Oscillatory behaviour of fluxes and surface temperature

[nichollsh]

The model sometimes exhibits oscillatory behaviour in F_atm, F_int, and T_surf. This is probably not physical, since it occurs when using the dummy modules. It also occurs when using SPIDER+JANUS.

We should investigate this further. My guess is that it is associated with the time-stepping scheme, which is currently quite crude.

Example from the dummy model in the image below. The behaviour looks particularly bad in this plot because of the log scale on the y-axis. We can see from the OLR lines that the magnitude of the oscillations is quite small (in this case, at least).

[timlichtenberg]

Is this still relevant with the current last version?

[nichollsh]

Yeah this is still an issue. It doesn't typically cause problems physically, but we should at least look into using a better time-stepping scheme in terms of model efficiency when the melt fraction is small. Maybe there are techniques that can be borrowed from ESM codes? Somewhat connected to https://github.com/ExPlanetology/aragog/issues/21.

I would suggest that we look into this soon, because it could potentially yield significant performance benefits.

[nichollsh]

Looking into this a bit more... I think this behaviour arises from the time-step being increased up to the point where the model starts to become unstable. For small time-steps, this simulation below is smooth. If we increase the tolerance on the time-steps (which makes them larger) then occasionally the fluxes will oscillate. The oscillation settles down because the time-step is then decreased.

I am sure there are some tried-and-tested ways to handle this. I guess one approach would be something like what Dan suggested here, where the model looks ahead to see what would happen to the evolution of the planet under a proposed step length.

[nichollsh]

Forcing the model to take large steps shows the following behaviour. This is what we'd expect given the logic above. This further indicates that the solution to stopping these oscillations can be found in a better time-step selection scheme.