitudes but not spatial patterns.
• Rich-club and homophily: Positive degree assortativity (AC = 0.16) and monotonically increasing rich-club metric indicate hubs preferentially connect with other hubs (rich-club phenomenon), enhancing risk of synchronous multi-continental droughts.
• ENSO as key driver: Composites for synchronous drought in Altiplano (Bolivian High) and South Africa show El Niño-like SST anomalies (0.6–0.8 °C positive in eastern-central equatorial Pacific) associated with widespread precipitation deficits over Northeast Brazil, South Africa, and parts of Australia, consistent with upper-tropospheric anticyclones and Rossby-wave teleconnections driving simultaneous droughts.
• Duration dependence: Long-term (>6 months) drought networks show higher DC magnitudes and dominance over regions such as South/East Australia, Southern Africa, Midwest USA, Sahel, and Western India, indicating stronger oceanic SST control. Short-term (≤6 months) synchronization is more pronounced in NW USA, consistent with land–atmosphere processes. Latitudinal averages show stronger linkage for long-term droughts in the Southern Hemisphere.
• Practical implications: Hubs with largest spatial scales are often ENSO-modulated, implying heightened synchronous risk during El Niño; CN-derived metrics offer new avenues for risk management and climate model evaluation.

The CN analysis directly addresses the research questions by revealing how droughts are regionally and globally linked through a heterogeneous, partially scale-free network with distinct hubs that act as connectors across continents. The strong DC–BC relationship and low CC at hubs demonstrate a core–periphery topology that facilitates long-range synchronization, explaining how planetary-scale connectivity emerges. MSD quantification clarifies that hubs operate across multiple spatial scales (up to ~10,000 km), consistent with known teleconnections. The homophily and rich-club behavior among hubs indicate that when one hub enters drought, others are statistically more likely to do so, elevating the risk of simultaneous, multi-continental droughts (e.g., over South America, Africa, Australia). Physical composites implicate ENSO—especially El Niño—as a primary driver of such synchronized events via SST-induced circulation anomalies (upper-level anticyclones, Rossby waves), while regional weather systems and topography modulate local-to-regional coherence. Duration-dependent differences highlight oceanic control on persistent droughts versus land–atmosphere processes for short-lived events. These findings are significant for forecasting and mitigation: identifying hubs and their teleconnection scales can inform early warning, agricultural adaptation, and resilient trade networks; CN metrics can augment process- and structure-based evaluation of climate models, potentially improving projections of synchronized drought risk under climate change.

This study introduces a global Complex Network framework based on Event Synchronization of ScPDSI drought onsets to uncover the topology of drought synchronization. Key contributions include identifying drought hubs and a core–periphery, partially scale-free structure; quantifying spatial scales of synchrony (MSD up to ~10,000 km); demonstrating rich-club homophily among hubs; and linking synchronous multi-continental droughts to ENSO-driven circulation anomalies. Robustness across time periods, indicators, and lag choices strengthens confidence in the results. Future research directions include: exploring multi-scale and variable lag structures; disentangling relative roles of oceanic teleconnections, land–atmosphere coupling, and anthropogenic water use; extending to other drought indices and impact metrics; integrating CN metrics into climate model evaluation and model development; and assessing future changes in synchronized drought risks and implications for global food systems and trade resilience.

• Fixed maximum time lag of 3 months for primary analyses (with a 6-month sensitivity), which may not capture all relevant multi-scale delays.
• Data length and quality constraints inherent to historical datasets, especially for drought studies.
• ScPDSI characteristics (long-tailed distributions and persistence) may influence inferred network structure; impacts not fully quantified.
• While composites implicate ENSO and other modes, attribution among general circulation, land surface processes, and anthropogenic water use remains to be partitioned in future work.
• Network construction relies on a stringent top 0.01% synchronization threshold (θ = 0.51); alternative thresholding or weighted networks could be explored for sensitivity.