1 of 24

Motivation for understanding extreme coastal water levels

S. Dukes

Gryphon Aerial Services

Southern California Flooding

Willapa Bay Washington Flooding

Northern Oregon Coast Flooding

2 of 24

Peter Ruggiero and Fernando Mendez

Dylan Anderson, Laura Cagigal, Ana Rueda, Jose Antolinez, Mark Merrifield, John Marra, Ian Robertson, Ayesha Genz, Meredith Leung, Patrick Barnard, Li Erikson, Andy O’Neill, Tyler Anderson, Zhenqiang Wang

………………………………………………….

Combining a stochastic climate emulator with surrogate models of dynamic coastal simulators to drive probabilistic coastal flood impacts research

3 of 24

 

San Diego, CA Observational Record (Relatively Short)

Serafin et al., 2014: short records preclude accurate and robust estimates of extreme return level events.

4 of 24

Objective: Develop a probabilistic approach to efficiently evaluate coastal flooding in both today’s climate and a range of possible future climates.

Approach: Hybrid statistical dynamical downscaling techniques taking advantage of a broad range of machine learning and data science advances.

Anderson et al., 2019, Parker et al., 2019, Cagigal et al., 2020, Anderson et al., 2021, Marra et al., 2022

Develop a Coastal Flooding ‘Digital Twin’ via:

Linking a stochastic climate emulator with surrogate models (emulators) of high-fidelity simulators

1.

2.

5 of 24

Stochastic Climate Emulator: Time-varying emulator for short- and long-term analysis of coastal flooding (TESLA; Anderson et al., 2019; JGR: Oceans)

6 of 24

Weather Typing: via identifying dominant modes of spatial variability through EOFs and K-means clustering with respect to Principal Components

Anderson et al., 2019

7 of 24

Chronology Model: Climate-based Autoregressive Logistic Model

Includes ENSO, seasonality, MJO together, and previous two days of synoptic weather together

Guanche et al. (2014), Antolinez et al. (2016)

 

8 of 24

Emulation of local metocean conditions: Sample from Gaussian Copulas for waves and storm surge

Individual waves, winds, surges, etc.

  • Probabilities of occurrence
  • Recurrence of extremes

TWL Hydrographs

9 of 24

TESLA Generated Monte Carlo Simulations of Multivariate Forcing

10 of 24

TESLA Generated Monte Carlo Simulations of Multivariate Forcing

11 of 24

TESLA Generated Monte Carlo Simulations of Multivariate Forcing

12 of 24

TESLA Generated Monte Carlo Simulations of Multivariate Forcing

13 of 24

Not all extreme total water level events are created equal!

There are an infinite combination of drivers (Hs, Tp, Dir, SS, tide, MMSL, and rainfall/river discharge) that can combine to generate the 100-year return period event

100s of 500-year synthetic time series simulated with TESLA

14 of 24

Extreme Value Application: Return Period Analysis of a Proxy of Flooding

Marra et al., 2022

15 of 24

El Ninos

La Ninas

Extreme Value Application: Assessing the influence of climate variability on extreme TWLs

Anderson et al., 2019

16 of 24

Resilience Metric Application: Compute the influence of seasonality on extreme TWL exceedances and impacts

Slope = 0.039

Toe elev. = 3.53 m

Crest elev. 6.14 m

(a)

(b)

(c)

 

Leung et al., in revision

17 of 24

Resilience Metric Application: Compute the influence of sea level rise on extreme TWL exceedances and impacts

Slope = 0.039

Toe elev. = 3.53 m

Crest elev. 6.14 m

(a)

(b)

(c)

 

Leung et al., in revision

18 of 24

Tesla derived input conditions (21 Dimensions)

Hs, Tp, θ,

SLP, U, V

tides

Numerical Model

Hazard Assessment

Teach a statistical model to “behave” like the numerical model

Surrogate models (Emulators) of high-fidelity simulators:

CoSMoS- D3D and XBeach

… machine learning approaches: Gaussian Process Regression;

Neural Networks;

Radial Basis Functions;

Random Forests; etc.

Jia & Taflanidis (2013); Jia et al. (2016); Parker et al. (2019); Goldstein et al. (2019); Anderson et al., 2021

19 of 24

Freeboard

Naval Base Coronado Bayside Application: Surrogate model-derived stochastic future pier use threshold exceedances

Mostly functional

Partially functional

NOT functional

Anderson et al., 2021

20 of 24

Beach width

Naval Base Coronado Oceanside Application: Surrogate model-derived daylight hours available for beach training

Anderson et al., 2021

21 of 24

Dune erosion

Naval Base Coronado Oceanside Application: Surrogate model-derived frequency and style of berm/dune impacts

Anderson et al., 2021

22 of 24

Barrier inundation

Naval Base Coronado Oceanside Application: Surrogate model-derived flood depths during El Niño event and 1.0 m of SLR at 2100

Marra et al., 2022

23 of 24

Naval Base Coronado Oceanside Application: Surrogate model-derived flood depths during El Niño event and 1.5 m of SLR at 2100

Marra et al., 2022

Barrier inundation

24 of 24

Computational reduction via machine learning derived hydrodynamic emulators

Final Thoughts:

Hybrid statistical dynamical modeling, utilizing a range of big data and machine learning approaches, is a powerful approach for producing stochastic, spatially varying extreme total water levels for probabilistic coastal flood risk analysis

Coupled with surrogate models of high-fidelity numerical simulators these approaches can provide a fully stochastic exploration of coastal flooding risk (including compound flooding)

https://github.com/teslakit/teslakit/

Wide range of coastal flood risk applications