Abstract
Abstract
How ocean-reanalysis choice changes inferred tropical cyclone (TC)intensity is rarely quantified. We evaluated seven products along 2,441 TCs(1993--2020) against storm-time drifter SST and gridded Argo on thefive-day pre-storm environment, propagating theirdifferences through the maximum and dynamic potential intensity (MPI, DPI). Surface biases are basin-dependent---warm GLORYS (Mercator globalreanalysis), cold HYCOM (Hybrid Coordinate Ocean Model). Thelayers decouple: the coldest surface product overestimates tropical cycloneheat potential most (\((+)\)3.4 kJ cm\((^{-2})\)), and the product ranking reordersfrom MPI to DPI, widening the spread from 1.78 to 2.69 m s\((^{-1})\). By stormstrength, the disagreement peaks at category 5, where two productssignificantly underestimate TCHP (up to 6.3 kJ cm\((^{-2})\)). The errorsare non-stationary: ORAS5 flips its bias sign across 2012, and the onlyreanalysis intensity trend significant at the nominal level, HYCOM's DPI decline(\((-)\)0.94 m s\((^{-1})\) decade\((^{-1})\)), is a step between 2010 and 2012 thatcoincides with the product's own production transition---a caution for trendwork built on any single reanalysis. Product choice is a structural,time-varying uncertainty---single-dataset studies may carry erroneousconclusions.