Presentation Abstract
There are serious concerns about ecological, social, and economic impacts in the Pacific Northwest due to Ocean Acidification (OA). We built a system to predict aragonite saturation state (Ω) of seawater in Netarts Bay, Oregon based on large scale forcing parameters. An artificial neural network – trained against a continuous, multiyear monitoring record of carbonate chemistry – learns a regression estimate of Ω based on seasonality, tides, and wind conditions. This approach is agnostic to the details of the underlying chemical and biological processes offering a distinct modelling perspective. The result is a conceptually simpler and more strictly empirical parameterization and a model that is flexible in application due to dependence on only easily obtainable parameters. Forecast validation by a cross validation method indicates good prediction performance, particularly for the high frequency content of the Ω time series, over periods of stable wind forecasting. Our forecast model demonstrates that the complex temporal dynamics of carbonate chemistry within an estuary can emerge from forcing operating on longer timescales. This further elucidates the management and commercial value of this model; experimental work with calcifiers suggests the details of these high frequency chemical dynamics are critical to the magnitude of stress imposed. Lastly, these forecasts, deployed as a web application, can facilitate OA mitigation strategies by providing aquaculturists with real-time predictions for consideration in operational decisions. Numerous sites, including on the Salish Sea, are poised to soon have viable training data for application of this method. Broader deployment promises to enable comparison between sites and expansion of direct aquaculture and management applications. Expansion to other sites is expected to require altered explanatory variables but this exercise may itself yield insight. Relatedly, we note the potential of this approach to help constrain timescales and sources (natural and anthropogenic) of contributions to physiological OA stress.
Session Title
Ocean Acidification: Modeling and Predictions
Keywords
Aragonite saturation, Ocean acidification
Conference Track
SSE5: Climate Change: Impacts, Adaptation, and Research
Conference Name
Salish Sea Ecosystem Conference (2018 : Seattle, Wash.)
Document Type
Event
SSEC Identifier
SSE5-655
Start Date
5-4-2018 2:15 PM
End Date
5-4-2018 2:30 PM
Type of Presentation
Oral
Genre/Form
conference proceedings; presentations (communicative events)
Contributing Repository
Digital content made available by University Archives, Heritage Resources, Western Libraries, Western Washington University.
Subjects – Topical (LCSH)
Ocean acidification--Oregon--Netarts Bay--Computer simulation; Neural networks (Computer science)--Oregon--Netarts Bay
Geographic Coverage
Netarts Bay (Or.)
Rights
This resource is displayed for educational purposes only and may be subject to U.S. and international copyright laws. For more information about rights or obtaining copies of this resource, please contact University Archives, Heritage Resources, Western Libraries, Western Washington University, Bellingham, WA 98225-9103, USA (360-650-7534; heritage.resources@wwu.edu) and refer to the collection name and identifier. Any materials cited must be attributed to the Salish Sea Ecosystem Conference Records, University Archives, Heritage Resources, Western Libraries, Western Washington University.
Type
Text
Language
English
Format
application/pdf
Included in
Fresh Water Studies Commons, Marine Biology Commons, Natural Resources and Conservation Commons, Terrestrial and Aquatic Ecology Commons
Omega Oracle: forecasting estuarine carbonate weather
There are serious concerns about ecological, social, and economic impacts in the Pacific Northwest due to Ocean Acidification (OA). We built a system to predict aragonite saturation state (Ω) of seawater in Netarts Bay, Oregon based on large scale forcing parameters. An artificial neural network – trained against a continuous, multiyear monitoring record of carbonate chemistry – learns a regression estimate of Ω based on seasonality, tides, and wind conditions. This approach is agnostic to the details of the underlying chemical and biological processes offering a distinct modelling perspective. The result is a conceptually simpler and more strictly empirical parameterization and a model that is flexible in application due to dependence on only easily obtainable parameters. Forecast validation by a cross validation method indicates good prediction performance, particularly for the high frequency content of the Ω time series, over periods of stable wind forecasting. Our forecast model demonstrates that the complex temporal dynamics of carbonate chemistry within an estuary can emerge from forcing operating on longer timescales. This further elucidates the management and commercial value of this model; experimental work with calcifiers suggests the details of these high frequency chemical dynamics are critical to the magnitude of stress imposed. Lastly, these forecasts, deployed as a web application, can facilitate OA mitigation strategies by providing aquaculturists with real-time predictions for consideration in operational decisions. Numerous sites, including on the Salish Sea, are poised to soon have viable training data for application of this method. Broader deployment promises to enable comparison between sites and expansion of direct aquaculture and management applications. Expansion to other sites is expected to require altered explanatory variables but this exercise may itself yield insight. Relatedly, we note the potential of this approach to help constrain timescales and sources (natural and anthropogenic) of contributions to physiological OA stress.
