Connecting dynamic vegetation models to data – an inverse perspective
Abstract
Dynamic vegetation models provide process‐based explanations of the dynamics and the distribution of plant ecosystems. They offer significant advantages over static, correlative modelling approaches, particularly for ecosystems that are outside their equilibrium due to global change or climate change. A persistent problem, however, is their parameterization. Parameters and processes of dynamic vegetation models (DVMs) are traditionally determined independently of the model, while model outputs are compared to empirical data for validation and informal model comparison only. But field data for such independent estimates of parameters and processes are often difficult to obtain, and the desire to include better descriptions of processes such as biotic interactions, dispersal, phenotypic plasticity and evolution in future vegetation models aggravates limitations related to the current parameterization paradigm. In this paper, we discuss the use of Bayesian methods to bridge this gap. We explain how Bayesian methods allow direct estimates of parameters and processes, encoded in prior distributions, to be combined with inverse estimates, encoded in likelihood functions. The combination of direct and inverse estimation of parameters and processes allows a much wider range of vegetation data to be used simultaneously, including vegetation inventories, species traits, species distributions, remote sensing, eddy flux measurements and palaeorecords. The possible reduction of uncertainty regarding structure, parameters and predictions of DVMs may not only foster scientific progress, but will also increase the relevance of these models for policy advice.
What the paper shows and why it matters
(AI-generated)
Dynamic vegetation models describe plant growth, competition, mortality and regeneration from basic processes. For vegetation not in equilibrium, for example under climate change, this gives them advantages over correlative species distribution models. The models’ parameters, however, are traditionally set directly from independent measurements, and model output is compared with field data only to validate. That route is appealing because processes can be tested independently, but it limits how many processes, species and places a model can cover, leaves large uncertainty, and leaves unused data, such as distribution or pollen records, that reflect many interacting processes at once. The problem the paper takes on is the assumption that a process-based vegetation model must be parameterised directly, independently of the data it is meant to predict.
The paper argues that Bayesian inference lets dynamic vegetation models be fitted to a much wider range of data while keeping their process basis. Direct knowledge about parameters enters as prior distributions; observations of the quantities the model predicts enter inversely, through likelihoods, which state how probable the observations are given the parameters and an assumed error model. The resulting posterior distribution, which weighs the prior knowledge against the data, supports parameter estimation, comparison of model structures and prediction uncertainty. The paper spells out what this requires: more complex error models for correlated data, diagnostics in which systematic mismatches can point to structural errors in the model or systematic errors in the data, and about 5 × 10⁴ to 5 × 10⁵ model runs for about 20 parameters, so that, with limited parallelisation, one simulation should take no more than about a minute. It reviews candidate data, from inventories to eddy flux (gas exchange) measurements and palaeorecords, and recommends combining data from different scales, because their errors tend to be less correlated.
The paper concludes that fitting vegetation models to data need not reduce them to a complicated correlative model that is blindly adjusted to data, because prior knowledge about parameters and model structure stays central. The authors expect better-constrained predictions with quantified uncertainty, which matters for policy, and tests of ecological theory at different scales. Van Oijen et al. (2013) cite the paper in support of combining prior knowledge with data in their Bayesian calibration, comparison and averaging of six forest models.
References
- van Oijen et al. (2013). Bayesian calibration, comparison and averaging of six forest models, using data from Scots pine stands across Europe. Forest Ecology and Management 289, 255–268. doi:10.1016/j.foreco.2012.09.043