Facilitating feedbacks between field measurements and ecosystem models

Abstract

Ecological models help us understand how ecosystems function, predict responses to global change, and identify future research needs. However, widespread use of models is limited by the technical challenges of model–data synthesis and information management. To address these challenges, we present an ecoinformatic workflow, the Predictive Ecosystem Analyzer (PEcAn), which facilitates model analysis. Herein we describe the PEcAn modules that synthesize plant trait data to estimate model parameters, propagate parameter uncertainties through to model output, and evaluate the contribution of each parameter to model uncertainty. We illustrate a comprehensive approach to the estimation of parameter values, starting with a statement of prior knowledge that is refined by species‐level data using Bayesian meta‐analysis; this is the first use of a rigorous meta‐analysis to inform the parameters of a mechanistic ecosystem model. Parameter uncertainty is propagated using ensemble methods to estimate model uncertainty. Variance decomposition allows us to quantify the contribution of each parameter to model uncertainty; this information can be used to prioritize subsequent data collection. By streamlining the use of models and focusing efforts to identify and constrain the dominant sources of uncertainty in model output, the approach used by PEcAn can speed scientific progress. We demonstrate PEcAn's ability to incorporate data to reduce uncertainty in productivity of a perennial grass monoculture (Panicum virgatum L.) modeled by the Ecosystem Demography model. Prior estimates were specified for 15 model parameters, and species‐level data were available for seven of these. Meta‐analysis of species‐level data substantially reduced the contribution of three parameters (specific leaf area, maximum carboxylation rate, and stomatal slope) to overall model uncertainty. By contrast, root turnover rate, root respiration rate, and leaf width had little effect on model output; therefore trait data had little impact on model uncertainty. For fine‐root allocation, the decrease in parameter uncertainty was offset by an increase in model sensitivity. Remaining model uncertainty is driven by growth respiration, fine‐root allocation, leaf turnover rater, and specific leaf area. By establishing robust channels of feedback between data collection and ecosystem modeling, PEcAn provides a framework for more efficient and integrative science.

Linking big models to big data: efficient ecosystem model calibration through Bayesian model emulation

“…We compared the emulator approach to the Differential Evolution Markov Chain with snooker update algorithm (DREAMzs) as it is one of the fastest converging algorithms known in the literature (Laloy and Vrugt, 2012). The implementation of DREAMzs was provided by the BayesianTools package (Hartig et al, 2017) which is called within the bruteforce data assimilation framework of PEcAn (v1.4.10), an ecosystem modeling informatics system (LeBauer et al, 2013). The emulator framework has also been implemented in PEcAn.…”

“…Meta-analysis priors were specified by fitting distributions to raw data collected from literature searches, unpublished data sets, or from expert knowledge (LeBauer et al, 2013). Direct mapping of previous information to model parameters allows us to account for the uncertainties in measurements derived from the collective weight of a large range of studies rather than arbitrarily choosing values from any one study (LeBauer et al, 2017).…”

“…Before these experiments, we conducted an uncertainty analysis (LeBauer et al, 2013;Dietze et al, 2014) to choose the model parameters for calibration. The parameters that can be constrained by data are those that contribute to the model uncertainty for that corresponding variable.…”

Changing How Earth System Modeling is Done to Provide More Useful Information for Decision Making, Science, and Society

“…For example, increasing the spatial resolution of a specific atmospheric physics model is justifiable in order to predict atmospheric dynamics more precisely (Shaffrey et al 2009;Palmer 2012) but if its computational requirements restrict the inclusion of other details then the model may be less accurate than had an alternative atmospheric model formulation been adopted to allow other component processes to be represented more accurately. The adequacy of a model structure, including the level of detail, can be assessed by the degree to which predictions can recapture the known (and relevant) dynamics of interest, although such assessments are not in widespread use (Judd et al 2008;Le Bauer et al 2013;Smith et al 2013).…”

“…Such methods will be important for propagating uncertainty into projections and enabling assessments of the value of alternative model formulations in terms of precision, accuracy, and overall confidence in how well the model captures reality (Vrught et al 2005;Berliner and Wikle 2007;Scholze et al 2007;Sexton and Murphy 2012;Le Bauer et al 2013;Smith et al 2013). New studies examining the tradeoffs between the level of model detail and the ability to quantify uncertainty would be informative in relation to this (Smith 2002;Ferro et al 2012;Palmer 2012).…”