Abstract
Interest in the global patterns of ecosystems, and their basis in plant physiological principles, goes back to the pioneering days of ecology in the late 19th and early 20th centuries. But this interest waned during subsequent decades, not to be rediscovered until “global change” emerged as a topic of concern.
Dynamic global vegetation models (DGVMs) merging physiological, biophysical, ecological and biogeographical processes were first developed in the 1990s. There are now many DGVMs, some originating in climate modelling centres and coupled into general circulation models. DGVMs have been used in analysis of the global carbon cycle and biosphere-climate feedbacks. The most advanced can predict patterns and changes in fire regimes, trace gas emissions and much else. Yet in some important aspects, such as predicting the effects of rising CO2 concentrations and global warming on productivity, there are stubbornly large differences among DGVMs. This situation reflects a lack of agreement on their underlying theoretical basis. The ways in which plants are categorized by “functional types” also remains rudimentary.
Meanwhile, plant functional ecology, Earth Observation, and atmospheric and ice-core measurements of the past and present concentrations of trace atmospheric constituents have made enormous progress, both in data collection and process understanding, which DGVMs have not taken on board. Hence the implication of my title: today’s global ecology is a shaky edifice, lacking foundations.
I will argue that the basis now exists for a new generation of predictive models for large-scale biospheric processes, based on a far more solid theoretical and empirical foundation. I will present examples for key processes, including photosynthesis, respiration and transpiration, where recent progress allows key processes at the leaf and plant levels to be represented both parsimoniously and accurately for global modelling. The next generation of DGVMs will be simpler than the previous one, and the model’s parameters and their uncertainties will be more transparent. A wide range of large-scale observations will provide benchmarks for their emergent properties.
But several obstacles need to be overcome. The almost complete separation of modelling and experimental communities is one. Improving models doesn’t only require application of engineering tools—benchmarking, Bayesian parameter optimization, uncertainty analysis and the like (valuable though these are). They also require experiments in controlled environments, e.g. to quantify the acclimation of photosynthetic temperature optima and Rubisco activity to environmental variations, and for these experiments to be conducted across the spectrum of plant functional types. On the other hand, modelling needs to be construed not only as an activity through which process knowedge is applied, but also as a means to facilitate the design and interpretation of experiments.