Correspondence Between Long Term Carbon Sequestration and Measurable Variables in a Global Land Surface Model

The response of net atmosphere-land carbon exchange under future warming and increasing CO2 is key to the projection of future climate change. However, current land-surface model differ widely in their prediction of the land carbon sink by 2100. These models are increasingly complex and entail a large array of mechanisms. Consequently, the number of “knobs”(i.e. model parameters) available to tune model results has increased drastically. In principal, objectively tuning all parameters of a model to the measurements at hand should yield a best configuration. But in practice, it is important to know structure of data that helps best to improve a model’s long-term carbon sink trajectory; or alternatively whether there are variables where a model data mismatch would not necessarily compromise the model outcome.

We performed a sensitivity analysis of LM3VN, a land surface model with a prognostic nitrogen cycle, by varying 60 parameters, and checked for correspondence between the sensitivity of the model’s long-term (1850-2100) carbon sink and contemporary (1980-2006) calibration variables. We found, that few parameters had a strong impact on the long term carbon sequestration, showing that the model entails a number of negative feedbacks. Importantly, the parameters to which the model was most sensitive were found to vary between individual gridcells, supporting the idea of point-specific and regional model assessment. The model’s prediction of the current total carbon inventory correlated well with the prediction of the long term carbon sink, indicating that evaluation of models against current carbon inventories could improve their prediction of carbon sequestration over the this century, although the aggregation of such data is challenging. A promising correspondence is that of the interannual variability of net carbon exchange, we found this the correlation to be significant in a majority of gridcells (73%) but weak if globally aggregated. Overall, such targeted sensitivity analysis may help to select data sets and inform observation networks in order to constrain the response of the terrestrial carbon cycle to global change factors, particularly if broadened across models and scenarios.