Complex Land Surface Models (LSMs) rely on a plethora of parameters. These parameters and the associated process formulations are often poorly constrained, which hampers reliable predictions of ecosystem dynamics and climate feedbacks. Robust and uncertainty-aware parameter estimation with observations is complicated by, for example, the high dimensionality of the model parameter space and the computational cost of LSM simulations. Herein, we adapt a novel Bayesian data assimilation and machine learning framework termed ‘calibrate, emulate, sample‘ (CES) to infer parameters in a widely-used LSM coupled with a demographic vegetation model (CLM-FATES). First, an iterative ensemble Kalman smoother provides an initial estimate of the posterior distribution (‘calibrate‘). Subsequently, a machine-learning-based emulator is trained on the resulting model-observation mismatches to predict outcomes for unseen parameter combinations (‘emulate‘). Finally, this emulator replaces CLM-FATES simulations in an adaptive Markov Chain Monte Carlo approach enabling computationally feasible posterior sampling with enhanced uncertainty quantification (‘sample‘). We test our implementation with synthetic and real observations representing a boreal forest site in southern Finland. We estimate a total of six plant-functional-type-specific photosynthetic parameters by assimilating evapotranspiration (ET) and gross primary production (GPP) flux data. CES provided the best estimates of the synthetic truth parameters when compared to data-blind emulator sampling designs while all approaches reduced model-observation errors compared to a default parameter simulation (GPP: -10% to -30%, ET: -4% to -6%). Although errors were also consistently reduced with real data, comparing the emulator designs was less conclusive, which we mainly attribute to equifinality and insufficient experiment complexity.

Extensive regions in the permafrost zone are projected to become climatically unsuitable to sustain permafrost peatlands over the next century, suggesting transformations in these landscapes that can leave large amounts of permafrost carbon vulnerable to post-thaw decomposition. We present three years of eddy covariance measurements of CH4 and CO2 fluxes from the degrading permafrost peatland Iskoras in Northern Norway, which we disaggregate into separate fluxes of palsa, pond, and fen areas using information provided by the dynamic flux footprint in a novel ensemble-based Bayesian deep neural network framework. The three-year mean CO2-equivalent flux is estimated to be 106 gCO2 m-2 yr-1 for palsas, 1780 gCO2 m-2 yr-1 for ponds, and -31 gCO2 m-2 yr-1 for fens, indicating that possible palsa degradation to thermokarst ponds would strengthen the local greenhouse gas forcing by a factor of about 17, while transformation into fens would slightly reduce the current local greenhouse gas forcing.

Stable boundary layers commonly form during Arctic polar night, but their correct representation poses a major challenge for numerical weather prediction (NWP) systems. To perform detailed model verification by probing the lower boundary layer, airborne fiber-optic distributed sensing (FODS), tethered sonde and ground-based eddy-covariance measurements are carried out during contrasting synoptic forcings in a fjord-valley system in Svalbard. The FODS-derived turbulent potential energy and static stability profiles are used to investigate the spatial and temporal evolution of different inversion types. The observed vertical temperature and wind speed profiles are compared to two configurations of the HARMONIE-AROME system with different horizontal resolutions of 2.5 km and 0.5 km. The higher-resolved model captures cold pool and low level jet formation during weak synoptic forcing, resulting in a well-represented vertical temperature profile, while the coarser model exhibits a warm bias in near-surface temperatures up to 8 K. During changing background flow, the higher-resolved model is more sensitive to misrepresented wind directions. The results indicate the importance of the ratio between nominal model resolution and valley width to represent stable boundary layer features. Kinetic and potential energy spectra are examined for the two model configurations to derive the effective resolutions. The higher-resolved model has also a higher effective resolution, but is more diffusive than the coarser model. Our results underline the substantial benefit of spatially resolving FODS measurements for model verification studies and underline the importance of model and topography resolution for accurate representation of stable boundary layers in complex terrain.

We measured COs-1 which corresponds to a growing season estimate of 0.04 to 0.16 g CH4 m-2. We find that this moss tundra emits about 60 gCO2-equivalents m-2 yr-1 of which CH4 is responsible for 7%. Air temperature, soil moisture and greenness index contributed significantly to explain the variation in ecosystem respiration (Reco) while active layer depth, soil moisture and greenness index best explained CH4 emissions. Estimate of temperature sensitivity of Reco and gross primary productivity showed that a modest increase in air temperature of 1 degree did not significantly change the NEE during the growing season but that the annual NEE would be even more positive adding another 8.5 gC m-2 to the atmosphere. We tentatively suggest that the warming of the Arctic that has already taken place is partly responsible for the fact that the moist moss tundra now is a source of CO2 to the atmosphere.