Spatiotemporal variation in the length–weight relationship of Pacific yellowfin tuna reflects biological and fishery factors and affects aggregate weight estimation

Length–weight relationships (LWRs) underpin biomass estimation in tuna stock assessments, yet they are typically treated as spatially and temporally invariant despite mounting evidence of biological and environmental heterogeneity. Using a 16-year (2009–2024) at-sea observer dataset from the Taiwan Observer Program (TOP), we investigated spatiotemporal variation in the LWR of Pacific yellowfin tuna (Thunnus albacares) and evaluated how model structure and data source influence weight-based biomass inference. We applied generalized additive models with robust error structures to (1) quantify biological, environmental, and operational drivers of weight-at-length; (2) identify the empirically supported size interval within which traditional linear log–log LWRs remain defensible (“linear safe zone”); (3) examine spatiotemporal patterns and long-term regional trends in relative condition (Kn); and (4) assess the sensitivity of sample-based aggregate whole weight outcomes to alternative LWR specifications. Weight-at-length was significantly shaped by sex, seasonality, fishing depth (hooks-between-floats), bait type, and spatial structure. Log–log predictions departed from the classical power-law assumption at size extremes, while a linear safe zone was identified between 98 and 161 cm fork length. Within this interval, linear approximations performed robustly; outside it, nonlinear effects and predictive uncertainty increased sharply. Concurrently, Kn declined significantly in the northwestern subtropical Pacific (Region 1), suggesting emerging physiological stress in a warming transition zone. Scenario-based aggregate whole weight comparisons revealed strong internal consistency among fishery-specific TOP-derived models (<3% divergence), but marked inflation (>20%) when static regional LWR parameters were applied. This asymmetry demonstrates that data source selection exerts a greater influence on biomass-relevant weight inference than model form alone. Our findings position LWRs not merely as conversion coefficients but as dynamic ecological interfaces linking growth, environment, and fishery practice. Incorporating spatially structured, fishery-specific LWRs into assessment workflows can enhance the robustness of biomass estimation while improving early detection of climate-driven shifts in population condition.