Cross‐scale integration of knowledge for predicting species ranges: a metamodelling framework

Abstract

Aim Current interest in forecasting changes to species ranges has resulted in a multitude of approaches to species distribution models (SDMs). However, most approaches include only a small subset of the available information, and many ignore smaller-scale processes such as growth, fecundity and dispersal. Furthermore, different approaches often produce divergent predictions with no simple method to reconcile them. Here, we present a flexible framework for integrating models at multiple scales using hierarchical Bayesian methods. Location Eastern North America (as an example).Methods Our framework builds a metamodel that is constrained by the results of multiple sub-models and provides probabilistic estimates of species presence. We applied our approach to a simulated dataset to demonstrate the integration of a correlative SDM with a theoretical model. In a second example, we built an integrated model combining the results of a physiological model with presenceabsence data for sugar maple (Acer saccharum), an abundant tree native to eastern North America. ResultsFor both examples, the integrated models successfully included information from all data sources and substantially improved the characterization of uncertainty. For the second example, the integrated model outperformed the source models with respect to uncertainty when modelling the present range of the species. When projecting into the future, the model provided a consensus view of two models that differed substantially in their predictions. Uncertainty was reduced where the models agreed and was greater where they diverged, providing a more realistic view of the state of knowledge than either source model. Main conclusionsWe conclude by discussing the potential applications of our method and its accessibility to applied ecologists. In ideal cases, our framework can be easily implemented using off-the-shelf software. The framework has wide potential for use in species distribution modelling and can drive better integration of multi-source and multi-scale data into ecological decision-making.

Inference of biogeographic history by formally integrating distinct lines of evidence: genetic, environmental niche and fossil

“…Such a framework would have the advantages of using the full range of data available, allowing for the filling in of temporal or spatial gaps, reducing the impacts of data limitations and biases, and quantifying the confidence in conclusions (Schaub and Abadi 2011, Pagel and Schurr 2012, Talluto et al 2016). At a minimum, this novel integrative modeling framework would need to: 1) cohesively and quantitatively combine data and understanding from different fields, 2) identify and model key processes, 3) quantify uncertainty in data and processes, and 4) be computationally tractable.…”

“…For example, Pagel and Schurr (2012) developed a Bayesian DRM composed of a sequence of conditional, probabilistic equations that describe abundance, detectability, dispersal, population growth, and the influence of the environment on carrying capacity. An alternative to a DRM is a hierarchical Bayesian metamodel that uses a mechanistic or correlative model to constrain an ENM using genetic, phenological, trait, pollen-vegetation, experimental, or other data (Talluto et al 2016). pollen).…”

Multi‐scale integration of tree recruitment and range dynamics in a changing climate

“…We used the framework outlined in Talluto et al (2016), which operates by first constructing an SDM using occurrence and environmental data (hereafter "naïve SDMs", indicating that they are informed only by presence-absence data), and by then further informing the parameters of this model using sub-models that relate species performance to the same variables used to calibrate the SDM. Here, we build SDMs using occurrences of adult trees, and then further constrain these with sub-models using data for seedlings at two different scales, considering (a) seedling survival using smaller-scale experimental results and (b) seedling recruitment dynamics, using data from the FIA.…”

“…This was accomplished by simulating presence-absence datasets y recruitment and y survival following Talluto et al (2016). This was accomplished by simulating presence-absence datasets y recruitment and y survival following Talluto et al (2016).…”

“…In this study, we apply a recently developed, integrated metamodelling framework (Talluto et al, 2016) to directly assimilate juvenile and adult tree climatic niches and recruitment rates into a cohesive framework for projecting current and future ranges. We assimilate data from experimental and observational datasets of seedlings and adults of five dominant tree species in the Rocky Mountain region of the western US, and project future ranges while accounting for underlying variation in recruitment in response to climate.…”