A cross-shelf, multi-product framework for delineating climatological upwelling centres in the Benguela Upwelling System

The identification, spatial extent, and number of upwelling centres (UCs) in eastern boundary upwelling systems (EBUS) are contingent upon the choice of sea surface temperature (SST) product, sensor class, and analytical threshold conventions. This study introduces a multi-product, cross-shore framework that applies a daily segmented-breakpoint regression to each transect of a fixed shore-normal mesh, consistently employed across an 11-year analysis (2015–2025) of the Benguela Upwelling System (BUS). Six Level-4 SST products, with native grid resolutions ranging from 0.01° to 0.25° and encompassing three sensor classes (infrared, infrared–microwave fused, and microwave-only), are processed symmetrically. A coherent two-regime cross-shore structure is identified in 87.3%–96.9% of profiles per product. Five of the six products locate the climatological inshore-to-offshore boundary at a median distance of 175–200 km offshore, while the microwave-only MW_OI product consistently places this boundary further offshore at 250 km, highlighting sensor class as a more critical determinant of UC structure than grid resolution. A ≥ 3/6 multi-product consensus robustly identifies the three central canonical BUS UCs (Lüderitz, Namaqua, Cape Columbine) across all four seasons, as well as a persistent JJA–SON UC at the Pelican Point shelf narrowing (22–23° S), which is absent from the canonical five-UC framework. However, Cape Frio and Cape Peninsula are not recovered, likely due to their proximity ( ∼ 2°) to the analysis-domain boundaries. A multivariate regression on the distances to the 200, 500, and 1000 m isobaths reveals the shelf-break distance as the primary geometric determinant of UC structure (mean partial r = + 0.24 for the climatological boundary distance, bootstrap-significant in 11 of 24 (product, season) regressions, p β < 0.05 ; r = − 0.24 for the inshore gradient roughness ratio, significant in 9 of 24). All statistical inferences employ block-bootstrap standard errors adjusted for the product-specific along-coast decorrelation length. This framework reconceptualises UCs as spatio-temporally dynamic maxima of upwelling-favourable conditions along a near-continuous coastline, rather than as static point features, and is readily adaptable to other EBUSs.