CARDAMOM (CARbon DAta MOdel fraMework) is a computer programme that retrieves terrestrial carbon (C) cycle variables by combining C cycle observations with a mass balance model. CARDAMOM produces global dynamic estimates of plant and soil C pools, their exchanges with each other and with the atmosphere, and C cycling variables for processes driving change. A Bayesian method is used to retrieve model parameters that statistically reduce the difference between model outputs and C observations for each model cell. The outcome is a probabilistic assessment of the fluxes, pools and process variables of the C cycle in each cell.
CARDAMOM produces a C cycle analysis consistent with C measurements and climate. It is well suited for using with global-scale satellite observations, for instance aboveground biomass or leaf area index. CARDAMOM produces a confidence interval on all retrieved quantities, so that there is a clear assessment of model-data consistency, and the value of assimilating extra datasets can be quantitatively assessed.
Conventional C cycle estimates rely on prescribed C cycle parameters; CARDAMOM retrieves parameters by combining models with data. The implementation of the CARDAMOM family of C models, (DALEC models), differs from typical model implementations: there is no spin-up for biomass and soil C, and no prescribed plant functional types or C pool steady state. CARDAMOM generates maps of key model parameters as an output, rather than as an input. Rather than being compared to data, the CARDAMOM outputs represent the best fit to all data within each CARDAMOM cell.
CARDAMOM has several components. First is a C mass balance model, DALEC, that has four vegetation pools and two dead organic matter pools (these pools can be adjusted to test alternate structures). There are around 20 model parameters associated with phenology (plant timing), allocation, residence times, temperature sensitivity and productivity. The second component is a system to arrange and organise spatial data related to forcing (climate, burned area, management) and to observations for assimilation (currently LAI, biomass, soil C). The third component is the Markov Chain Monte Carlo code, that operates on each model grid cell to retrieve model parameter vectors consistent with data. The fourth component is a structure to manage model-data interactions, distribute jobs to parallel computing nodes, retrieve state and process variables from optimized model parameters, and to summarise results for analysis.