|
Plant communities process and store large amounts of the world’s
terrestrial carbon, and so may act to amplify, or dampen, climate change
caused by anthropogenic CO2 pollution. To predict these effects, current
Earth system models scale directly from the leaf-level physiology of
photosynthesis – which is relatively well understood – to the
ecosystem-level fluxes and stores of carbon, which are what matters for
the global carbon budget. But this approach ignores the reality that
plant communities are composed of individual plants, which make
decisions about how much to CO2 to extract from the atmosphere and where
to allocate this carbon, and which compete with other individuals of the
same or different species. Surprisingly little is known about these
processes, and modelling capabilities for them are almost non-existent
at present. We are putting plant growth modelling on a new footing, by
combining process-based models, with large amounts of data, to produce
rigorously-parameterized models whose predictions can be trusted enough
to integrate into larger analyses. We specify models with explicit
representations of environment-dependent carbon fixation, allocation
(e.g. leaves vs stem vs root) and tissue losses (e.g. frost damage);
parameterize these models with direct measurements of the biomass of
compartments of individual plants (e.g. above-ground biomass, root
biomass); and then run simulations under altered climate and nutrient
regimes. We examine multiple plant species, and plants growing in
competition, which allows us, for the first time, to begin to build a
rigorous, bottom-up understanding of the relationship between
physiology, plant growth, biodiversity and ecosystem functioning. The
models take the form of coupled non-linear differential equations, which
are integrated thousands of times within a custom MCMC sampling scheme,
providing estimates of parameter values and parameter uncertainty. These
computational requirements help to explain why this approach has only
recently begun to be employed in ecology. Compared to other scientific
disciplines, ecologists tend to be much less familiar with mathematical
and computational techniques. Therefore, we have made a primary aim of
this project to package the MCMC algorithm within a simple interface
that allows the non-specialist to read in data; specify a biological
model; parameterize the model from the data; and carry out simulations;
all with limited or no programming.
Collaborators:
-
Drew Purves (Microsoft Research)
- Lindsay Turnbull, Andy Hector, Bernhard Schmid (Institute of
Environmental Sciences, Zurich)
|