Statistical-Based Hydrological Simulation and Inference

Hydrological models have been used as essential tools for water resources planning and understanding the mechanisms in the water cycle. The rapid advance in information science and computational power in recent decades has encouraged hydrologists to solve water problems through data-driven approaches. To date, various statistical methods have been developed for hydrological simulation and inference (i.e., explain reasonings behind model response). However, many challenges arise from enormous uncertainties and complexities in the hydrological systems that greatly limit the usefulness of statistical models in terms of simulation accuracy and inference. Thus, there is a global need for advanced statistical models for robust hydrological simulation and inference.In this dissertation research, a set of statistical-based hydrological simulation and inference methods has been developed. They have improved upon the existing simulation efforts and helped gain reliable inferences. These methods include (1) a stepwise clustered regression tree ensemble (SCRTE) model that addresses the spatial autocorrelation of daily streamflow; (2) a Wilks feature importance (WFI) method that provides reliable variable rankings for supporting hydrological inference and simulation (through a stepwise clustered ensemble (SCE) model); (3) a baseflow-filtered stepwise clustered ensemble (BFSCE)model that helps gain insights towards sub-hydrological processes (i.e., baseflow, overland flow and interflow) in irrigated watersheds; (4) a joint probabilistic rainfall-runoff (JPRR) model that addresses the high-to-peak flow simulations and projections under climate change. The major contributions of this research are summarized as follows: (1) the proposed SCRTE model can effectively address the temporal autocorrelation of daily streamflow (i.e., a hydrologic effect that was inadequately reflected through conventional statistical models); (2) outstanding simulation performance has been achieved through the proposed SCRTE model compared with many well-known and advanced statistical models; (3) The proposed stepwise clustered ensemble (SCE) model has significantly improved the streamflow simulation performance of stepwise cluster analysis (SCA) by 66.1% based on a large dataset (i.e., 673 basins); (4) The knowledge learned from WFI can be transferred to other statistical models to improve their simulation performance, indicating the “universal fitting” characteristic of WFI inference; (5) a process-based baseflow subtraction (from streamflow) has allowed statistical models to identify the critical information reflecting the overland flow and interflow process and thus help trace the origin of streamflow; (6) the hydrological inference from WFI has shown to be valid for both the entire streamflow process and its sub-processes (e.g., baseflow and overland flow); (7) the proposed JPRR model can be coupled with any existing statistical hydrological models to address high-to-extreme flow simulations and projections; (8) the hydrological projections from the JPRR model provide more reliable future flood risk assessment than conventional statistical models, enabling hydrologic infrastructural design under climate change. Findings and achievements from this research can help water-related decision-making, such as local flood risk management and irrigation schedule optimization.

Name thesis:degree_name

Doctor of Philosophy (PhD)

Level thesis:degree_level

Doctoral -- first

Discipline thesis:degree_discipline

Engineering - Environmental Systems

Grantor dc:publisher

Faculty of Graduate Studies and Research, University of Regina

Year dc:date.issued

2022

Author dc:creator

• Li, Kailong

Advisor dc:contributor.advisor

• Huang, Guo (Gordon)

Committee members dc:contributor.committeemember

• Deng, DianLiang
• Wu, Peng
• Zhu, Hua

Language dc:language.iso

en

DOI

https://doi.org/10.82465/3765

OAI identifier oai:identifier

oai:uregina.scholaris.ca:10294/15566