Evaluating explanations of land cover change using approximate Bayesian Computation
The ecology of the Mediterranean region today is the product of thousands of years of human intervention, combined with the effects of climatic change on its flora, and natural disturbance arising from its active fire regim e. Understanding the complex ways in which anthropogenic land- cover change, climate, and fire interact with each other could help guide public policy and land management decision- making, especially important to ensure sustainability in an uncertain future of climate change. A fruitful step in this endeavour will be to explain changes to landscape- scale vegetation ecology arising from the time when humans first began to compete with fire as a source of ecological disturbance 10,000 years ago – the agricultur al revolution. Paleoecological data provide a means to study ecosystem change in the distant past, and over timescales longer than a human lifetime. In particular, analysis of how the abundance of pollen produced by different plant functional types has changed over time can tell us much about the evolution of the composition of those functional types in the land- cover around a study site. Pollen time series can be viewed as fingerprints of the interacting processes of anthropogenic land- cover change, fire a nd ecological succession, against the backdrop of a changing climate. However, no amount of scrutiny of the data alone can elucidate the causal role each of the interacting factors had in creating them. We have developed a model which integrates theoretica l understanding of ecological and anthropogenic processes via an agent -based modelling framework. By simulating both natural and anthropogenic processes and allowing them to interact with each other in a spatially explicit virtual landscape, we can generat e dynamic pictures of landscapes evolving in time which are consistent with socio- ecological theory. However, as scientists we must be vigilant in our reporting of uncertainty about our models. Historical sciences provide a unique challenge in this regard, since theories cannot be tested by direct experimentation. In this talk, I will discuss the use of approximate Bayesian computation (ABC) as a means to unite acausal paleoecological data with a socio- ecological simulation model which encodes causal hypoth eses about Neolithic land -cover change, but whose likeness to reality is uncertain. ABC will be considered as a tool to support inference about which models best explain available empirical data, as well as indicating which aspects of those models are most uncertain.