We examine the oft-quoted relationship between the migration of Fraser Rive
r sockeye salmon around the northern end of Vancouver Island and sea surfac
e temperatures. We examine the methods used to estimate the northern divers
ion and conclude that the estimates have a sufficiently low expected error
to form a useful representation of sockeye salmon behaviour. The well-known
relationship with Kains Island sea surface temperature is explored and pro
blems are pointed out. In particular, we explore why Kains Island temperatu
res are good predictors of salmon behaviour in May when the sockeye can be
over 1000 km away, but the coastal temperatures are poor predictors in July
to September when the salmon are actually close by. We show that a more ro
bust predictor can be developed using open ocean temperature fields and we
show why Kains Island fails as a predictor during the summer months. Finall
y, we show by cross-validation that the northern diversion is predictable w
ith an r.m.s. error of about 0.1.