Assessing the role of parameter interactions in the sensitivity analysis of a model of peatland dynamics

Identifying the processes controlling the carbon balance in northern peatlands is crucial for their integration in global climate models. As a first step, current models of peatlands dynamics need to be adequately evaluated to verify the consistency of processes before their integration in global models.We compared here the sensitivity of the total carbon mass of a peatland simulated by the Holocene Peat Model (HPM) derived from different sensitivity methods: a ‘classic’ local sensitivity analysis method and a global approach with the screening method.We observed that the conclusions drawn by the different methods are very different and, moreover, that it is not advisable to draw any general conclusion from a local one-at-a-time (OAT) experiment because the model space represented is very limited. The results also stressed that the representation of several processes, such as runoff or peat decomposition under different saturation conditions, lack sufficient empirical data and knowledge to be adequately represented in the model. In addition to the evaluation of the model per se, the exploration of its behaviour allowed us to observe simulations of two fen-bog transition patterns.This study showed that interactions between parameters should be taken into account when evaluating peatland dynamics models. The results of the screening method are useful for model improvement and simplification. Moreover, this method enables the exploration of model processes in detail, thus providing insight into peatland system dynamics.

► We analysed the sensitivity of the Holocene Peat Model total carbon mass to the different model parameters.
► The results of two different methods (one local sensitivity analysis and a screening method) are compared.
► The method allows the assessment of dynamical processes and their comparison with the current knowledge.
► Different peatland development pathways could be identified in the results.
► Global sensitivity analysis should be preferred, so that interactions between parameters are taken into account.