Addressing Uncertainty in Understanding Hydroclimate, Hydrology and Hydraulics Across Scales
出版項
2020
說明
1 online resource (207 pages)
文字
text
無媒介
computer
成冊
online resource
附註
Source: Dissertations Abstracts International, Volume: 82-07, Section: B
Advisor: Ivanov, Valeriy Y
Thesis (Ph.D.)--University of Michigan, 2020
Includes bibliographical references
Outputs from the Global Climate Model are always used to study the impacts of climate change on the hydrological cycle. But the inferences suffer from significant uncertainties, which need to be addressed before any robust conclusion of climate change can be made. Specifically, I applied a Bayesian weighted averaging (BWA) method to investigate the change of peak annual runoff seasonality in the future period based on outputs from GCMs contributing to CMIP5. Because of the circular nature of the time variable, we modified the BWA method before we apply it. A high-quality daily runoff dataset is used in this BWA framework to reduce the bias of model projections. Based on the Bayesian inference, we identified a robust spatial pattern for the change of peak annual runoff timing, which is attributed to the change of snowmelt and soil wetness seasonality due to increased temperature. Evapotranspiration (ET) is a crucial component of the surface water, and energy balance, whose reliability of estimation at any scale remains challenging since observations of ET are sparse in space and time. Therefore, models are required to simulate ET at any scale using in situ or remote sensing observations. This dissertation applied a novel method based on the Maximum Entropy Production (MEP) theory to estimate ET, which requires only net radiation, temperature, and specific humidity as inputs. Using site-level eddy flux data in the Amazon rainforest, the MEP method shows high skill at the hourly, daily, and monthly scales. Consistent performance under different levels of land-surface dryness is revealed, hinting that drought signal is appropriately resolved. The site-level MEP-based estimates outperform the estimates of the MODIS ET product, which is commonly used for large-scale assessments. We then applied MEP to project the change of ET in the future with GCMs' forcing. MEP based projections are shown to be more robust than the original GCMs' ET projections, implying the uncertainty of ET in climate models can be reduced by using a parsimony algorithm. This dissertation has also presented a framework for quantifying the uncertainty of urban flooding simulation. A physically rigorous "hyper-resolution" hydraulic and hydrologic model - tRIBS-OFM - is used here to simulate flood propagation, advance numerical representation and understanding of interactions between flooding and the urban environment. Due to the steep computational cost and constraints associated with resolving the 2D Saint-Venant equations at very high resolutions, no effort has been made to address the uncertainty systematically. The uncertainty quantification remains challenging even the model is run in parallel with multiple cores. We approach this problem by training a surrogate model for tRIBS-OFM. Specifically, this surrogate model relies on polynomial chaos expansions, which creates a mapping of flooding outputs from the uncertain inputs. The surrogate model is very computationally inexpensive, therefore, the uncertainty in the inputs/parameters can be propagated to outputs efficiently through the surrogate model. Within this uncertainty quantification framework, we propose a real-time high-fidelity urban flooding forecasting framework, which is able to predict near-instantaneous quantities of interest (e.g., river discharge, inundation field) given a forecasted rainfall with uncertainty warranted. For this study, we reproduce a flooding event in a catchment located in the city of Houston during the 2017 Harvey event and validate the performance of the uncertainty quantification framework with streamflow, high watermarks, and inundation depth data
Electronic reproduction. Ann Arbor, Mich. : ProQuest, 2021
