The biological carbon pump regulates atmospheric CO₂ on centennial timescales, yet its in-situ quantification is biased by the reference depth chosen for export. The canonical export horizons are the euphotic depth (z_eu; the depth of 1% surface PAR) and the mixed layer depth (MLD). However, z_eu is systematically shallower than the depth of UV photodegradation, the detectable penetration depth of downwelling irradiance at 380 nm (Pd₃₈₀). This leaves a intermediate layer between z_eu and Pd₃₈₀, in which particulate organic carbon (POC) accumulates and is remineralized by bacteria and zooplankton without being captured by Zeu-referenced export. Using quality-controlled BGC-Argo profiles from 444 floats across all major ocean basins (2012–2026), we (i) show that Pd₃₈₀ is the physically relevant remineralization horizon, (ii) quantify the resulting global export-efficiency bias, and (iii) develop and validate a model that infers Pd380 from satellite-derived euphotic depth, making the horizon globally and repeatedly mappable.
To locate where particle degradation accelerates, we fit the Martin flux law with the normalization depth z* treated as a free parameter rather than fixed at 100 m. For each profile cluster we select the reference depth that maximizes the power-law fit quality and yields the most stable, physically bounded attenuation exponent b. This fitting plays a dual role. Empirically, the optimal z* marks the transition from the production-dominated surface to the remineralization-dominated mesopelagic. It also validates the Pd380 hypothesis, since the optimal Martin depth tracks Pd₃₈₀ closely. UV-light attenuation thus marks where remineralization intensifies, and the z_eu-Pd₃₈₀ layer is the intermediate degradation zone.
From apparent oxygen utilization we then assessed globally the bias incurred by referencing Zeu instead of Pd₃₈₀. Profiles fall into four cases according to the relative position of MLD, z_eu and Pd₃₈₀. Globally we find 14% overestimation of export efficiency on average, but the signal is strongly regionally structured. The South Subtropical gyre shows the largest bias (40%), driven by the extreme UV transparency of its oligotrophic waters, while Arctic and North Subpolar regions show the lowest (2%), due to deep winter mixing. 
Because Pd380 cannot be retrieved directly from standard ocean colour products, we developed a predictive model that links it to satellite-derived z_eu through a joint regression whose slope and intercept vary smoothly with latitude and day of year. For an operational validation independent of the fitting data, we took in-situ Pd380 as ground truth, retrieved the co-located, contemporaneous, remotely-derived z_eu. 
The framework delivers a physically grounded, optically retrievable reference depth for POC export and a route toward more consistent, biogeochemically anchored export-flux estimates from basin to global scale.