data assimilation with openda

Data assimilation with OpenDAOpenDa workshop

November 2012

eScience center

8-11-12

Nils van Velzen, Martin Verlaan, Stef Hummel, Julius Sumihar, Alja Vrieling, Albrecht Weerts, Ghada El Serafy, Herman Gerritsen

Outline

Data assimilation and calibration in one sheet

What is OpenDA

OpenDA architecture

Coupling models into OpenDA

Example of Calibration

Example of Data assimilation

OpenDA and Parallel computing

Summary

2Data assimilation with OpenDA

Data assimilation and Model calibration in one sheet

• A generic toolbox for data-assimilation

– (real time) data assimilation

– model calibration

– impact of observations

– reconstruction of sources


What is OpenDA

• Motivation for a toolbox:

– The algorithms are model/observation independent BUT(!)

– DA methods work on top of models → serious changes to model code.

– Spaghetti code

– Difficult to change algorithm

– Maintenance problems

– Difficult to test algorithms

– Difficult/impossible to reusecode

– Expensive!


What is OpenDA

• Content:

– Set of interfaces that define interactions between components

– Library of data-assimilation algorithms

– DA philosophy

– Building blocks only need to be implemented once


What is OpenDA

• Open source (LGPL)

• Why OpenDA?

– More efficient than development for each application

– Shared knowledge between applications

– Development of algorithms with e.g. universities

– Easier to test, which should result in fewer bugs

– Optimized building blocks

– Development template


OpenDA Website


• www.openda.org• Downloads

– Users

– Developers

• Documentation

• Wrappers for models

• OpenDA association

OpenDA Main Application


Single program for

• Editing input files

• Running simulations and calibrations

• All models and observation sources

• GUI and command line

• Sequential and parallel

OpenDA architectureObject Oriented Approach


Models in OpenDA

Formal form of a model

State of model instance

Object oriented concepts:

Data encapsulation; state cannot be directly changed only through the methods like:

GetState, AxpyState, Compute...

Multiple instances of model

Algorithm has no knowledge on model internals


dxdt

=M (x (t ) ,u (t ) , p ,w(t ))

x (t) , u (t ) , p ,w(t)

Models in OpenDA

• Models in Data assimilation context are stochastic!

• Normal dynamic model (codes) are not


Models in OpenDA

• A model wrapper is needed for each model:

– Black box

– Java

– C/Fortran

– C#

• Testing OpenDA models

– Java

– Matlab

– C#


Models in OpenDA


OpenDA

MyStochModel MyWrapper

TheirModel.exe

MyStochModel

Black Box wrapperOpe

nDA

OpenDABridge

java

C/Fortran

other

own

3rd p

arty

MyModel MyModel

Models in OpenDA


OpenDA

Black Box wrapperOpe

nDA

java

C#

own

3rd p

arty

MyModel

OpenDA

Black Box wrapper OpenMI Wrapper

OpenMI Model

Generate

Example of model calibration SWAN 3rd generation wave model


Wave breaking and interaction over a bar

0.4

0.3

0.2

0.1

0

0.5m

X

Z0 10 20 30m

WAVES

STATIONS01 02 0403 0705 06 08

totS =wind input+ non-linear interactions (quadruplets & triads)+ whitecapping + bottom friction+ depth induced wave breaking

Example of model calibration

SWAN 3rd generation wave model


Example of data assimilationOperational storm surge forecasting


EnKF semi parallel

07-11-12 20Data assimilation with OpenDA

ξif (t k )=M (ξi

a(t k−1))+wi(t k )

x f (t k )=1N

ξif (t k)

E f (t k )=[ξ1f (t k )−x

f (t k) ,ξ2f (t k )−x

f (t k) , ... ,ξNf (t k)−x

f (tk )]⋮

Semi parallelCommunication

volume

Cm+n N

0

ξia(t k)=ξi

f (t k)+K (t k)[ y (tk )−H (t k)ξif (t k)+v i(tk )]

n N

Only parallelize model steps:• Not scalable, often sufficient

EnKF semi parallel


Lotos-euros air quality model

EnKF semi parallel


Generic semi parallel due to OO concepts

EnKF parallel


ξif (t k )=M (ξi

a(t k−1))+wi(t k )

x f (t k )=1N

ξif (t k)

ξia(t k)=ξi

f (t k)+K (t k)[ y (tk )−H (t k)ξif (t k)+v i(tk )]

Stochastic model time steps

Mean of Ensemble

Update (for all ensemble members)

EnKF parallel


Column wise distribution

Separate Combined

n Np

n log2( p) n log2( p)

n N log2( p−1)

Row wise distribution

Separate Combined

C m

n Np

+n N log 2( p−1)

0 C m+n Np

00

0n Np

EnKF parallel


WAQUA shallow water model

Comparison of parallelization strategies for RRSQRT (Roest et al.)

EnKF parallel

Full parallel using OO concepts:

All filters run same code

State vectors are distributed (parallel vector)

Operations on parallel vectorsby a parallel vector implementation

NO need to change the model andfilter code

All complexities hidden in genericsupport layers/implementations


Parallel computing with black box models

Black box models

No change to model code

Data exchange using files

Note: disk can be slow/more data written than needed



Swan model for wind generated waves

Operational model + DA for the north sea

Black box model

1 hour, 8 cpu's 10 min 50% IO

EnKF implementation (?)

Parallelization of filter



Black box model is normal model

Semi parallel+full parallel

Note disk speed,

Use local disks:

Faster

No sequentialbottleneck


Summary

• OpenDA: Data assimilation toolbox for Data assimilation/calibration.

• Object oriented design allows easy exchange of building blocks.

• End users do not need to do any programming for experimenting.

• Various ways for adjusting (your) model for use in OpenDA.

• Conceptual support for all kinds of parallel computing.


data assimilation with openda

Documents