Interacting through BMI can lead to odd soil_temperature behavior.

[BSchilperoort]

I used the example dataset from zenodo, and STEMMUS_SCOPE 1.6.1 through the BMI.

I set the temperature at one time step with:

        model.set_value_at_indices(
            "soil_temperature",
            np.array([10, 11, 12]),
            np.array([50, 50, 50]),
        )

But depending on at which model time step I do this I get weird results:

There are a few time steps where the set temperature impulse disappears completely, or the temperature jumps around.

[MostafaGomaa93]

I think we need to confirm if this bug is really from BMI or from the code itself. It could be from the code itself because in some of my trials (as below), I also receive extremely high-temperature values (up to 300 C) which of course there is a bug, and sometimes when I change the soil parameters, this bug is gone.

The example below runs without BMI and without groundwater coupling also

[BSchilperoort]

I think we need to confirm if this bug is really from BMI or from the code itself

it's for sure due to the model code. the BMI is just a way to get the model to misbehave clearly.

[yijianzeng]

I think we need to confirm if this bug is really from BMI or from the code itself

it's for sure due to the model code. the BMI is just a way to get the model to misbehave clearly.

Hi Bart, not sure how the test is done. Is it using an existing site? If you provide a site name, we can have a further look. The problem is the numerical convergence from two potential parts:

1. 'ebal' in SCOPE cannot reach the energy balance criteria, and the coupled model does not converge. We'd observed this in quite some situations.

2. in STEMMUS, originally, we allowed the model to have a flexible time step ranging from 1s to 30 mins. Such treatment is to address the fact that some dramatic changes in state variables (either matric potential, soil temperature, or atmospheric pressure) will lead to numerical divergence. With a flexible time step, then the changes in state variables will be reduced, and the model can come to numerical stability again.

But now, since SCOPE has a fixed 30-minute interval, we'd fixed the time step of STEMMUS-SCOPE as 30 minutes as well, which may eventually lead to numerical divergence.