Fig. 5: Workflow to identify ecosystem service bundles.

The numbered boxes distinguish the individual ecosystem service indicators. We first established the optimal number of clusters for our methods to resolve to using consensus clustering, in this case four. Next, we used six clustering methods to create bundles of ecosystem services (ES). Co-occurring ES have the same colour. We then created an algorithm to relabel co-occurring ES, so bundles containing similar ES were given the same label. Lastly, we built the final bundles from the ES which clustered within a particular bundle most frequently. We calculated the robustness of these bundles by taking the average number of times each constituent ecosystem service was found in its bundle—if every ecosystem service within a bundle was assigned to it by six of the six methods, its robustness would be 100%.