Finding a home for experimental data in terrestrial biosphere models: An empiricist’s perspective

Abstract:

Terrestrial biosphere models are necessary to project the integrated effects of processes and feedbacks on the climate system in 100 years. A tension exists between the representation of ecosystem processes in terrestrial biosphere models, which must be necessarily coarse, and the overwhelming complexity of processes that empiricists observe in the natural world. Working together, modelers and empiricists can diffuse this tension by targeting the experiments and observations needed to resolve model uncertainty. I have learned a few lessons in the recent realm of model-experiment interaction ‘Mod-Ex’: (1) Complaining about ‘bad’ or unrealistic representation of processes in models is unhelpful. Modelers are already in a position where they need to have expertise in any number of disciplines; no one person can be an expert in all. Instead, we (empirical scientists) need to proactively provide the information needed for model parameters and processes. This may require a global database. (2) Model needs are nearly always broader than narrow empirical questions. What ecologists might think of as ‘the boring background information’—meteorology, soil processes, site history—are all necessary to put important ecological processes in a modeling context. (3) Data collected to inform the model is more meaningful if it considers the way that models necessarily function (e.g., reaching an equilibrium state before projection into the future can begin). For example, the SPRUCE experiment was designed as a regression design, rather than an ANOVA design, to allow for models to predict response thresholds, rather than the experiment providing a ‘yes’ or ‘no’ answer. (4) Empiricists have an important role to play in guiding and provide constraints on scaling their small-scale measurements to the temporal and spatial scales needed for large-scale global models. This interaction will be facilitated by a move to trait-based modeling, which seeks to capture the variation within a system rather than mean responses, and allows the development of predictive trade-offs between certain traits, and relationships between traits and environment.