Predicting the behaviour of complex dynamical systems by extrapolating from historical patterns is problematic. This is particularly true for predicting future population consequences to climate change, as we are asked to extrapolate to mostly novel environmental conditions. Optimization theory may allow biologists to move from describing patterns and mechanisms to being able to predict from how organisms should behave. Understanding the rules that allow organisms to optimize their behaviour should thus help us to better predict individuals’ and thereby populations’ responses to novel conditions. However, rigorous empirical tests of whether organisms indeed adapt their behaviour to the rapidly changing conditions as predicted by optimization models and whether this accurately predicts the population consequences are rare. Here we test whether a long-lived shorebird adapts its habitat preference and settlement patterns in a rapidly changing environment. Using a game theoretical model parameterized with 30 years of demographic data, we show that birds should adapt their habitat selection preference during their life in a deteriorating environment. Breeder removal experiments and observations on prospecting behaviour of nonbreeders conducted during different stages of population decline suggest that birds have only partly adapted their preferences, indicating that prospecting individuals may use only particular cues to detect rapid changes in habitat quality. We show that understanding the behavioural mechanism helps us to better predict the future population development under climate change, but at the same time highlight the counterintuitive result that maladaptive behaviour at the individual level can be good for population viability.