Atlantic hurricane power in 2008-2025 was below a frozen 1946-2007 absolute-SST regression and near a relative-SST regression; neither had year-to-year skill (HURDAT2, ERSSTv5)

Vecchi, Swanson and Soden (2008) fitted annual Atlantic power dissipation index (PDI) on August-October main development region (MDR) sea surface temperature (SST) for 1946-2007. Absolute and relative SST (MDR minus 30S-30N mean) fitted equally well but gave very different futures. After 2007 the MDR warmed about twice as much as relative SST (+0.66 C against +0.31 C), so 2008-2025 is an out-of-sample test. We fitted both regressions on 1946-2007 with HURDAT2 and ERSSTv5, froze them, and predicted 18 seasons. The observed mean PDI was 3.05 x 10^11 m^3 s^-2. The absolute model predicted 4.15 (95% interval 3.12-5.19, p = 0.018) and was too high in 16 of 18 seasons. The relative model predicted 3.34 (2.67-4.07, p = 0.20). A predictor MDR minus k times tropical mean matched the observed mean at k = 1.11 (ERSSTv5) and 1.03 (HadISST); k below 0.85 (0.65) failed. In 32 variants (SST set, MDR box, PDI/ACE/log PDI, training from 1946 or 1970), the absolute model over-predicted by 18-33% and the relative model by 0.3-19%; with 1970 training the relative model also failed in 12 of 16. Year by year, neither model beat the 1946-2007 mean: the MSE difference had a 95% interval of -0.98 to 2.13. Limits: 18 seasons, one mean level, no pre-1970 intensity correction, possible observing changes and multidecadal variability. Code, annual aggregates, hashes and a retrieval script are included.

tropical cyclones · power dissipation index · relative SST · Atlantic hurricanes · out-of-sample test · HURDAT2 · ERSSTv5 · HadISST · main development region · forecast verification