Default time step is too long for high RHi cases

[sdeastham]

The default transport and ice growth time steps are currently 10 minutes, which is too long for high RHi cases. See below - when ice crystals are able to grow large, they settle through many grid cells in a single step. As a result, water in the intervening cells is not taken up, resulting in "pulse-like" behaviour:

Reducing the time step to one minute improves (but does not fully fix) this issue:

Not only is this an imperfect fix, it increases run times significantly (~4x in the given test case, going from 30 seconds to 2 minutes). It would be better to find other fixes (e.g. include simplified ice growth in transport sub-cycling).

[sdeastham]

It looks like 5 minutes may be a reasonable compromise. Going even lower dramatically increases run times but does not significantly change the total ice mass as a function of time, at least for this case. More testing is needed.

[sdeastham]

This issue appears to be (somewhat) less severe at lower RHi; in this case, at 110% RHi and 10 K lower (the loRH cases):

[Calebsakhtar]

Just to follow up on this. I have conducted my own tests on a 125% RHi ISSL case (as opposed to the 140% RHi ISSL performed by @sdeastham above). Here are the results:

Ice mass (kg / m):

Integrated Optical Depth (m):

Number of ice crystals (#/m):

Variation of the time-integrated intOD with timestep:

Observations
With increasing the timestep from 0.5 mins, the behaviour "converges". From about 5 mins upwards, the behaviour "diverges".

The convergence is likely numerical diffusion, as @sdeastham mentioned. I might choose to manually verify this by checking APCEMM's transport equation at a later date.

This competes with the pulse-like behaviour induced divergence, which becomes increasingly problematic at higher timesteps.

Conclusion
4-5 mins seems to be when these effects cancel each other out. This concurs with @sdeastham's analysis at 140% RHi.

I recommend the use of a 5 minute timestep for both ice growth and transport

[chinahg]

Wanted to add my observations here as I started seeing some interesting (concerning?) behavior regarding timestep choice on resulting contrail properties.

I ran test cases for T = 205K, 218K, 225K and RHi = 110%, 130%, with a lapse rate of 2.5 K/m. The transport timestep was 1, 5, or 10 minutes and the ice growth timestep was held constant at 1 minute.

Note 1: The ice growth timestep is not consistently implemented and can lead to confusion as to what growth timestep scheme is actually being implemented. In short, the ice growth is being triggered for every transport timestep (unless ice growth step > transport step). See

APCEMM/Code.v05-00/include/Core/TimestepVarsWrapper.hpp

Line 111 in dc763ec

inline bool checkTimeForIceGrowth()

and

APCEMM/Code.v05-00/src/Core/LAGRIDPlumeModel.cpp

Line 69 in dc763ec

SimStatus LAGRIDPlumeModel::runFullModel() {

As for the actual results there are large differences in the resulting contrail properties depending on the chosen timesteps. Regardless of integrated optical depth convergence I think the model being this sensitive to timestep is concerning.

Legend format is [transport timestep in minutes]T[ice growth timestep in minutes]I. Note that the true ice growth timestep is equal to the transport timestep (as explained above).
Quantities on the y-axis are per meter flown.
Ignore the truncated lifetimes of the 205K and 218K 130% cases, the runs are in progress and are taking an exceptional amount of time (see #81)

Some part of these discrepancies could be linked to increasing numerical diffusion at smaller timesteps or a Courant number > 1 for larger timesteps. Curious to see if anyone has other thoughts on this.