2021 CSDMS meeting-108: Difference between revisions

From CSDMS
(Created page with "{{CSDMS meeting personal information template-2021 |CSDMS meeting first name=Michael |CSDMS meeting last name=Itzkin |CSDMS meeting institute=University of North Carolina at C...")
 
No edit summary
 
(3 intermediate revisions by 2 users not shown)
Line 50: Line 50:
|State=Oregon
|State=Oregon
|CSDMS meeting coauthor email address=sally.hacker@oregonstate.edu
|CSDMS meeting coauthor email address=sally.hacker@oregonstate.edu
}}
{{CSDMS meeting authors template
|CSDMS meeting coauthor first name abstract=Paige
|CSDMS meeting coauthor last name abstract=Hovenga
|CSDMS meeting coauthor institute / Organization=Oregon State University
|CSDMS meeting coauthor town-city=Corvallis
|CSDMS meeting coauthor country=United States
|State=Oregon
|CSDMS meeting coauthor email address=hovengap@oregonstate.edu
}}
}}
{{CSDMS meeting abstract template 2021
{{CSDMS meeting abstract template 2021
|CSDMS meeting abstract=Coupled process-based numerical models have the potential to greatly enhance our understanding of the drivers of coastal change by allowing for detailed simulations of the processes involved within each model core of the coupling. However, producing accurate hindcasts and forecasts with these coupled frameworks can be challenging due to a wide array of parameters that interact nonlinearly across and within the individual model cores and the potentially substantial computational cost that limits both the number and duration of simulations that can be reasonably performed. Additionally, many model parameters (e.g., wave asymmetry and skewness or sediment transport coefficients) that are critical for model calibration are unitless coefficients in the model formulations and thus cannot be readily measured in the field. Here, we use Windsurf, a coupled beach-dune modeling system that includes Aeolis, the Coastal Dune Model, and XBeach, paired with two surrogate neural network models, to produce a pair of hindcasts and forecasts to replicate observed modes of dune and beach morphological change on a developed barrier island on the US Atlantic coast (Bogue Banks, North Carolina). The first neural network aids in the calibration process by allowing for the prediction of Windsurf’s error surface over thousands of potential parameterizations to rapidly identify a potential best calibration before actually running the model. Windsurf is then run within a genetic algorithm to further hone the collection model parameter settings. Once Windsurf is finished running, we use the output to train a second neural network which contains a Long Short-Term Memory (LSTM) layer to produce five-year forecasts of dune crest height and dune toe elevation. We test our results by comparing them to observed data collected in the field between 2016-2020 using Real-Time Kinematic Global Positioning System (RTK-GPS) and find our forecasts (from the hindcasts) produce reasonably accurate predictions of dune morphology change at interannual scale.
|CSDMS meeting abstract=Coupled process-based numerical models have the potential to greatly enhance our understanding of the drivers of coastal change by allowing for detailed simulations of the processes involved within each model core of the coupling. However, producing accurate hindcasts and forecasts with these coupled frameworks can be challenging due to a wide array of parameters that interact nonlinearly across and within the individual model cores and the potentially substantial computational cost that limits both the number and duration of simulations that can be reasonably performed. Additionally, many model parameters (e.g., wave asymmetry and skewness or sediment transport coefficients) that are critical for model calibration are unitless coefficients in the model formulations and thus cannot be readily measured in the field. Here, we use Windsurf, a coupled beach-dune modeling system that includes Aeolis, the Coastal Dune Model, and XBeach, paired with two surrogate neural network models, to produce a pair of hindcasts and forecasts to replicate observed modes of dune and beach morphological change on a developed barrier island on the US Atlantic coast (Bogue Banks, North Carolina). The first neural network aids in the calibration process by allowing for the prediction of Windsurf’s error surface over thousands of potential parameterizations to rapidly identify a potential best calibration before actually running the model. Windsurf is then run within a genetic algorithm to further hone the collection model parameter settings. Once Windsurf is finished running, we use the output to train a second neural network which contains a Long Short-Term Memory (LSTM) layer to produce five-year forecasts of dune crest height and dune toe elevation. We test our results by comparing them to observed data collected in the field between 2016-2020 using Real-Time Kinematic Global Positioning System (RTK-GPS) and find our forecasts (from the hindcasts) produce reasonably accurate predictions of dune morphology change at interannual scale.
|CSDMS meeting posterPDF= CSDMS_Meeting_2021_Poster_Itzkin.pdf
|CSDMS meeting posterPNG= CSDMS_Meeting_2021_Poster_Itzkin.png
}}
}}
{{blank line template}}
{{blank line template}}

Latest revision as of 12:51, 17 May 2021


No registration fee




Log in (or create account for non-CSDMS members)
Forgot username? Search or email:CSDMSweb@colorado.edu


Browse  abstracts


Modeling interannual dune growth and recovery on a developed barrier island

Michael Itzkin, University of North Carolina at Chapel Hill Chapel Hill North Carolina, United States. mitzkin@unc.edu
Laura Moore, University of North Carolina at Chapel Hill Chapel Hill North Carolina, United States. laura.moore@unc.edu
Peter Ruggiero, Oregon State University Corvallis Oregon, United States. peter.ruggiero@oregonstate.edu
Sally Hacker, Oregon State University Corvallis Oregon, United States. sally.hacker@oregonstate.edu
Paige Hovenga, Oregon State University Corvallis Oregon, United States. hovengap@oregonstate.edu



Coupled process-based numerical models have the potential to greatly enhance our understanding of the drivers of coastal change by allowing for detailed simulations of the processes involved within each model core of the coupling. However, producing accurate hindcasts and forecasts with these coupled frameworks can be challenging due to a wide array of parameters that interact nonlinearly across and within the individual model cores and the potentially substantial computational cost that limits both the number and duration of simulations that can be reasonably performed. Additionally, many model parameters (e.g., wave asymmetry and skewness or sediment transport coefficients) that are critical for model calibration are unitless coefficients in the model formulations and thus cannot be readily measured in the field. Here, we use Windsurf, a coupled beach-dune modeling system that includes Aeolis, the Coastal Dune Model, and XBeach, paired with two surrogate neural network models, to produce a pair of hindcasts and forecasts to replicate observed modes of dune and beach morphological change on a developed barrier island on the US Atlantic coast (Bogue Banks, North Carolina). The first neural network aids in the calibration process by allowing for the prediction of Windsurf’s error surface over thousands of potential parameterizations to rapidly identify a potential best calibration before actually running the model. Windsurf is then run within a genetic algorithm to further hone the collection model parameter settings. Once Windsurf is finished running, we use the output to train a second neural network which contains a Long Short-Term Memory (LSTM) layer to produce five-year forecasts of dune crest height and dune toe elevation. We test our results by comparing them to observed data collected in the field between 2016-2020 using Real-Time Kinematic Global Positioning System (RTK-GPS) and find our forecasts (from the hindcasts) produce reasonably accurate predictions of dune morphology change at interannual scale.
CSDMS Meeting 2021 Poster Itzkin.png

Click on the poster to enlarge