Understanding and predicting how multiple ecosystem services (ES) covary, particularly in relation to social and ecological drivers of change, is a research imperative for guiding sustainable environmental management for human well-being. The identification of ‘ecosystem service bundles’ has been operationalised to facilitate a search for general rules determining ES associations in relation to social-ecological drivers. An increasingly popular approach for identifying and mapping ES bundles, largely based on cluster analysis, has been applied across the world to a range of cultural landscapes to facilitate cross-study comparisons of ES associations and their drivers. We reviewed the use of this correlative approach in recent studies and synthesized its application into four steps that capture the plurality of methods used. To critically assess the strengths and limitations of current approaches, we then applied these steps to a cross-study comparison (North and South French Alps) of relationships between social-ecological variables and ES bundles, as it is widely advocated that cross-study comparisons are necessary for achieving a general understanding of predictors of ES associations.
We found that even within the French Alps, there is enormous variation in the degree to which different social-ecological variables can explain the distributions of ES bundles. A key assumption with correlative approaches is that consistency in the spatial congruency between ES likely emerges from common social-ecological drivers. In actuality, ES may trade-off against each other simply because they compete for space (e.g. agricultural production and timber production), rather than being causally linked (e.g. agricultural production reducing water quality through diffuse pollution). The former trade-off can easily be represented through current correlative approaches but the latter ultimately requires a process-based description of the system to detect and quantify them. We therefore conclude that the approach reviewed is poorly suited to enabling sound understanding and prediction of ES bundles. A more hypothesis-driven approach than is currently taken is required to make real progress in predicting relationships between ES bundles, and we outline a roadmap of the types of research required to enable such an understanding to emerge.