Faced with climate change and associated species range shifts, effective conservation strategies that suitably represent biodiversity need to account for both present and future species distributions and be robust to future warming. We consider this principle in relation to the Solitary Islands Marine Park (SIMP), a warming region in subtropical eastern Australia. Over the last decade, the habitat-forming kelp Ecklonia radiata has markedly declined in this region, corresponding with changes in herbivory and a tropicalisation of fish communities. Further evidence of tropicalisation includes a southwards extension of anemonefishes and host sea anemones since the mid-1990s. In light of these changes, can conservation planning for biodiversity representation be undertaken for decadal time-scales, the scale at which zoning plans in multiple-use Marine Protected Areas often operate? In support that it can, strong cross-shelf patterns in reef fish assemblages (inshore, mid, offshore) have been maintained over the decadal scale in the SIMP. These patterns have been used to help refine a Habitat Classification System (HCS) for application in conservation planning (e.g. using Marxan) through surrogate biodiversity representation. Loss of kelp habitat on mid reefs reduced heterogeneity of the habitat mosaic present in the region, impacting diversity and thereby affecting representation. However, persistence in broad assemblage patterns indicate that a well-designed HCS is a robust and informative tool for representing biodiversity in conservation planning in the face of climate change. Likewise, predictive models of species/functional group distributions and abundances have considerable utility for longer-term planning, although caution is required in their interpretation. Predictive models support the importance of distance from shore and depth as categories in the HCS. Marxan analyses using predictive models versus HCS categories indicate advantages and disadvantages in both approaches. If possible, a combination of both approaches is preferred, especially when also incorporating local knowledge of sites with exceptionally high biodiversity.