Multi-fidelity Modeling and Machine Learning Approaches to Compound Flood Risk Assessment

Interdisciplinary Areas: Data and Engineering Applications

Project Description

Characterizing the hazard associated with compound flooding from one or more drivers (e.g., storm surge, rainfall, riverine overflows) is challenging due to the mix of models and statistical techniques commonly used in studies of one driver or another. These differences are particularly stark between coastal and inland studies, but transition zones exist with appreciable hazard from storm surges, pluvial flooding, fluvial flooding, and co-occurring events of each type. Risk analyses to inform policy also may wish to consider how hazard will change under time amidst uncertainty about future climate conditions or with mitigation infrastructure (e.g., levees) in place. This can yield a large number of landscapes to model, limiting the number of events that can be simulated on each landscape.

 
This postdoctoral project will work to overcome these limitations in two ways: 1) advancing new techniques for multi-fidelity, multi-model simulation (i.e., running a large number of simulations through a computationally cheap, lower-fidelity model to inform samples of events to simulate with a higher-fidelity model); and 2) developing machine learning approaches to predict inundation as a function of storm parameters, landscape parameters, and boundary/antecedent conditions (e.g., sea levels, base inflows, soil moisture) as well as to directly predict inundation hazard curves.
 

Start Date

July 2024
 

Postdoc Qualifications

Experience with the following techniques and models is desirable:
- ADCIRC+SWAN and/or the HEC suite of models (-RAS, -HMS, -FIA, etc)
- Statistical techniques for flood risk assessment and mapping (JPM-OS, POT, etc)
- Machine learning techniques for surrogate/response surface modeling
- Advanced to expert proficiency in Python and/or R
- PhD in civil engineering, hydrology, risk analysis, or related fields 
 

Co-Advisors

David R Johnson (davidjohnson@purdue.edu) - School of Industrial Engineering (primary) and Department of Political Science (minority)

Venkatesh Merwade (vmerwade@purdue.edu) - Lyles School of Civil Engineering 
 

Short Bibliography

Al Kajbaf, A. and Bensi, M., 2020. Application of surrogate models in estimation of storm surge: A comparative assessment. Applied Soft Computing, 91, p. 106184.

Jafarzadegan, K. and Merwade, V., 2019. Probabilistic floodplain mapping using HAND-based statistical approach. Geomorphology, 324, 1 January 2019, p. 48-61.

Johnson, DR; Fischbach, JR; Geldner, NB; Wilson, MT; Story, C; Wang, J. Coastal Master Plan: Attachment C11: 2023 Risk Model. Version 2, p. 1-33. (https://coastal.la.gov/our-plan/2023-coastal-master-plan/2023-plan-appendices/)

N. Geldner, D. Johnson, G. Villarini, B. Yuill, A. Saharia, S. Zou, L. Grimley, N. Young, M. McManus, H. Roberts, S. Misra. “Applied Joint Probabilistic Modeling of Compound Coastal Flood Hazard: An Extension of the Joint Probability Method with Optimal Sampling.” 14th International Conference on Applications of Statistics and Probability in Civil Engineering, ICASP14. Dublin, Ireland. July 9-13, 2023.