CHARACTERIZATION OF UNCERTAINTY IN MODEL PARAMETER AND PRECIPITATION DATA ACROSS SEVERAL HEADWATER CATCHMENTS IN THE CANADIAN ROCKIES: A LARGE-SAMPLE HYDROLOGY APPROACH

Abstract

Hydrologic modelling and prediction in the Canadian Rookies are hampered by the sparsity of hydro-climatic data, limited accessibility, and the complexity of the cold regions hydrologic processes. Previous studies in this region have mainly focused on very few heavily instrumented catchments, typically with limited generalizability to other catchments in the region. In this thesis, I adopt a “large-sample hydrology” approach to address some of the outstanding issues pertaining to data uncertainty, model parameter identifiability, and predictive power of hydrologic modelling in this region. My analyses cover 25 catchments with a range of physiographic and hydrologic properties located across the Canadian Rockies. To address forcing data uncertainty, which is commonly considered as the most dominant source of uncertainty in the hydrology of this region, I processed and utilized three different gridded-data products, namely ANUSPLIN, CaPA, and WFDEI. To make the problem tractable, I applied an efficient-to-run conceptual hydrologic model to simulate the hydrologic processes in this region under a variety of parameter and input data configurations. My analyses showed significant discrepancies in precipitation amounts between the different climate data products with varying degrees across the different catchments. Runoff ratios were quite variable under the different products and across the catchments, ranging from 0.25 to 2, highlighting the significant uncertainty in precipitation amounts. To handle precipitation uncertainty in hydrologic modelling, I developed and tested two strategies: (1) implementing a correction parameter for each data product separately, and (2) developing and parameterizing a linear combination of the different data products to have a unified, presumably more accurate data product. These new precipitation-correcting parameters along with a selected set of the hydrologic model parameters were analyzed and identified via Monte-Carlo simulation, considering three model performance criteria on streamflow simulation, namely Nash-Sutcliffe Efficiency (NSE), NSE on log-transformed streamflow (NSE-Log), and Percent Bias (PBias). Overall, the hydrologic model showed adequate performance in reproducing observed streamflows in most of the catchments, with NSE, NSE-Log, and PBIAS ranging in 0.36-0.87, 0.43-88, and 0.001%-34%, respectively. However, most of the model parameters showed limited identifiability, limiting the power of the model for the assessment of climate and land cover changes. Overall, WFDEI climate data provided the best performance in parameter identification, while demonstrating a superior performance in reproducing observed streamflows.