default number of water iterations too low

[ericbarefoot]

At the moment, the water iteration process (the number of times parcels are routed, then weights recalculated) only happens one time.

We should have it go at least 5 times, since this allows the water surface and flux fields to more realistically simulate water flow. For example, see the following three plots of eta after 11 timesteps each.

itermax = 1

itermax = 2

itermax = 5

[amoodie]

can you confirm the behavior in the matlab version? I think you told me it was 5? If so, yeah I think change it!

[elbeejay]

I understand this being the case when the flow field is more of a 'jet' than a delta. After a distributary network is developed is this difference still apparent? I think we should probably test a few cases/longer scenarios to get a better idea of what impact the water iteration number has.

[ericbarefoot]

I think we should probably test a few cases/longer scenarios to get a better idea of what impact the water iteration number has.

Yeah that's a good point @elbeejay. I suspect that the jet vs spread flow is probably more important initially, but I suspect (with no evidence atm) that the flow field evolves to some steady state solution with more iterations, so finding the optimum number would be important.

If so, yeah I think change it!

Lol. I believe in you. I can confirm that it was 5 in Matlab originally, where the hydro loop is done here. l certainly think this is perhaps the easiest issue to resolve, but we should be clear on the impacts, because it will definitely break this test

[amoodie]

summary

With the help of our handy new preprocessor matrix expansion (#69), this was a really easy test to do. The yaml config at the end of the comment was used to do 500 timesteps runs (summarized below, and all results attached as a zip).

I guess my overall interpretation is that @elbeejay is probably right. While itermax strongly influences the jet formation early on in the simulation, the effects later are much less obvious, if at all. Thoughts? Should we do a finer-itermax-interval set of experiments? Longer simulation?

results

eta

itermax	timestep 5	timestep 350	timestep 495
1
2
5

velocity

itermax	timestep 5	timestep 350	timestep 495
1
2
5

config

out_dir: itermax_comparison
Length: 80.0
Width: 200.0
seed: 0
dx: 1.0
L0_meters: 3.0
S0: 0.0002
u0: 0.5
N0_meters: 5.0
h0: 1.0
H_SL: 0.0
f_bedload: 0.4
C0_percent: 0.1
toggle_subsidence: False
sigma_max: 0.0
save_eta_figs: True
save_discharge_figs: True
save_velocity_figs: True
save_dt: 5
save_strata: True
timesteps: 500

matrix:
  itermax:
    - 1
    - 2
    - 5

[elbeejay]

Thanks for spinning those up and looking into this @amoodie. My thought is that for the default value, we might want itermax to be set low so that the default configuration runs "faster".

[amoodie]

Yeah, I agree. To me, it seems like itermax: 2 strikes a balance between "jet" and "radial symmetry" early on in the run, and since this seems to work to produce a good looking delta after 500 timesteps, I think it's probably a good default choice.