1 / 32

COSTA An Introduction Nils van Velzen email: c.vanvelzen@tudelft.nl

www.costapse.org. COSTA An Introduction Nils van Velzen email: c.vanvelzen@tudelft.nl. www.costapse.org. Outline. Data assimilation and calibration Existing software and programming issues COSTA Application of COSTA with WAQUA/TRIWAQ Conclusions. Data assimilation and calibration.

elam
Download Presentation

COSTA An Introduction Nils van Velzen email: c.vanvelzen@tudelft.nl

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. www.costapse.org COSTAAn Introduction Nils van Velzen email: c.vanvelzen@tudelft.nl Data Assimilation Summer School, Sibiu, 6th August 2009

  2. Data Assimilation Summer School, Sibiu, 6th August 2009 www.costapse.org Outline • Data assimilation and calibration • Existing software and programming issues • COSTA • Application of COSTA with WAQUA/TRIWAQ • Conclusions

  3. Data Assimilation Summer School, Sibiu, 6th August 2009 Data assimilation and calibration • (Dynamic) simulation models • Weather forecast • Air quality • Shallow water (storm surge) • Ocean modelling • Reservoir • Predictions are not perfect!

  4. Data Assimilation Summer School, Sibiu, 6th August 2009 Data assimilation and calibration • Measurements: • Satellites • Buoy • Radar • Weather stations

  5. Data Assimilation Summer School, Sibiu, 6th August 2009 Data assimilation and calibration • Make model perform better using the available observations: • Calibration: change model parameters in order to reduce the difference between the model predictions and observations • Data assimilation: combine the model forecast and observations into a mixed forecast

  6. Data Assimilation Summer School, Sibiu, 6th August 2009 Existing software and programming issues • Simulation software is complex • More complex is combined with Data assimilation and/or Calibration methods • Expensive to develop and maintain • Splitting up software is in general a good idea • Less complex • Possible to reuse (parts of) the code

  7. Data Assimilation Summer School, Sibiu, 6th August 2009 Existing software and programming issues • Why people think they have to develop a custom implementation of a DA method: • Computational efficient • Need to handle model specific issues • Are these assumptions correct? Issues are: • Which method to implement? • Expensive • Incompatible • Potentially full of bugs

  8. Data Assimilation Summer School, Sibiu, 6th August 2009 COSTA • A problem solving environment for data assimilation and calibration • Components and their interface • Data assimilation methods • Platform for exchanging models and methods • Free software (LGPL)‏

  9. Data Assimilation Summer School, Sibiu, 6th August 2009 Steps to prepare model codes for implementing data assimilation • Identify the state-vector of your model and those parameters and forcing (you think) you want to use • Isolate the model time step --This is 90% of the work and has to be done for a custom data assimilation implementation as well --COSTA provides tools to simplify this work

  10. Data Assimilation Summer School, Sibiu, 6th August 2009 COSTA • A generic environment where user can easily try out different DA-methods to their models • A generic environment where method developers can easily implement their method to different models

  11. Data Assimilation Summer School, Sibiu, 6th August 2009 COSTA • Most data assimilation actually only requires information about model state and observation at the same time. • This makes it possible to implement model, observation and data assimilation methods separately as independent components.

  12. Data Assimilation Summer School, Sibiu, 6th August 2009 COSTA components Model, observations and DA-method

  13. Data Assimilation Summer School, Sibiu, 6th August 2009 Model Component • Model component in COSTA (formal) • State of a model instance • Methods to get or change the state of model instance

  14. Model Component • Propagate the model state-vector • Get, set, axpy for • GetObsValues: Data Assimilation Summer School, Sibiu, 6th August 2009

  15. Data Assimilation Summer School, Sibiu, 6th August 2009 Using COSTA

  16. Data Assimilation Summer School, Sibiu, 6th August 2009 Using COSTA • Use the “costawb” program to combine a model with a DA-method without any programming • Model components are dynamic libraries and are linked to the “costawb” program

  17. Data Assimilation Summer School, Sibiu, 6th August 2009 WAQUA/TRIWAQ • Simulation model for shallow water • Used operational by the Dutch Rijkswaterstaat • Long history of data assimilation • Too expensive • Inflexible

  18. Data Assimilation Summer School, Sibiu, 6th August 2009 WAQUA/TRIWAQ • Domain decomposition and parallel computing • COSTAmodel component • RRSQRTFilter

  19. Data Assimilation Summer School, Sibiu, 6th August 2009 The WAQUA/TRIWAQ model • Deterministic model: • Stochastic model:or

  20. Data Assimilation Summer School, Sibiu, 6th August 2009

  21. Data Assimilation Summer School, Sibiu, 6th August 2009

  22. Data Assimilation Summer School, Sibiu, 6th August 2009 Deterministic and Stochastic • We want a separation between deterministic and stochastic model • Use deterministic model for other applications • Play with alternative noise models

  23. Data Assimilation Summer School, Sibiu, 6th August 2009 Deterministic and Stochastic • WAQUA/TRIWAQ uses ADI-scheme: • One time step consist of two half time steps • Time correlation applied each half time step • New approach: • Time correlation each whole time step • Different results!

  24. Data Assimilation Summer School, Sibiu, 6th August 2009 Challenge • Can we: • Separate filter and method Method? • have same results? • Similar performance? • Is the COSTA modelinterface sufficientfor this task?

  25. Data Assimilation Summer School, Sibiu, 6th August 2009

  26. Data Assimilation Summer School, Sibiu, 6th August 2009 Conclusions • COSTA offers a flexible framework for using and developing data assimilation methods. • Successful coupling with “real” models: • WAQUA/TRIWAQ • Lotos-Euros • Chimere

  27. Data Assimilation Summer School, Sibiu, 6th August 2009 Conclusions • Easy to use • Complete basic functionality available: • Implementation available for all basic building blocks • Growing number of available methods • Support for parallel computing

  28. Data Assimilation Summer School, Sibiu, 6th August 2009 Thank you!

  29. Data Assimilation Summer School, Sibiu, 6th August 2009 Existing software and programming issues

  30. Data Assimilation Summer School, Sibiu, 6th August 2009 Tools for creating COSTA models • Simple model builder • Handles the “object oriented” aspects of the model • You only need to implement 4 functions • Model combiner • Combine multiple COSTA models in one composite model • Create a stochastic model from a deterministic model • Black Box model builder • Create a COSTA component for a model without changing the model code • Support for parallel models

  31. Data Assimilation Summer School, Sibiu, 6th August 2009 Automatic parallelization

  32. Data Assimilation Summer School, Sibiu, 6th August 2009 Using parallel models

More Related