Condensation fireworks

[milankl]

The current condensation scheme produces negative values in the upper troposphere with funky "fireworks" that move against the wind... The white stuff is convection!

humidity.mp4

[milankl]

snapshot from last time step (colorbar is inverted though compared to the video) shows the negative humidity

[milankl]

This is a numerical issue because a $\tfrac{q^\star - q}{\Delta t}$ condensation tendency on a single time step is numerically very oscillatory with our leapfrog time stepping scheme, e.g. here starting at T=300K and 110% humidity

The implicit condesantion correction that accounts for the higher saturation humidity $q^\star$ due to the latent heat release helps already and should be used in any case. Reducing the tendency by a factor of $1/n$ so that $\tfrac{q^\star - q}{n\Delta t}$ with $n=7$ (just a multiple of a few time steps) and using a flux limiter of 0.2K per time step to limit the latent heat release however already helps massively

approaching 100% within a few tens of time steps which sounds reasonable. At T85 that's equivalent to a time scale of 1 hour.

[milankl]

With these changes the fireworks are gone, humidity can still be negative though, but that's tolerated for now

humidity3.mp4

[milankl]

Surface humidity in comparison (previous was at k=3 of 8 levels)

humidity3.mp4

[milankl]

Snapshots of 6-hourly precipitation stitched together with the new scheme, large scale + convective. I blame the current UniformCooling longwave radiation scheme for making the tropics too often slightly unstable causing large patches of convective precipitation there. So I believe we need to wait for a better radiation scheme to correct that.

precip2.mp4

[milankl]

Update: For large-scale condensation the tendencies should be evaluated at the previous time step, because that changes the above to

meaning no wiggles anymore in the implicit case, the explicit has some but muuuuch less than before. I guess the same applies for surface fluxes probably too that should be evaluated on the previous time step?