Hernández, Felipe
(2019)
Integrated High-Resolution Modeling for Operational Hydrologic Forecasting.
Doctoral Dissertation, University of Pittsburgh.
(Unpublished)
Abstract
Current advances in Earth-sensing technologies, physically-based modeling, and computational processing, offer the promise of a major revolution in hydrologic forecasting—with profound implications for the management of water resources and protection from related disasters. However, access to the necessary capabilities for managing information from heterogeneous sources, and for its deployment in robust-enough modeling engines, remains the province of large governmental agencies. Moreover, even within this type of centralized operations, success is still challenged by the sheer computational complexity associated with overcoming uncertainty in the estimation of parameters and initial conditions in large-scale or high-resolution models.
In this dissertation we seek to facilitate the access to hydrometeorological data products from various U.S. agencies and to advanced watershed modeling tools through the implementation of a lightweight GIS-based software package. Accessible data products currently include gauge, radar, and satellite precipitation; stream discharge; distributed soil moisture and snow cover; and multi-resolution weather forecasts. Additionally, we introduce a suite of open-source methods aimed at the efficient parameterization and initialization of complex geophysical models in contexts of high uncertainty, scarce information, and limited computational resources. The developed products in this suite include: 1) model calibration based on state of the art ensemble evolutionary Pareto optimization, 2) automatic parameter estimation boosted through the incorporation of expert criteria, 3) data assimilation that hybridizes particle smoothing and variational strategies, 4) model state compression by means of optimized clustering, 5) high-dimensional stochastic approximation of watershed conditions through a novel lightweight Gaussian graphical model, and 6) simultaneous estimation of model parameters and states for hydrologic forecasting applications.
Each of these methods was tested using established distributed physically-based hydrologic modeling engines (VIC and the DHSVM) that were applied to watersheds in the U.S. of different sizes—from a small highly-instrumented catchment in Pennsylvania, to the basin of the Blue River in Oklahoma. A series of experiments was able to demonstrate statistically-significant improvements in the predictive accuracy of the proposed methods in contrast with traditional approaches. Taken together, these accessible and efficient tools can therefore be integrated within various model-based workflows for complex operational applications in water resources and beyond.
Share
Citation/Export: |
|
Social Networking: |
|
Details
Item Type: |
University of Pittsburgh ETD
|
Status: |
Unpublished |
Creators/Authors: |
|
ETD Committee: |
|
Date: |
19 June 2019 |
Date Type: |
Publication |
Defense Date: |
21 March 2019 |
Approval Date: |
19 June 2019 |
Submission Date: |
5 April 2019 |
Access Restriction: |
1 year -- Restrict access to University of Pittsburgh for a period of 1 year. |
Number of Pages: |
188 |
Institution: |
University of Pittsburgh |
Schools and Programs: |
Swanson School of Engineering > Civil and Environmental Engineering |
Degree: |
PhD - Doctor of Philosophy |
Thesis Type: |
Doctoral Dissertation |
Refereed: |
Yes |
Uncontrolled Keywords: |
Hydrologic modeling, calibration, data assimilation, forecasting, remote sensing, evolutionary computation, Gaussian graphical models, k-means |
Date Deposited: |
19 Jun 2019 13:01 |
Last Modified: |
19 Jun 2020 05:15 |
URI: |
http://d-scholarship.pitt.edu/id/eprint/36358 |
Metrics
Monthly Views for the past 3 years
Plum Analytics
Actions (login required)
|
View Item |