Our new perspective paper on “Connecting dynamic vegetation models to data – an inverse perspective” is now available online at the Journal of Biogeography (see also this post by my coauthor James Dyke). The paper is a part of a special issue on distribution models “The ecological niche as a window to biodiversity” that will appear later this year. I’ll most likely do a post about the issue when it goes online, so I won’t say more about it here, but I think it has turned out to be a really nice collection of papers, many of which deal with the question of how we can move from purely correlative to more process-oriented distribution models (this links to any posts related to this special issue).
Across the range of modeling approaches in the special issue, our paper is at the very end of the process-oriented spectrum. What we look at is dynamic vegetation models, i.e. process-based global and local vegetation models that are used to predict community composition and species distributions from basic factors and processes such as resource availability, photosynthesis, competition and life history (see Fig.1).
Figure 1: Dynamic vegetation models and their connection to the concept of the fundamental and realized niche. From Hartig et al., in press. Copyright see publisher.
So, actually you can’t get much more process-based than that. The question for those models is more: how can we tie them more closely to data, and in particular: is it possible to use existing data on various scales together with process-based models to infer species characteristics and in some sense the niche of a species in a similar way as it is currently done with statistical species distribution models?
We argue that the answer is yes, and that Bayesian methods are the most logical approach to do this. The latter is not because we’re fundamentally advocating Bayesian methods over alternative statistical approaches, but simply because the Bayesian approach seems best suited to the type of task (fitting process-models to a variety of data types at different scales that might be updated over time) that we envision to be on the horizon. A special advantage of Bayesian methods is that they naturally allow merging information from direct measurements of model parameters such as photosynthetic rates, etc. (through priors) with any inversely generated information.
So, I hope that by providing ideas, examples, references and explanation, this paper will encourage more people to go this route. We give a quite gentle and general introduction to Bayesian statistics with process-based models (see also Fig.2). We also discuss the merits and technical challenges of different data types, such as vegetation inventories, species trait databases, species distribution databases, remote sensing, eddy flux measurements and palaeorecords, that could be used to infer model parameters. We have all these data nowadays, but it is difficult to combine results from statistical analyses of these data in a coherent picture of a species. The hope that we express is that fitting process-based models will act as a catalyst to synthesize heterogeneous data into ecological meaningful parameters. This might not only improve predictions of ecosystem reactions to environmental change, but also to a better fundamental understanding of the functional responses and diversity of plants.
Figure 2: The Bayesian modeling cycle. From Hartig et al., in press. Copyright see publisher.