Calibration of a Land Surface Model to adjust the carbon balance using History Matching.

Land exchanges carbon with the atmosphere through numerous processes, such as photosynthesis and vegetation respiration. As land carbon uptake is greater than land carbon emissions, land surface represents nowadays a carbon sink, absorbing around a third of total anthropogenic carbon emissions every year. Land surface models, used in global climate models, model these processes and provide estimates of net carbon fluxes between the atmosphere and land surfaces. However, simulations of these models still contain significant uncertainties. Other methods, known as Atmospheric CO2 Inversion, exist to estimate these fluxes, and are not consistent with estimates given by land surface models which usually provide a stronger carbon sink in the tropics than Atmospheric CO2 Inversions. These methods cannot provide simulations of the future that are needed for future projections of Climate Models. For that reason, it is important to understand and improve key processes controlling ecosystem carbon budgets, and to embed this understanding in predictive models. Land surface models typically use many free parameters to describe vegetation, which need to be rigorously calibrated or tuned. Here, we aim at calibrating ORCHIDEE, the land surface model used by the IPSL Earth system model, on the atmospheric inversion fluxes (taken as data-driven constraints) in order to study ORCHIDEE's ability to reconcile with Atmospheric CO2 Inversion by finding physically acceptable parameter sets, or detect models’ inability to recover the same spatio-temporal distribution of carbon fluxes. Calibration usually requires many model simulations, which are very costly. Emulators, and especially Gaussian processes, can replace the computationally time-consuming model and help us to run a large number of simulations to fill the parameter space and rule out parameter subspace that give inconsistent simulation. This method, called History Matching, is emerging in the climate community and has shown many advantages. We show the capacity of History Matching to calibrate ORCHIDEE on global simulations using different targets: Known, using twin experiments, which leads to a very rich source of information on parameter sensitivity, uncertainty, equifinality and global and specific knowledge of the model. Unknown, using fluxes from atmospheric CO2 inversion which could also be combined with vegetation activity data (i.e, such as vegetation fluorescence) to add a physical constraint to parameter calibration. This calibration can provide parameter sets that reconcile to a certain extent bottom-up and top-down approaches, or key information on missing processes in ORCHIDEE that need to be added or modified. In both cases, this is highly instructive and leads to a better understanding of the model and processes being modeled and highlights the potential of current land surface models to simulate carbon flux distribution compatible with existing atmospheric CO2 observation (in situ or from satellite).