Feedback versus uncertainty

Session C1: Compexity, Sensitivity, and Uncertainty Issues in Integrated Environmental Models

Even without uncertainty about the model structure or parameters, the output of a hydrological model run still contains several sources of uncertainty. These are: measurement errors affecting the input, the transition from continuous time and space to discrete time and space, which causes loss of information about the input, discretization of the model equations resulting in errors due to the discretization scheme and the use of finite precision calculations in model evaluation. Interval analysis can provide upper bounds on the output error due to all of these sources. This paper focuses on tracking uncertainty about input values and the effects of finite precision calculation. A conventional hydrological model was recoded in interval arithmetic and run with small, but non-zero uncertainty bounds on the input time series values to see whether the internal feedback loops would keep the output uncertainty bounded.