Uncertainty and sensitivity analysis
Increasingly, stakeholders in carbon cycle analyses – model developers, policy makers and buyers of carbon credits – are recognizing the importance of including estimates of uncertainty in their analyses. Quantifying the uncertainty of model output and indicating how the uncertainty can be apportioned to different sources (sensitivity analysis) is necessary for model validation, and uncertainty analysis is required under various reporting and trading schemes (such as the 2006 IPCC Guidelines for National Greenhouse Gas Inventories and California’s Climate Registry Forest Project Protocol). In addition, effective applications of the model can be enhanced if the user is made aware of inputs to which the model is particularly sensitive, and thus where future efforts in reducing uncertainty should be concentrated.
The sensitivity of the dead-organic matter sub-module of the operational scale version of the Carbon Budget Model of the Canadian Forest Sector (CBM-CFS3) to variation in model parameters has been assessed in two ways:
- The first approach was based on the use of sensitivity and uncertainty analysis software packages that are freely available on the internet and accessible to model users. Impacts of parameter variation on stocks and fluxes were assessed using four simulated landscapes and three species. Two rotation lengths were used to evaluate the effects of interactions between modeled scenarios and assumptions about parameter variability. Results of the analysis showed that the model was sensitive to variation in parameters controlling the foliage and fine root pathways, but the sensitivity differed depending whether a softwood or hardwood landscape was being simulated. These findings indicate that inferences drawn from sensitivity and uncertainty analysis of stock and flow forest carbon models are specific to the landscapes and time horizons being modeled.
- The second approach assessed the impact on slow pool stocks by systematically varying each of parameters in the dead organic matter sub-module by 5%. More parameters were tested using this approach, but only 1 parameter could be varied at a time. Base decay rates, temperature modifiers, decay proportions to the atmosphere and fuel consumption coefficients were included for sensitivity testing, using model simulations representing soil plot data in Alberta and Ontario. The model was shown to be sensitive to changes in slow pool decay rates, and changes in parameters controlling foliage and fine root pathways. Sensitivity differed depending on whether a softwood or hardwood landscape was being simulated.