-
Notifications
You must be signed in to change notification settings - Fork 28
Thetis meeting agenda 2020 03 20
Date 2020-03-20 14:00UTC
Lawrence Mitchell, David Ham, Athanasios Angeloudis, Joe Wallwork, Lucas Mackie, Mariana Clare, Matt Piggott, Simon Warder, Tuomas Karna
Remote meetings via webex (since hangouts free tier has 10 person limit) (see slack channel).
Action: DH to add Mariana and Matt to webex team for hosting purposes.
TK presented a comparison of mimetic vs. DG for baroclinic vorticity. Mimetic is better from numerical results, but speed is about 4x slower (2x for solving, 2x for timestep restriction).
DH: what is the cost for same "effective resolution"? TK: unclear exactly what the right error metric is.
TK: Presented Columbia plume application. Got a bunch of interest from coastal ocean modellers, because the low numerical dissipation is of high importance.
MP: Why not FVCOM for these people? TK: Not a lot of experience, but view is that the dissipation is high.
TK: Plans to compare Thetis with MPAS-Ocean. DH: Aim for comparisons with "coastlines", so that one isn't giving them an easy pass due to doing things in boxes.
Conclusion is that one really has to do comparisons for time to error, rather than time to solution at given mesh resolution. Otherwise the more complex numerics Thetis runs don't look like they are advantageous.
TK: starting to work on 3D wetting and drying. Looking for realistic test cases to run for benchmarks. Both 2D and 3D.
MP: Quite a lot of data and models/bathymetry for severn estuary (2D)
AA/SW: Dam break idealised testcases
Action SW: send TK severn estuary calibration data/paper.
Not immediately clear that these map onto 3D.
Tuomas: update 3d bottom friction PR #192
Please comment on PR.
Mariana: conservative tracer 2d PR #191
Please comment on PR.
One test is failing, although MC does not think that this is unexpected. Fails only overshoot part: suggestion to remove overshoot check, since it is not expected for conservative schemes?
Periodic equatorial channel. From paper comparing FVCOM and ROMS. Thetis appears to preserve peaks better than other models.
Some new test fails: perhaps related to Firedrake changes?
Started looking at Monge-Ampere approach of Andrew. Now wanting to do parabolised version (a la Chris Budd) for time-dependent things. This requires ALE solvers in 2D.
Status of firedrake-adjoint issue #184
Action JW: switch back on adjoint tests and see if things still fail (some fixes went into firedrake master recently).