Predicting soil-atmosphere CH₄ and CO₂ fluxes across several drained peatland forests with JSBACH-HIMMELI

Abstract. Drainage and other land-use practices, including forestry, substantially influence soil carbon dynamics and associated greenhouse gas (GHG) exchange between boreal peatland soils and atmosphere. Besides the land-use, several site-specific environmental factors, such as water table level (WTL), soil temperature and vegetation regulate the peatland soil GHG balance. It is important to understand the interactions among these co-existing factors at site scale to improve estimates of soil GHG exchange at larger (e.g. regional) spatial scales. Process-based models are useful tools to make large scale estimations, but knowledge about how well the models can capture the variation among drained peatlands, and thus accurately simulate the soil GHG fluxes, are lacking. We simulated key environmental variables (water table level i.e. WTL and soil temperature) and soil–atmosphere methane (CH₄) and CO₂ (heterotrophic respiration) fluxes at six Finnish forestry-drained peatlands with a process-based JSBACH-HIMMELI model. Specifically, our focus was on evaluating the model performance and identifying the sources of uncertainty across multiple sites by using measurement data over several years from each site. The model predicted key environmental variables adequately well at each study site. In addition, we found that the there was more between-site variation in the prediction accuracy with the simulated CH₄ fluxes (NMRSE 1.02–9.25) than with the simulated CO₂ fluxes (NMRSE 0.62–1.24). Our study highlights WTL as the primary source of uncertainty in simulating the CH₄ fluxes. The sites characterized by consistently deep WTL exhibited substantially better agreement between the model and measurement with the soil–atmosphere CH₄ fluxes (RMSE 0.69–1.02 nmol m^-2 s^-1), than sites characterized by higher WTL (RMSE 1.04–9.70 nmol m^-2 s^-1). These discrepancies at high-WTL sites were associated with high anaerobic respiration peaks, which were enhanced by relatively thick modelled acrotelm (i.e. fast decomposing topsoil layer). Our results demonstrate that CH₄ parameterization has only a minor influence on modelled CH₄ fluxes, compared to WTL, but can nevertheless improve prediction accuracy, especially at sites characterized by high WTL.