|Date||Topics & Readings||Note|
||12:00-13:30. We retake the
Hydrology problem and the proposed solution of dealing with model
inadequacy by adding time-dependent parameters which get then run
through the model (Peter Reichert); A similar approach has been taken
in a Biology model (Ariel Cintron-Arias), which will be presented at
some future meeting. This extra meeting will also be WebEx and
Teleconference accessible. You can remind
yourself of the model, issues, and preliminary results in Peter's
presentation in 11-06-2006.
The slides of the 2-1-2007 presentation can be downloaded here.
meet again at the regular time, 11:00-13:00; we'll look into Jim
Crooks' developments, and also continue discussion of issues
raised by Peter's hydrologic model. Jim Crooks' slide is
||We meet at the usual time and
So far, we have explored two possibilities for dealing with bias: a) an unstructured (input dependent), external bias term, added to the model output (approach taken by the MUCM, NISS Project team, Dave, Herbie ... etc), which might have troubles extrapolating, and b) a (time dependent) bias term added to other input parameters, (approach taken by Peter Reihert and others); this should extrapolate better (since it goes inside the model), but requires access to the code and is computationally expensive. A middle of the road approach 'linearizes' this model taking the input-dependent bias term out of the model, but adding much more structure to the bias term; this is specially useful when the simulator is very expensive to run. Rui Paulo will be presenting details in a simple example.
|2-19-2007||We continue discussion of the use of derivatives of the model and the GASP emulator in the spirit of the previous meeting. Recall that one of the uses could be interpreted as an approximate linearization of the model used by Peter Reichert for time dependent parameters; Peter applied his model to the "Toy Example" which Rui was using (the example is described in Section 1.3 of this paper, and analyzed with the 'usual' methodology in subsequent sections). This is a summary of Peter's results. Rui will present results of the approximation and we'll discuss the difficulties encountered. These are the slides of his presentation.|
||Cari Kaufman will report
progress on use of Covariance Tapering methods to deal with huge
matrices in GASP's, and
Jim Crooks will report on progress on both a particular mixture of GASP's and using copulae to interpolate between
probability distributions (not necessarily normal). You can download both Cari's slides and Jim's slides.