Skilful multi-year predictions of Atlantic hurricane frequency

North Atlantic hurricane activity has increased substantially since the 1970s (refs 1,2), but whether this is attributable to natural internal variability(1,3) or external forcing(4-7) has not been resolved(8). Either way, hurricane frequency is potentially predictable, because climate models can directly simulate year-to-year variations in Atlantic tropical storm frequency, if forced by observed sea surface temperatures(9). However, skilful predictions have been limited to lead times of one season(10), and evidence for external forcing of hurricane frequency has been indirect, relying on statistical relationships(4) or external influences on related environmental factors(5-7). Here we extend skilful climate model predictions of hurricane frequency to lead times of several years, using decadal predictions(11) with nine variants of a general circulation model. In our experiments, the recent increase in tropical storm numbers was not caused by internal variability alone. This provides physically based model evidence of externally forced changes in hurricane frequency, albeit from a single modelling system. Initialization of the model with the observed state of the climate improves forecast skill, mainly through better predictions of tropical Pacific and North Atlantic ocean conditions, in line with previously documented teleconnections(1,3,12-15). Our results show that predictions of hurricane frequency are viable beyond the seasonal scale, and further elucidate causes of hurricane variability.