ML21064A436
Text
Probabilistic Assessment of Multi-Mechanism Floods in Coastal Areas Due to Hurricane Induced Storm Surge, precipitation and River Flow Somayeh Mohammadi1, Michelle Bensi1, Shih-Chieh Kao2, Scott DeNeale2, Elena Yegorova3, Joseph Kanney3 and Meredith L Carr4 (1) University of Maryland College Park, Department of Civil and Environmental Engineering, College Park, MD, United States (2) Oak Ridge National Laboratory, Environmental Sciences Division, Oak Ridge, TN, United States (3) Nuclear Regulatory Commission, Rockville, MD, United States (4) U.S. Army Corps of Engineers, ERDC/CHL, Vicksburg, MS, United States 6th Annual NRC PFHA Research Workshop Online (Feb 22-25, 2021)
Topics addressed Research objective Challenges for multi-mechanism flood (MMF) hazard MMF hazard estimate in this study Probabilistic model (Bayesian motivated approach)
Predictive models Next steps 1
Research objective Estimating flood hazard due to simultaneous occurrence of storm surge, precipitation, and river flow in coastal areas 2
Challenges in Analyzing Multi-Mechanism Floods
- 1. Capturing dependency structure Statistical approaches o Copula o Direct estimation of joint distributions Bayesian approaches
- 2. Capturing the interaction between the flood mechanisms Computationally expensive models 3
MMF analysis in this study Tropical cyclone Storm surge Precipitation Changes in river discharge River base flow Total river discharge Methods for developing joint distributions Bayesian motivated approaches Copula-based approaches Direct estimation of joint distributions Graphical model for this study 4
Probabilistic Model (Bayesian Motivated Approach)
Estimate probability of exceedance for different values of target response variable and generation of hazard curve Calculate joint probability for different combination of discretized values by considering dependency between variables and conditional probabilities Discretizing the error of each model and correcting response variables Discretize the distributions Develop numerical, surrogate or analytical models to predict response variable and conditional distributions Fit distributions to input variables and discretizing them Determine physical relationships between involved variables 5
Predictive models 6
Rmax Xo
Vf CPD Surge Model 1 Vw PRCP Model3 QStorm Model 4 Model 2 Model 5 QRiver Total Q
Predictive models Model 1 (Surge model)
- Surrogate model for predicting surge height using hurricane parameters Model 2 (Wind model)
- Statistical model for predicting maximum wind velocity using central pressure deficit Model 3 (Precipitation model)
- Statistical and empirical model for predicting hurricane induced precipitation Model 4 (Precipitation-induced Discharge Model)
- Statistical model for predicting precipitation induced discharge Model 5 (Surge-induced Discharge Model)
- Statistical model for predicting surge induced river discharge 7
Model 1: Surge Model Predictive (X) variables Vf, CPD, X,,
Response (Y) variable Surge height Model 1 Rmax CPD
Vf Xo Surge Model 1 Vw PRCP Model3 QStorm Model 4 Model 2 Model 5 QRiver Total Q
Delaware River at Trenton 8
Data source: https://chswebtool.erdc.dren.mil/
Save point: 5373
Model 2: Wind model Response (Y) variable Max wind speed Predictive (X) variable Central pressure deficit Model 2 Rmax CPD
Vf Xo Surge Model 1 Vw PRCP Model3 QStorm Model 4 Model 2 Model 5 QRiver Total Q
= 42.4807 0.00842 + 2.9752
= 18.66 Source: North Atlantic Coast Comprehensive Study (NACCS), Nadal-Caraballo et al. 2015 9
Model 3: Precipitation Model Rmax CPD
Vf Xo Surge Model 1 Vw PRCP Model3 QStorm Model 4 Model 2 Model 5 QRiver Total Q
Predictive (X) variables Vf, Vw, X, Response (Y) variable Precipitation depth Model 3 Basin Average daily rainfall 10
Along-Track precipitation Total daily along-track precipitation Along-track precipitation at different hours after landfall 11
Model 4: Precipitation-Induced Discharge Predictive (X) variable PRCP (mm)
Response (Y) variable Q (cfs)
Model 4 Rmax CPD
Vf Xo Surge Model 1 Vw PRCP Model3 QStorm Model 4 Model 2 Model 5 QRiver Total Q
Data :
Simulated daily streamflow, runoff and precipitation data from 1980-2015 (Rapid-VIC model)
Correlation = 0.9831; RMSE_Test = 3.0915e+03 (cfs) 12
Model 5: Surge induced discharge Data:
15 minute data from USGS (Discharge and Water Level)
Station number: 01463500 Rmax CPD
Vf Xo Surge Model 1 Vw PRCP Model3 QStorm Model 4 Model 2 Model 5 QRiver Total Q
Predictive (X) variable Surge height (ft)
Response (Y) variable Q (cfs)
Model 5 Correlation = 0.9831; RMSE_Test = 3.0915e+03 (cfs) 13
Computing Total Discharge (Q)
Rmax CPD
Vf Xo Surge Model 1 Vw PRCP Model3 QStorm Model 4 Model 2 Model 5 QRiver Total Q Superposition:
= +
14
Defining model input parameters N
o.
Parameter Distribution type 1
CPD Doubly truncated Weibull distribution 2
Rmax Lognormal distribution 3
Vf Normal distribution 4
Heading direction Normal distribution 5
Landfall location Uniform distribution 15
Defining input model parameters 16
Statistical analysis of baseflow Gathering daily discharge time-series Removing hurricane event dates from record Randomly sampling a subset of data Performing statistical assessment to define distribution Lognormal distribution as the best fit distribution 17
Decreasing discretization error Surrogate Model Monte Carlo simulation
Generating conditional probability tables (CPTs)
Reducing the impact of discretization error 18
Inclusion of modeling error (epistemic uncertainty)
For all models:
Normal distribution assumed for error Error discretized in to 9 bins Correcting predicted response variables:
= +
19
Estimating annual exceedance frequency
(> )
20
Next steps Considering the non-linearity between surge induced and precipitation induced discharge in the analysis Inclusion of the tides in the analysis Conducting the analysis using other methods (for comparison) oCopula based oDirect estimation of the joint distribution 21
Thank you Somayeh@terpmail.umd.edu mbensi@umd.edu