Abstract:
My dissertation focuses on evaluating model uncertainties when numerically simulating surface water-groundwater (sw-gw) interactions at different scales. To do so, I mainly use HydroGeoSphere, a physically-based distributed finite element model that fully couples variably saturated subsurface flow with surface water flow. I evaluate three predominant model uncertainty types at three different scales of sw-gw interaction. For each of these investigations, I selected a corresponding study site.
Firstly, I evaluate structural (conceptual) uncertainty from delineating baseflow contribution areas to gaining stream reaches, or stream capture zones, at the catchment-scale. I investigate how the delineated stream capture zone (in the Alder Creek watershed) can differ due to the chosen model code and delineation method (Chow et al. 2016). The results indicate that different models can calibrate acceptably well to the same data and produce very similar distributions of hydraulic head, but can produce different capture zones. The stream capture zone is highly sensitive to the post-processing particle tracking algorithm. Reverse transport is an alternative and more reliable approach that accounts for local-scale parameter uncertainty and provides probability intervals for the stream capture zone. The two methods can be combined to enhance the overall confidence in the delineated stream capture zone.
Secondly, I evaluate parameter uncertainty when simulating meander-scale hyporheic exchange by conducting a model-based sensitivity (bathymetry) and uncertainty (subsurface K-distribution) analysis. I select the Steinlach River Test Site for demonstration. I conduct a sensitivity analysis to determine the aspects of river bathymetry that have the greatest influence on the predictive biases (Chow et al. 2018). Results indicate that simulating hyporheic exchange with a high-resolution detailed bathymetry using a 3D fully coupled sw-gw model leads to nested multi-scale hyporheic exchange systems. A poorly resolved bathymetry will underestimate the small-scale hyporheic exchange, biasing the simulated hyporheic exchange towards larger scales, thus leading to overestimates of hyporheic exchange residence times. The detailed river slope alone is not enough to accurately simulate the locations and magnitudes of losing and gaining river reaches. Thus, local bedforms in terms of bathymetric highs and lows within the river are required. Incorporating local bedforms will likely capture the nested nature of hyporheic exchange, leading to more physically meaningful simulations of hyporheic exchange fluxes and transit times.
Additionally, I conduct an uncertainty analysis to evaluate the trade-offs between intrinsic (irreducible) and epistemic (reducible) model errors when choosing between homogeneous and highly complex subsurface parameter structures (Chow et al. 2019). Results indicate that, if the parameter structure is too simple, it will be limited by intrinsic model errors. By increasing subsurface complexity through the addition of zones or heterogeneity, we can begin to exchange intrinsic for epistemic errors. Thus, choosing the appropriate level of detail to represent the subsurface parameter distributions depends on the acceptable range of intrinsic structural errors for the given modelling objectives and the available site data.
Thirdly, I evaluate data uncertainty at the bedform-scale in a fractured rock setting by modelling a conservative tracer experiment at the Eramosa Bedrock River Site. I use a stochastic discrete fracture network framework to represent the subsurface fractured bedrock connectivity, and in doing so produce a probabilistic distribution of the potential hyporheic exchange extents and residence times. The results indicate that the coincidence of fractures and hydraulic gradients determine the spatial extents of bedform-scale hyporheic exchange. Furthermore, hyporheic exchange residence times in bedrock rivers at the bedform-scale are potentially orders of magnitude longer when compared to fluvial rivers (i.e., months to years vs. minutes to hours).
In the age of highly-parameterized integrated hydrological models, there is an increasing need to understand whether we are getting the right forecasts for the right reasons. Modelling is only a single tool in the scientific toolbox that works best when combined with others, e.g., field and laboratory experiments. The next generation of integrated hydrological modellers will face evermore challenging objectives, which to achieve will require multi-disciplinary teams. Regardless of the origins of your knowledge-base, it is important to always approach numerical models with a healthy level of skepticism. Ultimately, scientists should approach models with a relentless fearlessness to falsify them, which is an absolute necessity if we wish to continually push the envelope of scientific knowledge.