Data driven modelling/prediction of Coastal State Indicators

1
Data driven modelling/prediction of Coastal State
Indicators Kenneth Kingston
2
Data Driven Modelling Philosophy

• Attempts to model the essence of a system or
  system parameter by capturing the inherent trends
  in a data time series.
• Little or no system knowledge hence cannot
  utilize physical principles to establish models
  of the system
• Models are typically statistical in nature
• some system knowledge helps a lot

Kenneth Kingston
3
Data Driven Modelling Objectives of this talk

• Introduce concepts of a statistical approach to
  modeling a system neural networks
• Illustrate the utility of the approach for an
  Argus derived CSI

Kenneth Kingston
4
Data Driven Modelling Neural Networks

• Input output approach to modeling a system
• if a relationship exists between the inputs and
  outputs of the system being modeled, the feed
  forward neural network will be able to establish
  it!

Kenneth Kingston
5
Data Driven Modelling The system

• Time series of a given Argus Derived CSI
• Adopt data driven philosophy of making no apriori
  assumptions based on system knowledge
• 2 years of data at monthly intervals
  (June 1999 June 2001)
• 15 alongshore estimates of the CSI

Kenneth Kingston
6
Data Driven Modelling The system
CSI
Kenneth Kingston
7
Data Driven Modelling The system
Kenneth Kingston
8
Data Driven Modelling NN Implementation
Assume tn f ( tn-1, tn-2, , tn k)
Input
Output
Time
Kenneth Kingston
9
Data Driven Modelling NN Implementation
Window size 6 months
Kenneth Kingston
10
Data Driven Modelling NN Implementation
Window size 6 months
Kenneth Kingston
11
Data Driven Modelling NN Implementation
Window size 9 months
Kenneth Kingston
12
Data Driven Modelling NN Implementation
Window size 9 months
Kenneth Kingston
13
Data Driven Modelling 2001 - 2003
Kenneth Kingston
14
Data Driven Modelling CSI ?

• Chosen CSI Argus derived measure of beach
  volume at Jan Van Speijk
• Howard over to you!

Kenneth Kingston