Combined neural network and numerical modeling for extreme storm surges

Transcription

1 Leichtweiß-Institute for Hydraulic Engineering and Water Resources Department of Hydromechanics and Coastal Engineering Platzhalter für Bild, Bild auf Titelfolie hinter das Logo einsetzen Combined neural network and numerical modeling for extreme storm surges Mohamed Tayel and Hocine Oumeraci 10 March FZK-Kolloquium

2 Contents Motivation and Background. Objectives. Methodology. First results. Summary. modeling for extreme storm surges Page 2

3 Motivation and background Short- term Effects Astronimical Tide Seiches (resonance) deterministic + Corilios force Inverse barometric effect + Wind setup Static storm surge Corilios force Wave setup External surge Meteorological Stochastic) + Corilios force + Corilios force River discharge Tide s, Storm Surge, Wave setup, External factors ) = Extreme water level modified after Oumeraci, 2009 Sea level rise Effect of Morphology and structures Long- term Effects modeling for extreme storm surges Page 3 Nonlinear interaction and until now not completely solved

4 Objectives Developing an operational model which combines ANN and numerical modelling. Determination of the worst extreme water level, which is physically possible in the 21st century at a given site. Building ANNs model to relate and predict extreme water levels at a given site (e.g. Sylt) using data from a neighbouring site (e.g. Cuxhaven). Filling water levels data gaps for given sites. Grossmann I., K. Woth and H. Von Storch Die Küste 2005 modeling for extreme storm surges Page 4

5 Methodology meteorology forces from present day re-analysis and hindcast A Hydrodynamical model (Telemac 2D) covering the North Sea area Storm surge forecasted at Cuxhaven and Sylt meteorology forces from climate change part Climatic change part GCM using A2 and A1B scenarios from IPCC(2007) Different RCM for North Sea (see Woth k., R. Weisse and H. Von Storch GKSS 2005/1) Astronomical tidal constituents NARX neural network model account for nonlinear interaction of all water levels constituents Total water level at Cuxhaven and Sylt External surge at Aberdeen and Elbe river discharge ANNs models Water levels at Cuxhaven and Sylt using hydrodynamic model every 6 hours ANNs model predicting hourly values of Meteorological forces for Cuxhaven and Sylt ANNs model for hourly water level for Cuxhaven and Sylt Water levels at Cuxhaven and Sylt modeling for extreme storm surges Page 5

6 First key results I First phase Optimal structure and input parameters Input Layer External surge time series Pressure time series Wind velocity time series Hidden Layer Output Layer Second phase: Effect of previous water levels Wind direction time series Tide time series Wave setup time series River discharge time series b 1 b 2 b 3 Forecasted water level Previous water levels modeling for extreme storm surges Page 6

7 First key results II Previous time steps Correlation Coefficient Error Index (EI) Previous time steps Correlation Coefficient Error Index (EI) modeling for extreme storm surges Page 7

8 First key results III cc= 0.99 and EI= modeling for extreme storm surges Page 8

9 Summary Combining ANNs with numerical modelling allows to: account for nonlinear interaction of water levels constituents. reduce the training data for ANNs and provide water levels predictions every hour instead of every 6 hours by RCM meteorological forces. provide a powerful and computationally efficient modelling techniques. Applications as an operational model for: filling the gaps in long-term data series by using sequential time series predictions at given site. water levels retrieval at given remote site (e.g. Sylt) based on a long-term data recorded at neighboring sites (e.g. Cuxhaven). modeling for extreme storm surges Page 9

10 Vielen Dank für Ihre Aufmerksamkeit Thank you modeling for extreme storm surges Page 10