Brian Powell - Research Notes

4DVAR!

Mon, 22 May 2006 19:21:31 CDT

After countless attempts, I finally have a weak constraint 4-D variational data assimilation case working for the Intra-America Seas. With a terrible first-guess field (from monthly climatology), I assimilated a SSH field and went back 5 days.

This plot shows the improvement in the initial misfit. Now that the mechanism is working, it is time for some real experiments. Finally!

New Cluster

Sat, 20 May 2006 13:19:07 CDT

We are ordering our new cluster this week. 100 CPUs of Opteron computational power. Oh my, oh my, I have so many experiments lined up to run on this beast. I am giddy to get it going and take the load from the current 32 CPU cluster.

ROMS is such an excellent parallel platform that we will be able to keep this cluster fully loaded, and the amount of time waiting for experiments to complete will be drastically reduced.

Mothballing Matlab

Sat, 20 May 2006 13:16:35 CDT

Can someone say that? Get rid of matlab? Well, it is becoming more possible by the day. It isn't without work. No, this is not some clone/copy of matlab, but an entirely new environment built using python. A real language. Open Source. No annual fees.

Rob Hetland at Texas A&M has been carrying this torch and has slowly been grinding me down on these very points, and I have to say that I may have finally caved. Numerical Python (numpy) provides the foundation for python to become a numerical powerhouse. BLAS/Atlas? Used natively by numpy. FFTW? Used natively by numpy. MatPlotLib provides the plotting engine, and we are mostly there.
[...]