Quantifying permafrost C-cycling by fusing process-models and observations

Globally permafrost soils store huge quantities of carbon (C) in dead organic matter (DOM). Currently, the permafrost region is estimated to be a small net C sink. However, as the climate warms permafrost soils have begun to thaw, making a massive quantity of DOM available for potential decomposition and likely shifting the region to a net source of C. Process-models of terrestrial ecosystems are a vital tool in evaluating our understanding of ecosystem function, but also in generating forecasts of C emissions under varied climate change scenarios in support of decision support. But different models contain competing hypothesise of ecosystem functioning, leading to divergent forecasts despite convergent estimates of contemporary net C emissions. These process-models also result in contrasting estimates of the internal C-cycling. We currently lack a consistent, rigorous observational constraint on ecosystem C-stocks and dynamics (particularly below ground) due to varied challenges across both in-situ and satellite-based Earth Observation (EO). Here, we present a Bayesian model-data fusion approach (CARDAMOM) which combines diverse observations of terrestrial ecosystems (e.g. leaf area, soil C, biomass, net C exchange) to calibrate an intermediate complexity model (DALEC). CARDAMOM generates a probabilistic estimates of DALEC parameters at pixel scale based on local information. Using these local calibrations, DALEC offers a probabilistic, data-constrained estimate of current ecosystem C-cycling including its internal dynamics, which can be used to evaluate large scale process-models. We evaluate process-model estimates of key ecosystem properties, e.g. DOM residence time, and their climate sensitivity. Through this process we can identify and exclude process-models which are inconsistent with data from forecast analyses.