Network motifs shape distinct functioning of Earth's moisture recycling hubs

The study addresses how atmospheric moisture flows are organized globally and how their network topology influences regional hydrological functioning and sensitivity to change. Using high-resolution reanalysis-based moisture tracking, the authors represent evaporation-to-precipitation connections as a directed network to uncover structural features—particularly network motifs—that govern moisture redistribution. The purpose is to identify major terrestrial moisture recycling hubs and characterize their differing network topologies (directed vs reciprocal), which has implications for how land-use change and climate warming may propagate effects on precipitation. This is important because precipitation over land originates roughly equally from land and ocean, so both atmospheric change and land-surface alterations (e.g., deforestation, irrigation) can modify precipitation patterns, yet the arrangement and functioning of these moisture flows remain insufficiently understood.

The paper builds on advances in atmospheric reanalysis (e.g., ERA5) and Lagrangian moisture tracking that allow high-resolution mapping of moisture flows. Prior work produced global monthly mean flows between 0.5° grid cells for 2008–2017. Complex network analyses have been used widely in climate science, often using statistical similarity to define links; however, motifs—small, overrepresented connectivity patterns—are known to facilitate information transport and cascading dynamics in many complex systems (web, gene networks, food webs). Earlier research on tipping elements showed feed-forward loops (FFLs) can facilitate cascading transitions by lowering thresholds. The study leverages these insights to analyze directed, causal moisture connections, aiming to reveal new aspects of the hydrological cycle not captured by degree metrics alone.

Data and tracking: The authors use the UTrack Lagrangian atmospheric moisture tracking model driven by ERA5 reanalysis to trace moisture parcels from evaporation to precipitation. For each mm of evaporation, 100 parcels are released 50 hPa above the surface within each 0.25° grid cell. Parcel trajectories are updated every 0.1 h using 3D wind fields (25 pressure levels). Parcels are stochastically redistributed vertically on average every 24 h, weighted by the vertical moisture profile. Parcels are tracked for up to 30 days or until 99% of their moisture precipitates. Precipitation attribution: At each time step, the fraction of a parcel that precipitates is proportional to local precipitation divided by total precipitable water (P/TPW). The model updates parcel moisture and the fraction originating from a given source. Monthly moisture flows m_ij from source grid cell i to target cell j are aggregated and stored at 0.5° resolution for 2008–2017 as monthly means. Uncertainties: Two main uncertainty sources are addressed: (1) quality of atmospheric forcing (vertical resolution is crucial; ERA5 preferred), and (2) tracking model assumptions (Eulerian vs Lagrangian, time steps, number of parcels, interpolation). The 0.5° resolution cannot resolve very localized features (e.g., fine-scale orography, land use). Evaporation estimates are more uncertain than precipitation. Validation with deuterium excess indicates correspondence in tropical rainforests. Network construction: Grid cells are nodes; directed edges represent moisture transported from evaporation source to precipitation target. To construct an unweighted network emphasizing strong pathways, links are included in descending order of flow strength until the cumulative transported moisture exceeds a threshold p of total global transport. Main analyses use p=25% (with sensitivity at p=20% and 30%). Both all-to-all (including ocean–land and land–ocean) and land-to-land sub-networks are considered. Network measures and motifs: In- and out-degree quantify numbers of incoming/outgoing connections. Three motifs are analyzed per node: feed-forward loops (FFLs, directed triangular structures focusing on a target node), zero loops (ZLs, reciprocal bidirectional pairs), and neighboring loops (NBrs, a ZL plus an additional source creating directed reciprocity/corridor). Motifs are counted as: FFLs when a node is the converging target; ZLs as counts of bidirectional interactions; NBrs when a node is the center of the neighboring loop. Motif counts per node are normalized by the respective global maximum to obtain motif strengths. Regional aggregation and comparisons: Spatially aggregated motif strength differences (FFL−ZL and FFL−NBr) are computed within four predefined focus regions: Amazon Basin (AB), Congo Rainforest (CR), Indonesian Archipelago (IA), and South Asia (SA). Seasonal (monthly) analyses of FFL−ZL complement annual means to assess ITCZ and monsoon influences. Sensitivity analyses vary p between 20% and 30% and adjust regional box sizes to assess robustness.

• Identification of four terrestrial moisture recycling hubs in the land-to-land network: Amazon Basin (AB), Congo Rainforest (CR), Indonesian Archipelago (IA), and South Asia (SA).
• Degree structure: In-degree distributions are broader than out-degree, indicating the presence of super-receivers but absence of super-distributors of moisture. The tropics exhibit highest in- and out-degree consistent with regions of intense evaporation and precipitation.
• Motif prevalence and organization: The network is highly organized by motifs. In the all-to-all network, each edge is on average part of ~100 motifs and each node of ~5000 motifs; in the land-to-land network, ~40 motifs per edge and ~250 per node. Motifs occur about two orders of magnitude more frequently than in randomized networks (tested at p=25%).
• Regional motif signatures: • Amazon Basin: Dominated by FFLs (directed lens), indicating strong directed moisture propagation. This suggests efficient downwind propagation of perturbations (e.g., deforestation, drought) along preferred pathways. • Congo Rainforest and Indonesian Archipelago: Dominated by reciprocal motifs (ZLs and NBrs), behaving like washing machines (reciprocal cycles) and directed corridors (NBrs), implying more local, bidirectional dependencies. • South Asia: Also exhibits many ZLs (a weaker washing machine), with hotspots associated with Himalayan and other mountainous and monsoon regions.
• All-to-all vs land-to-land: Land-to-land directedness is largely carried by NBrs (moisture corridors), whereas ocean-to-land connectivity prominently features FFLs (moisture lensing). Switching ZLs to NBrs in comparisons increases the relative importance of FFLs in all-to-all but not in land-to-land networks.
• Seasonality: Monthly analyses show significant seasonal variability linked to ITCZ shifts and monsoon systems. Despite monthly changes (e.g., AB behaves as a washing machine in November), the annual mean pattern persists: AB remains a lens; IA and CR are washing machines; SA is a weak washing machine.
• Sensitivity and robustness: Results are robust for p between 20% and 30%. Increasing p strengthens the Amazon’s FFL dominance in the all-to-all network due to inclusion of more ocean-involved FFLs.
• Implications: Regions characterized by FFLs (notably the Amazon) are more prone to cascading impacts from perturbations, while ZL-dominated regions may experience more localized, reciprocal feedbacks affecting both precipitation and evaporation.

The directed network motif analysis uncovers structural features of the global hydrological cycle not captured by degree metrics alone. The Amazon’s unique status as a directed lens (FFL dominance) implies heightened vulnerability to cascading impacts: weakening of upwind links (e.g., via deforestation or drought) can propagate efficiently downwind, expanding affected areas and potentially lowering thresholds for tipping cascades. In contrast, ZL-dominated regions (CR, IA, SA) exhibit strong local reciprocity; perturbations can feed back locally, potentially increasing sensitivity to local changes but also enabling local recovery (e.g., reforestation restoring both evaporation and precipitation). These findings align with theory that FFLs facilitate cascading transitions in networks of tipping elements. Climate warming, by increasing atmospheric water vapor more than evaporation/precipitation, may lengthen moisture recycling distances, altering network topology and potentially amplifying directed propagation of changes. The use of directed causal connections (rather than statistical similarity) allows clearer interpretation and opens avenues for further research, such as identifying connectors via higher-order centralities, examining temporal evolution (e.g., ENSO phases), and constructing projected networks under climate change to assess future Earth system functioning.

This study translates a decade of ERA5-driven Lagrangian moisture tracking into a directed global network, revealing four major terrestrial moisture recycling hubs and their distinct motif-based topologies. The Amazon Basin stands out as a globally unique directed lens dominated by feed-forward loops, suggesting strong downwind propagation of perturbations and elevated risk of cascading impacts. The Congo Rainforest, Indonesian Archipelago, and South Asia are characterized by reciprocal motifs (zero and neighboring loops), indicating more localized, bidirectional dependencies. Contributions include: establishing motif-based diagnostics for hydrological network functioning; demonstrating the overrepresentation and organizational role of motifs; and differentiating ocean–land versus land–land connectivity structures. Future research should: integrate motif structure with dynamical ecosystem models to quantify tipping risks; analyze higher-order network measures (e.g., betweenness) to find key connectors; assess temporal variability and climate-mode influences (e.g., ENSO); and develop climate-change scenario networks to evaluate shifts in moisture recycling topology.

• Forcing data and resolution: While ERA5 provides high vertical resolution critical for accurate moisture transport, the 0.5° horizontal resolution of outputs cannot resolve fine-scale orographic effects and local land-use heterogeneity.
• Model assumptions: Lagrangian tracking choices (time step, number of parcels, vertical redistribution scheme) and interpolation introduce uncertainties. Although parameter choices were tested for stability, they remain approximations.
• Data biases: ERA5 precipitation biases in the tropics can affect tracking results; evaporation estimates carry larger uncertainty due to limited global observations.
• Network thresholding: Constructing an unweighted network by retaining strongest links up to a cumulative transport threshold p emphasizes backbone pathways but omits weaker connections; while results are robust for p=20–30%, different thresholds could slightly alter motif balances (notably FFLs in the Amazon all-to-all network).
• Spatial aggregation: Results for focus regions depend on chosen box extents; sensitivity tests suggest robustness, but exact aggregated values may vary with region definition.