CTCD: Strand 1 Introduction

Strand 1: Carbon Flux Calculations and their Uncertainty

The core of CTCD’s mission is to provide best possible estimates of biospheric carbon fluxes and to quantify their uncertainty, partitioning uncertainty according to its sources in internal parameters, input data, initial conditions and model inadequacy. Strand 1 concentrates on the construction of the framework within which data can influence models and uncertainty can be attached to model calculations, together with model development and the actual calculations providing uncertainty estimates. The work is based on state-of-the-art global vegetation modelling and new Bayesian statistical techniques.

Individual Work Packages within the Strand test and refine vegetation models. A programme of comparisons of global and local (SDGVM and stand-based FR) models against each other and against both ground and satellite-derived data from a range of sites over Europe is helping to quantify imprecision in generalized global models that results from process simplification, and is contributing to refinement of the models. The results are contributing to deeper understanding of specific issues; for example, to assessment of the effect of management practices on carbon sequestration at UK and European scales. Further projects are constructing UK carbon flux and uncertainty maps and setting the parameters for development to European and global scales.

Other Work Packages within the Strand focus on development of Bayesian methodology for uncertainty and sensitivity analysis and model calibration. A key concept in this work is that of an emulator: a representation of a model in terms of a random function whose distribution serves both as an approximation to the model and simultaneously gives a description of its own precision. (In Bayesian statistical terms the emulator is to the model as a posterior or prior distribution is to an unknown parameter.) In their role as approximations emulators typically run many times faster than the full model, and so offer important advantages for explorations of model properties traditionally relying on multiple runs, such as sensitivity and uncertainty analyses. Even more importantly, the notion of Bayesian updating of an emulator gives a natural and powerful framework for quantifying uncertainty about models and for model–data fusion. See Bayesian emulation and Emulation and Uncertainty Analysis. Developments of emulator theory and the associated Bayesian inference methodology and implementation are taking account of the dynamic nature of vegetation models and the consequent propagation of uncertainty, giving a rigorous accounting system for the uncertainties in applications.

Key outputs from the Strand are maps of carbon fluxes and their uncertainties at regional to continental scales, with uncertainty partitioned according to its causes, and based on utilization of the full range of available data that can constrain calculations and reduce uncertainties. The uncertainty calculations rely on the statistical methods developed in the Strand and they will absorb many of the products derived in the other Strands. With them it will be possible to quantify the potential reduction in uncertainty from use of EO and other data sources. Important ancillary outputs include the statistical tools themselves (which will be made available to the wider community), and steadily improving and more powerful biospheric models.