The uncertainty module estimates the level of uncertainty associated with the quality of available data as well as the quality of the predictive model.
Types of uncertainty
The quality of any predictive model is contingent on several factors of knowledge uncertainty: the level in which we understand the processes, the quality of the model structure,the quality of the parameters, & the quality of the data.
The first factor relates to the level of scientific knowns & unknowns regarding a particular process. The second factor involves the scientific rigor of the assumptions on how driving forces influence the output of model simulations. The third & fourth factors relate to the quality of the available data used to develop parameters of environmental drivers & model inputs.
Approach used in uncertainty analysis
The approach used in the CLEAR 2006 uncertainty analysis attempted to isolate the different sources of errors & study their individual as well as combined contribution & relative significance. This was achieved through a simulation-based analysis where the model structure & formulation was fixed & a preselected 'true' set of model parameters was used to run the model. The output of this reference run was considered as 'true' field measurements of salinity values in the different sub-basins. Then, different sources of errors were introduced into the input & calibration data. This approach made it possible to analyze the contribution & interaction of different sources of errors & how they impacted the retrieval of the 'true' model parameters & the model prediction
A number of computational techniques can be used for uncertainty analyses. The CLEAR 2004 uncertainty analysis used the Monte-Carlo method, which is based on the assumption that all the uncertain model parameters & input & output variables are random variables. It is also assumed that the marginal & joint probability distributions of these variables are known. Such distributions are used to generate realizations of the uncertain model variables. The generated values are then used to perform model simulations & produce the desired predictions. This process is repeated many times (typically a few hundred times). The repeated simulations can then be used to construct a probability distribution of the model output.
For more information on Uncertainty, view these publications:An Uncertainty Analysis framework for the CLEAR Ecosystem Model: Using Subprovince 1 as Test Domain & Skill assessment
Uncertainty Assessment of Salinity Predictions in Barataria Estuarine System, Louisiana
Habib, E., et al. 2008. Effect of rainfall spatial variability & sampling on salinity prediction in an estuarine system. Journal of Hydrology 350: 56-67.
Habib, E., et al. 2008 Assessing Effects of Data Limitations on Salinity Forecasting in Barataria Basin, Louisiana, with a Bayesian Analysis. Journal of Coastal Research 23(3):749-763.
An Uncertainty Analysis Framework for the CLEAR Ecosystem Model: Using Barataria Basin as the Test Domain and Skill Assessment
Uncertainty Analysis of the CLEAR Ecosystem Model: Issues Needs, & Future Directions
Model Uncertainty & Limitations