IWLS call 17 July (11 AM 18 July NZ)

some thoughts on the collaborative paper on ML approaches to seasonal MSLA forecasting

Transcript

1. IWLS  call  17  July     Nicolas  Fauchereau   Sco6  Stephens  
2. Agenda     •  Opera>onal  forecas>ng  and  ensembles   (Rashed)   •  Paper  collabora>on  (Nico  /  Sco6  [NIWA])    
3. Paper  collabora>on   •  Suggested  )tle:  Machine  Learning  approaches   to  the  predic1on  of  seasonal  Mean  Sea  Level   Anomalies  in  the  Pacific     •  To  be  submi0ed  to:  ?   •  Authors:  Nicolas  Fauchereau,  Sco>  Stephens,   Judith  Wells,  Rashed  Chowdhury,  John  Marra,   William  Sweet,  Doug  Ramsay,  ???  
4. Paper  structure   •  Introduc)on   –  Extreme  sea  level   –  Societal  benefits  of  opera>onal  extreme  sea  level  risk  calendar   –  Review  of  exis>ng  ini>a>ves   –  interest  of  (sta>s>cal)  ML  predic>on:  Open-­‐source,  lightweight     –  …     –  Possible  approaches:     •  EVT   •  Signal  Decomposi)on  (trend  +  )de  +  MSLA  +  high-­‐frequency),  and  forecast  individual  components     •  Data  and  methods   –  Data  sources     –  Data  processing  (predictors  /  predictands)     –  ML  algorithms     –  Model  evalua>on  (cross-­‐valida>on  and  metrics)   •  Results     –  Regression   •  OLS   •  MARS     •  NN     –  Classifica>ons     •  LDA   •  SVM     •  RF   •  Conclusions    
5. •  Introduc>on:     –  Everyone,  John  leading     •  Data  and  Methods:     –  Data  (predictand):  sources,  QC,  decomposi>on:  Judith,  Sco>,  Rashed   –  Data  (predictand):  discre>za>on  for  classifica>on:  Nico   –  Data  (SST):  sources,  decomposi>on  (EOF,  ICA):  Nico       –  Methods:  ML  algorithms,  cross-­‐valida>on,  metrics:  Nico,  Sco>  (NN),   Judith,  Rashed  (LDA)   •  Results   –  Regression:     •  OLS:  Nico,  Judith     •  MARS:  Nico     •  NN:  Sco6     –  Classifica>on   •  LDA:  Rashed,  Nico   •  SVM:  Nico     •  RF:  Nico     Who  does  what  ?  
6. •  Google  docs  ?   •  GIT  ?   •  Other  ?   How  to  collaborate  ?