Conventional methods for uncertainty propagation, such as the standard Monte Carlo and Latin Hypercube Sampling methods, may be computationally prohibitively expensive, as described in more detail in Chapter 2. The development of alternative methods that are applicable to a wide range of transport-transformation models could substantially reduce the computational costs associated with uncertainty propagation (in terms of both the time and the resources required). In fact, such methods could facilitate the uncertainty analysis of complex, computationally demanding models where the application of traditional methods may not be feasible due to computational and time limitations. Additionally, there is a need for an extensive evaluation of the alternative methods with realistic case studies. Such an evaluation addresses the issues associated with the wider applicability of the methods. Further, the ease of use of any method is an important factor in its applicability. The extra effort required in understanding and applying a new method can sometimes offset other advantages that the method has to offer. Hence, attention must be paid to the development of auxiliary tools that facilitate the easy use of the new methods.
Thus, the primary objective of this thesis can be stated as
``the development of computationally efficient alternative methods for uncertainty propagation that are applicable to a wide range of transport-transformation models, and the development of auxiliary tools that facilitate easy use of these methods.''
Another objective of this thesis is to study the uncertainties associated with model structure and formulation, which fall beyond the scope of input uncertainty propagation. Such a study provides help in the selection of the appropriate level of model detail. Further, in conjunction with uncertainty propagation, it can be used to identify the relative magnitudes of uncertainties associated with model inputs and model formulation. This provides information as to where available resources should be focused: filling data gaps versus refining the model used for the problem.
These objectives are accomplished via the development of the Stochastic Response Surface Method (SRSM), and its evaluation for a range of transport-transformation models. Additionally, an easy to use, web-based interface is developed to facilitate black-box use of this method. Furthermore, the SRSM is coupled with an existing sensitivity analysis method, ADIFOR (Automatic DIfferentiation of FORtran), in order to further improve the computational efficiency of the SRSM. Finally, uncertainties associated with model structure and formulation are addressed in the framework of photochemical air quality modeling.