Link to the University of Pittsburgh Homepage
Link to the University Library System Homepage Link to the Contact Us Form

Understanding Hydrologic Processes and Correlations using Modeling and Machine Learning with Remote Sensing and In-Situ Wireless Sensor Network Data

Villalba Fernandez de Castro, German Augusto (2019) Understanding Hydrologic Processes and Correlations using Modeling and Machine Learning with Remote Sensing and In-Situ Wireless Sensor Network Data. Doctoral Dissertation, University of Pittsburgh. (Unpublished)

Download (4MB) | Preview


This work addresses three challenging issues about the overall applicability of hydrologic modelling. The first challenge is improving the collection of sub-surface data. Our approach uses a long-term deployment of wireless sensor network with environmental sensors. This approach is cost-effective when compared with the use of data-loggers and more flexible as it allows real-time monitoring of environmental variables. The plot scale environmental data is collected from our own WSN, deployed in western Pennsylvania, currently composed by 104 nodes and over 240 sensors including commercially available soil moisture, water potential and temperature sensors along with lab-made xylem sap flow sensors.
The second challenge is improving the availability and accuracy of continuous streamflow time-series estimates. The hydrometric network is modelled as a sparse Gaussian graphical model where each site represents a node in a graph. The graph model will have an edge between two sites only when their streamflow time-series are conditionally dependent given the other sites. A novel algorithm is presented, estimating a sparse graph by imposing sparsity to the precision (covariance inverse) matrix via the Graphical Lasso algorithm. The resulting graph is used for inference and a second algorithm determines which gauges can be removed with the least loss of information. The estimated streamflow time-series have better accuracy that other methods based on geographic proximity (least distance) or marginal correlation.
The third challenge is estimating the soil-water characteristics from biased and noisy observations of soil moisture. A novel method is presented for the simultaneous estimation of soil moisture and soil-related parameters. The simulation of soil moisture is performed using the Noah and the VIC models. The simulated site is a well-documented testbed in the state of Oklahoma. The calibration of the soil-related parameters uses Machine Learning techniques such as clustering, regression and classification, and soil-water correlations, providing physical and statistical constrains in the parameter space. Thus, the search is made within a reduced parameter space which makes the parameter calibration approach more effective and realistic. The performance of
the calibration algorithm is assessed regarding the quality of the soil moisture estimations while keeping the parameters in a feasible range


Social Networking:
Share |


Item Type: University of Pittsburgh ETD
Status: Unpublished
CreatorsEmailPitt UsernameORCID
Villalba Fernandez de Castro, German Augustogev5@pitt.edugev50000-0001-6808-6558
ETD Committee:
TitleMemberEmail AddressPitt UsernameORCID
Committee ChairLiang, Xxuliang
Committee MemberKhanna, Vkhannav@pitt.eduKHANNAV0000-0002-7211-5195
Committee MemberLin,
Committee MemberMao,
Date: 10 September 2019
Date Type: Publication
Defense Date: 15 July 2019
Approval Date: 10 September 2019
Submission Date: 25 July 2019
Access Restriction: 3 year -- Restrict access to University of Pittsburgh for a period of 3 years.
Number of Pages: 159
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: WSN Machine Learning Hydrological Modeling Graphical Lasso Streamflow SMC Soil Moisture Content
Date Deposited: 10 Sep 2019 18:53
Last Modified: 10 Sep 2022 05:15


Monthly Views for the past 3 years

Plum Analytics

Actions (login required)

View Item View Item