Predicting the effects of landscape structure (composition and configuration) on ecosystem services (ES) remains a major challenge. The simplest approach is to assume land cover alone is a sufficient proxy for services (simple benefits transfer). However, this approach ignores configuration effects and can provide misleading results when known sites are poor predictors for new sites. Field studies and process-based models of the underlying mechanisms of the social-ecological system from which ES arise can provide more insights, but are data-hungry,and are therefore only usually applicable for well understood case studies. Here, we outline a new approach for modelling ES that involves coupling simplified simulated landscapes with different land-cover types at different spatial scales. Our hypothesis is that this general, relatively simple approach enables us to capture key elements of the spatial complexity of the social and ecological variables that predict distributions of ecosystem services. We present initial analyses in which we model pollinator-dependent agricultural production by coupling two binary landscapes. First, pollinator abundance is modelled as a function of the structure of natural land cover; this is then an input into a second binary landscape where agricultural output is modelled as a function of the resultant pollinator abundance.
The initial results from our coupled spatial simulation model show that optimum ES provision (here agricultural production) is found in landscapes with intermediate levels of randomly distributed natural land cover. These results are qualitatively similar to those obtained from more complex landscape simulation models, and fit with our theoretical expectations of the effects of landscape structure on this type of ES. This suggests that our simple intuitive approach for modelling the effects of landscape structure ES has considerable promise. We discuss how our modelling framework can be extended to better understand and predict how landscape structure at multiple spatial scales is likely to affect a wide suite of ES, and how the results of our modelling work can be empirically tested.