debug NaNs in 4 year longrun

[juliasloan25]

In #908, we ran the global snowy land model for 4 years. This resulted in some NaNs in a few regions, namely western Canada, Patagonia, southern Greenland, and coastal Antarctica (see plots below). We need to investigate why these come up and fix the model before we can couple the full land model with atmos.

To Do

• run regional simulations of these areas with frequent output to determine which variable first becomes NaN

Plots from 4 year longrun

These plots render too small in Github to see the NaNs, but zooming in will show where the NaNs appear.

[juliasloan25]

After the Jacobian approx. improvement in #953, we see improved stability across the board in our longrun simulations. In the snowy land longrun, we now see no NaNs visually in plots after 4 years, though we should look at the simulation outputs interactively to quantitatively verify this. We do see some NaNs in Greenland in the soil-only longrun after 4 years

We also tried increasing our timestep from 450s to 900s, but this introduced more NaNs in coastal areas and the midwestern US (great lakes?). See the below plots for some results with different timesteps after the jacobian update:

LEFT: snowy land, dt=450s (build) ||| RIGHT: snowy land, dt=900s (build)

Next steps

• Interactively view longrun output and check if any NaNs at 4 years
• Try running longruns with dt=600s, check if any NaNs at 4 years
• Investigate cause of coastal NaNs with dt=900s - parameter maps?
• Add automated check to determine number of NaNs at end of longrun simulations

[juliasloan25]

I ran the longruns using dt=600s for 4 years. The simulations look largely stable, but in the snowy land longrun we see small regions of NaNs in the horn of Africa and the northwestern coast of Africa. In the soil-only longrun, we see an increased number of NaNs in Greenland as compared to using dt=450s.

LEFT: snowy land, dt=450s (build) ||| RIGHT: snowy land, dt=600 (build)

Note the slight changes in the horn of Africa and the northwestern coast of Africa.

LEFT: soil, dt=450s (build) ||| RIGHT: soil, dt=600 (build)

Note the larger region of NaNs in Greenland when using dt=600s.