DSD-INT 2016 Keynote Lecture 2016: From global to local: the latest developments - Verlaan

November 1st 2016

From global to local: the latest

developments

[email protected]

Outline

Motivation – global to local A Global Tide and Surge Model (GTSM): Delft3D FM

Initial model development Calibration with OpenDA Applications Towards the next version

Regional storm surge models Regional SAL effects

Innundation modelling (local) Innundation risk in Mozambique Xynthia flood

Impacts of storm December 6th 2013

Global effects of South pole ice

Present day ice line Weddell sea covered by ice

A Global Tide and Surge Model

Grid in Delft3D FM (D-Flow FM module)

This D-Flow FM grid uses triangles and rectangles for local grid refinement. Resolution is based on Courant number.

Unstructured approach - with grid refinement in shallow areas

July 5 2016

Computational performance

Calibration GTSM version 1

Model calibration

ModelpParameters ym Model output

yo Measurements

- CompareJ(p)Cost function

• Open source (LGPL) data-assimilation toolbox

• Library of data-assimilation algorithms:

– Simplex, conjugate gradient, LBFGS, DUD, ...

– EnKF, ENSR, steady-state, particle filter, ...

• Set of interfaces: models, observations, algorithms,...

• Platform for cooperation

• Association

11www.openda.org

What is OpenDA?

Model connectors

Hydrodynamics

• D-Flow FM (Delft3D FM Suite)

• Delft3D-FLOW

• EFDC

• Nemo

Waves

• SWAN / Delft3D-WAVE

Water quality

• DELWAQ* / Delft3D-WAQ

• HSPF

Rainfall runoff

• HBV

• OpenStreams*

• Mike-SHE

• MCRM

• Sacramento

Sewer systems

• SOBEK-FLOW (Delft3D FM Suite)

• WANDA

Groundwater

• Modflow*

CFD

• OpenFOAM*

Oil-gas reservoirs

• Eclipse*

Air quality

• Lotos-Euros

• Chimere

Calibration of the Model

FES2012

Altimeter observations:- assimilated FES2012 gridded data- very accurate on deep water

Calibration parameters: - Depth & Friction- 7 regions

Automated calibration:- 21 parameters- deep water series Jan2007

347 locations

Calibration results for deep water

Region Before After

Arctic 5.1 cm 3.2 cm

N. Atlantic 9.4 7.4

S. Atlantic 12.1 8.4

N. Pacific 8.1 6.2

S. Pacific 11.2 7.3

Indian Ocean 11.7 8.2

S. Ocean 12.4 10.2

Total 10.2 7.4

Reanalysis

July 5 2016

Highwater exceedance estimates

Reanalysis with ERA-interim meteo-forcing (1979-present)

Cooperation:Sanne Muis

July 5 2016

Validation of return periods

July 5 2016

GTSM-ERA-interim

GTSM-spiderweb

Without coastal refinement

Dietrich et al 2009

GLOSSIS operational forecasts

Hurricane Matthew

GLOSSIS operational forecasts: Hurricane Matthew

GLOSSIS operational forecasts: Hurricane Matthew

Storm Matthew Oct-07-2016 // FL, GA, SC

GLObal Storm Surge Information Systemsnapshot storm surge forecast

Max potential inundation level

Towards GTSM-v2

Grid refinement

Bathymetry gradient based refinement:

Sensitivity to resolution

2x finer in deep water

2x finer in shallow water

Tidal dissipation through generation of internal tides (IT)

IT dissipation – Anisotropic phenomenon

Barotropic tidal dissipation sources:• Bottom friction (dominant in shallow waters)• Generation of internal waves when flowing over rough topography in

stratified oceans (dominant in deep waters)

New anisotropic numerical implementation

Dissipation=f(Stratification, bathymetry gradient, cross-slope flow)

• Mid-ocean ridges and trenches (e.g. Mid-Atlantic Ridge)

• Continental shelves (e.g. Bay of Biscay)• Island chains (e.g. Hawaiian Ridge)

Literature values: = 1TW (4TW total)

( )1ITτ = CρNκ h u h- Ñ Ñg

Model sensitivity to grid resolution

Internal Tides dissipation (W/m2) distribution:

Model is now sensitive to user defined parameters for IT dissipation: Tuning possible.

RMSE reduction with IT dissipation:

• Deep waters:~3 cm• Coast: ~3.2 cm

Self Attraction and Loading (SAL)

SAL – Implementation in GTSM

SAL is the sum of three effects:

Deformation of the seafloor under the weight of the column of water (Earth is a elastic body)

Redistribution of Earth mass changes the gravitational field

Gravitational attraction induced by the mass of the ocean on the ocean itself

LOADING

SELF ATTRACTION

Literature ~ 10% of astronomical tide Reality: Spatial and temporal significant variability.

SAL potential in spherical harmonics Calculation of field at every time step, analogous to Tide Generating Forces

Reduction of errors in tidal representation of ~50 % in coastal stations and ~60% in deep waters

Minimal increase in computational times.

SAL tide spatial distribution

β approximation too simplistic

SAL tide Tide

Regional impact of SAL

Real-time data-assimilation

DCSM-v6 assimilation locations

Impact Kalman filter +6h

Computation for Vlissingen at Dec 5 19u

Results

Tide gages available through IOC/GLOSS

Satellite Data

JASON-3 Sealevel example

Todo

ASCAT Scatterometer winds

ASCAT Scatterometer winds

ASCAT Scatterometer winds

Data assimilation - future

Localization

“OpenDA-NEMO framework for Ocean Analysis and Predictions”, N. Van Velzen, M. U. Altaf, M. Verlaan, and A.W. Heemink, submitted to Ocean Dynamics

Sealevel

Observed Sealevel

Ensemble Kalman filter

Parallel scaling

Good scaling with standard OpenDA code for up-to 16 nodes (each 8 cores)

SWAN wave model

Floodrisk Mozambique

Synthetic cyclone tracks

Regional model

Example TC track

Preliminary results

Example TC track and surge

Preliminary results

Synthetic cyclones

Maximum surge levelfor 6 sample tracks out of more than 1000 cyclones that enter the area

Flooding example Mozambique

Xynthia storm

Xynthia storm surge

February 27 2010

Flooding computations

Grid 90m approx. 700x1000

Computation time (no MPI):• Delft3D FM (10minutes)• Lisflood-FP (18minutes)

Results sensitive to:• DEM• Boundary conditions• Friction

Future plans

• Complete GTSM version 2 including calibration with OpenDA• Incorporate real-time satellite observations in Delft-FEWS • Develop real-time data-assimilation of satellite data• Incorporate more accurate tropical cyclone estimates in global risk

estimates• Incorporate wave effects for coastal flood risk

• Develop coastal inundation models• Develop strategy for systematic validation of inundation models

Questions?

Thanks to …

Arthur van DamDirk EilanderEdward MelgerFirmijn ZijlHerman KernkampHessel WinsemiusJoao Lima RegoKun JanMaialen Irazoqui ApececheaSander van der PijlSanne MuisTony Minns

Albrecht WeertsAlja VrielingAnnette ZijderveldArno KockxArjo SegersDaniel TwigtJulius SumiharLora BuckmanMarieke EleveldNils van VelzenSandra Gaytan AguilarSimone de KleermaekerStef Hummel

…and others…