10.20381/ruor-13150
Parker, Geoff
Uncertainty metrics for coupled watershed models
Université d'Ottawa / University of Ottawa
2009
Engineering, Civil.
Université d'Ottawa / University of Ottawa
Université d'Ottawa / University of Ottawa
2013-11-08
2013-11-08
2009
2009
en
Thesis
Source: Dissertation Abstracts International, Volume: 71-05, Section: B, page: 3218.
http://hdl.handle.net/10393/29809
Predictions of river water quality models are subject to substantial uncertainties, which depend not only on parameterization and calibration strategies but also on the structure of the conceptual model itself. Equally, the uncertainty associated with a non-point source (NPS) model can also be attributed to both the parameterization and model structure. The work presented here evaluated the relative importance of these effects and the associated implications for coupled watershed/water quality models. Particular emphasis was placed on model application and interpretation for stochastic-type problems. Investigations were conducted using real-world data from the non-tidal Potomac basin in the Eastern United States for data analysis and case study purposes. In the first sub-project, two commonly used conceptual representations of real-world water quality processes were used, and their simulation of dissolved oxygen (DO), biochemical oxygen demand (BOD) and ammonium (NH4) components were scrutinized. A Generalized Likelihood Uncertainty Estimation (GLUE) approach to the inverse problem was then used to examine how uncertainty changed along the river network for each conceptual model. Differences were observed not only between deterministic instances of each conceptual model, but also between their response surfaces as a whole. In addition, response patterns showed substantial sensitivity to the selection of calibration data used. This methodology was able to identify surface response differences between two conceptual models, and could be used to identify or resolve conceptual model issues for specific applications and/or datasets. Of the two conceptual models examined, the simpler Mike11 model showed marginally better performance at the outlet in most respects, and substantially increased sensitivity to data proximity (i.e. location) throughout the watershed when compared with the more complex QUAL2E representation. Both 'high-fitness' calibrations showed an aggregate coefficient of variation (ACV) of approximately 7 x 10-2 near the outlet. A second sub-project examined instead the suitability of deterministic calibration criteria for stochastic model calibration and uncertainty analysis. Here, the suitability, relative benefits and substantial disadvantages of 'determinstic-optimization' approaches in stochastic contexts were examined. Three alternate calibration strategies, suitable for water quality modeling under uncertainty, were proposed, and then demonstrated. The proposed strategies, based on absolute relative error (ARE) measures, place increased emphasis on lossless memes and the inherent subjectivity of any calibration criteria. Calibration based on these criteria generally showed equal or better performance for stochastic applications when contrasted to root mean square error (RMSE). The findings suggest opportunities for potential improvement. In current calibration paradigms can be obtained by using ARE and related strategies. The areal nature of NPS models and the implications associated with the direct relationship between model structure and parameterization sampling were examined in a third sub-project. The approach was to lind limitations in scale for use of a conceptual model, then examine the scales at which a suitable stochastic depiction of key parameter sets could be obtained. The overlapping regions for model scale are optimal (and possibly the only suitable regions) for conducting meaningful stochastic analysis with a given NPS model. The NPS model examined in this thesis, AnnAGNPS, demonstrated substantial limitations with respect to scale that seem to trump potential parameter uncertainty in many applications. A narrow range of discretizations (in the 4--8.5 km2 scale range) suitable for conceptual model stability was identified. Response to parameter variations within this range was also examined and appears to be minimized within the structurally stable region, near discretizations 8 km2. The final portion of this thesis was a holistic evaluation. Coupled hydrology and water quality models are today an important tool used in the understanding and management of surface water and watershed areas. Such problems are generally subject to substantial uncertainty in parameters, process understanding, and data. Component models, drawing on different data, different concepts, and different structures, are affected differently by each of these uncertain elements. The fourth, and final, subproject draws on the previous three sub-projects, and further proposes a framework wherein the response of component models to their respective uncertain elements call be quantified and assessed, using a hydrological model and water quality model as two exemplars. Appendix A of this thesis provides an overview of the proposed framework in the context of predictive watershed quality and quantity model construction. The resulting assessments can identify model coupling strategies that permit more appropriate use and calibration of individual models, as well as better coupled model construction and interpretation. The results of this portion showed that both QUAL2E and MIKE1l may couple well with the AnnAGNPS model uncertainty at relatively high fitness calibrations. Additionally, structural parameter uncertainty was shown to be at least as significant as parameter-based uncertainties. The results of this work provide a reasonable, manageable, and widely applicable framework for the use of coupled deterministic models to assess uncertain systems. The suggested methodologies, with or without use of the specific codes and tools mentioned, can be readily employed to further understanding of the uncertainty issues, and uncertainty balance, surrounding contemporary, future, and legacy moods.