Habitat restoration frequently focuses on reaching an idealized steady state, but this is unrealistic for disturbance-dependent ecosystems where temporal variability is inherent and habitat conditions are expected to fluctuate. Understanding the ways in which the outcomes of restoration change over time in disturbance-dependent ecosystems can better inform adaptive management plans and increase the likelihood that restoration efforts will be effective. We conducted a decade-long restoration experiment to test how restoration efforts to increase disturbance levels impact habitat quality and populations of an endangered butterfly over time. We show that changes in plant communities as a response to disturbance vary depending on time since restoration, with target host plants initially increasing and peaking several years postrestoration but then declining. In the absence of further disturbance, butterfly population sizes follow a similar pattern, with population declines concurrent with declines in host plants. Due to this nonlinear response, management actions within disturbance-dependent ecosystems need to include long-term monitoring in order to accurately capture changes in habitat response, as well as active, adaptive planning that shifts according to current system stability. Restoration efforts within these dynamic habitats are more likely to succeed when temporal variability is explicitly tracked and multiple cycles of restoration are considered as part of management actions.