Marine heatwaves and cold spells as evolving heat-transport networks: life-cycle topology in the East/Japan Sea

Marine heatwaves (MHWs) and cold spells (MCSs) are usually described as spatiotemporal objects whose internal heat pathways remain implicit. We represent an event as a daily weighted, directed network whose nodes are its 0–50 m grid cells and whose edges are the anomaly heat fluxes across cell faces, closed by atmospheric and subsurface reservoir nodes; a node's net strength is its heat-flux convergence. For four large offshore events in the East/Japan Sea the mean net convergence tracks the mean anomaly tendency of the members (r = 0.68–0.85). The external inflow fraction follows the perimeter-to-area ratio of the object; per face, the boundary carries 0.4–0.8 of the interior flux near the peak, where the anomaly is weaker at the rim. Recirculation and betweenness bottlenecks peak with the event. The cold spells recirculate more than the heatwaves, which pass their anomaly through persistent corridors; a check against 672 events confirms this contrast. Plain Language Summary Marine heatwaves and cold spells are periods when the upper ocean stays unusually warm or cold for days to months.Theyarenormallymappedasblobsthatgrow andshrink,whichhideshow heatmovesbetween neighboring patches of water while the event develops. Here we treat each event as a network: every grid cell of the event is a node, and the heat carried by currents between neighboring cells is a link with a direction and a strength. Two extra nodes stand for the atmosphere and the deeper ocean. For four large events in the East/Japan Sea we find that the share of heat arriving from outside the event is largely set by how much edge the event has relative to its area, and that each face of the edge carries less heat than a face inside the event, where the anomaly is weaker at the rim. Mature events show more closed loops and a few bottleneck cells. The cold spells recirculate their anomaly more than the heatwaves, which pass theirs through along a few persistent pathways. The daily networks are a diagnostic of how an event is organized and a candidate input for graph-based prediction.