Re-evaluating how well tropical cyclone activity can be predicted over the 20th century from sea-surface temperatures

Multi-decadal variability of Atlantic tropical cyclone (TC) correlates with the SST differences between the Atlantic main development region and the entire tropics. Despite general consistency, the observed Atlantic TC count is lower than the expected value computed from existing SST products in the 1930-1960s. Such a divergence could be a result of deficiencies in the predictive model or errors in TC counts associated with ship-based, pre-satellite observational systems. Another possibility is that observational estimates of SSTs contain biases at decadal time scales. Using a recently-developed bucket SST dataset that corrects for relative offsets amongst groups of ships, we show that, over the 1930-1960s, Atlantic SST is colder after correcting for warmly offset German measurements, whereas Pacific SST is warmer after accounting for a truncation-induced cold bias in Japanese observations, suggesting overall better agreement between TC counts and SST-based predictions. Furthermore, accounting for group-wise SST offsets demonstrates greater SST uncertainties than previously recognized, impacting the agreement between TC counts and SST-based predictions. We explore the relationship between simple SST-based predictions and the degree to which systematic deviations remain when accounting for biases and uncertainties in SSTs.