Add per-turbine timeseries air density support

[bjarketol]

Summary

• Pass density from wind_resource as the Air_density simulation kwarg to PyWake's wind farm model, enabling DensityScale to correctly scale power output
• Drop density from the XRSite dataset where PyWake was silently ignoring it
• Extract per-turbine or uniform density in the timeseries branch, mirroring the existing operating flag pattern
• Enhance the density test to verify that density actually affects AEP results

Changes

• wifa/pywake_api.py: Drop density from site dataset, initialize air_density, extract density in timeseries branch, add Air_density to sim_kwargs
• tests/conftest.py: New shared test fixture make_timeseries_per_turbine_system_dict with per-turbine density data
• tests/test_pywake.py: Enhanced test comparing AEP with and without density
• pyproject.toml: Add tests to pythonpath so conftest is importable

Test plan

• test_pywake_dict_timeseries_per_turbine_with_density passes and verifies density affects AEP
• All 11 pywake tests pass with no regressions
• Full test suite passes (except pre-existing test_cs.py failures unrelated to this change)

🤖 Generated with Claude Code

[bjarketol]

@kilojoules can you review this one?

[bjarketol]

@SchmJo

[SchmJo]

@bjarketol I rebased this on #48 such that now also foxes passes a similar turbine-based timeseries test with air density

[SchmJo]

I am not a fan of numpy < 2, since it blocks future updates of numpy. Can we convince the wayve team to replace np.trapz by np.trapezoid instead? @koendev

[koendev]

As far as I can tell, it's essentially the same function, just renamed? If that's the case I can refactor the code to handle that, yes. Would any other changes be needed?

[bjarketol]

@koendev, you can use something like the code below to support both numpy 1 and 2:

        _trapezoid = np.trapezoid if hasattr(np, "trapezoid") else np.trapz
        U3 = _trapezoid(np.interp(z, zs, us), z) / (15.0e3 - h)
        V3 = _trapezoid(np.interp(z, zs, vs), z) / (15.0e3 - h)

[kilojoules]

Adresses #31

[kilojoules]

Looks great

[kilojoules]

I would like to support numpy > 2

[bjarketol]

Yes, agree, once @koendev has a numpy 2 update ready for wayve ready we can change the pin

[bjarketol]

Loosened the numpy pin to "numpy>=1.22,<3.0"

[kilojoules]

@koendev, you can use something like the code below to support both numpy 1 and 2:

        _trapezoid = np.trapezoid if hasattr(np, "trapezoid") else np.trapz
        U3 = _trapezoid(np.interp(z, zs, us), z) / (15.0e3 - h)
        V3 = _trapezoid(np.interp(z, zs, vs), z) / (15.0e3 - h)

I thied this and I can't get it to work. Tests are still failing.

#49

[bjarketol]

@koendev, you can use something like the code below to support both numpy 1 and 2:

        _trapezoid = np.trapezoid if hasattr(np, "trapezoid") else np.trapz
        U3 = _trapezoid(np.interp(z, zs, us), z) / (15.0e3 - h)
        V3 = _trapezoid(np.interp(z, zs, vs), z) / (15.0e3 - h)

I thied this and I can't get it to work. Tests are still failing.

#49

It needs to be fixed upstream in wayve

[koendev]

Hi all, wayve has been updated to use numpy >2. Please use the main branch, the couplings to foxes and PyWake are available there as well.