pvlib.bifacial.infinite_sheds.get_irradiance_poa output does not change when changing some input parameters

[kurt-rhee]

Describe the bug
https://pvlib-python.readthedocs.io/en/stable/reference/generated/pvlib.bifacial.infinite_sheds.get_irradiance_poa.html#pvlib.bifacial.infinite_sheds.get_irradiance_poa

The output of the function above does not change when the the following parameters change:

• height
• pitch
• model
• dni_extra
• iam # this one may be intended?

                rear_irradiance_group = get_irradiance_poa(
                    surface_tilt=group_df["rear_surface_tilt"],
                    surface_azimuth=group_df["rear_surface_azimuth"],
                    solar_zenith=group_df["apparent_zenith"],
                    solar_azimuth=group_df["azimuth"],
                    gcr=gcr,
                    height=99,
                    pitch=99,
                    ghi=group_df["ghi"],
                    dhi=group_df["dhi"],
                    dni=group_df["dni"],
                    albedo=ALBEDO,
                    model="hay_davies",
                    dni_extra=group_df["dni_extra"],
                    iam=1.0,
                    npoints=6,
                    vectorize=True,  # Vectorize is fine for time-series inputs
                )

To Reproduce
Steps to reproduce the behavior:

Some sample data

string_id met_name                      time  surface_tilt  surface_azimuth         ghi        dhi         dni  apparent_zenith    azimuth    dni_extra  pitch  racking_equipment_id  racking_controls_gcr  module_equipment_id  bifaciality_factor  pile_height  rear_surface_tilt  rear_surface_azimuth
122780        429       05 2025-08-18 07:15:00-06:00     11.563849        90.000000   64.606815  23.909445  482.167134        85.023666  77.323903  1327.346081   5.05                     8                   0.4                   12               0.791          1.6         168.436151            270.000000
122781        429       05 2025-08-18 07:20:00-06:00     13.798609        90.000000   79.229623  26.120372  526.216676        84.090495  78.089598  1327.346081   5.05                     8                   0.4                   12               0.791          1.6         166.201391            270.000000
122782        429       05 2025-08-18 07:25:00-06:00     16.098271        90.000002   94.731742  28.557150  563.347112        83.150932  78.853213  1327.346081   5.05                     8                   0.4                   12               0.791          1.6         163.901729            270.000002
122783        429       05 2025-08-18 07:30:00-06:00     18.475625        90.000000  117.185185  39.581374  579.055416        82.206119  79.615158  1327.346081   5.05                     8                   0.4                   12               0.791          1.6         161.524375            270.000000
122784        429       05 2025-08-18 07:35:00-06:00     20.947126        90.000002  136.620012  46.861257  596.118306        81.256884  80.375843  1327.346081   5.05                     8                   0.4                   12               0.791          1.6         159.052874            270.000002
122785        429       05 2025-08-18 07:40:00-06:00     23.533385        90.000000  158.033562  55.620140  612.808995        80.304018  81.135684  1327.346081   5.05                     8                   0.4                   12               0.791          1.6         156.466615            270.000000
122786        429       05 2025-08-18 07:45:00-06:00     26.261737        90.000000  172.369276  50.339806  664.439619        79.348009  81.895102  1327.346081   5.05                     8                   0.4                   12               0.791          1.6         153.738263            270.000000
122787        429       05 2025-08-18 07:50:00-06:00     29.168602        90.000000  190.162325  53.228972  684.073245        78.389286  82.654525  1327.346081   5.05                     8                   0.4                   12               0.791          1.6         150.831398            270.000000
122788        429       05 2025-08-18 07:55:00-06:00     32.304751        90.000001  215.508648  69.363785  674.533014        77.428165  83.414384  1327.346081   5.05                     8                   0.4                   12               0.791          1.6         147.695249            270.000001
122789        429       05 2025-08-18 08:00:00-06:00     35.744402        90.000001  226.806553  59.080368  719.509943        76.464956  84.175121  1327.346081   5.05                     8                   0.4                   12               0.791          1.6         144.255598            270.00000

Expected behavior
A clear and concise description of what you expected to happen.

I would expect that a higher height would give a larger irradiance value

Screenshots
If applicable, add screenshots to help explain your problem.

Versions:

• pvlib.__version__: 0.13.0
• pandas.__version__: 2.2.3
• python: 3.12

Additional context

[mikofski]

Hi Kurt, the paper (https://doi.org/10.1109/PVSC40753.2019.8980572) looks mainly at center rows which are far enough from the front or back that they don’t have a clear view of the ground or the ground doesn’t have a clear view of the sky outside the array. Also because the ground is integrated first variation in irradiance doesn’t affect the profile of view factor along the length of the module. This was deemed acceptable because as height increased from zero to 2[m] above the ground the profile of Fsky->grnd flattened, and had the same integral for all heights.

A suggested improvement to the model is to use superposition to treat the ground as two surfaces one for shade and another for unshaded (or just split into 100 segments like Marion) but I think the largest improvement would be to consider how far away the front row & back rows are. The functions will account for ground and sky outside the array if you set those limits. I believe at one time there was a FIXME in the code to model the 2nd, 3rd, 4th rows, etc.

Or to put this another way, I feel this is a shortcoming of the Marion model & experiments with arrays of only 5 rows so it’s easy for the center row to see beyond the array, which in my opinion is unrealistic for a 100[MW] system. Or put even another way, infinite sheds has the ability to model infinite rows both in front and behind the row under consideration. Whether this is a strength or weakness depends on the system being modeled which is why I added the FIXME to the code. You might need to go back to my original branch to find it.

[kurt-rhee]

Thanks Mark! Do you happen to have a copy of the paper for me to read? I am not an IEEE member this year. If not, no worries I can ask my workplace to purchase it for me.

[williamhobbs]

@kurt-rhee, here's a pre-print: https://mikofski.github.io/pvsc46_solarfarmer_bifacial/SolarFarmerBifacialModeling.pdf

[kurt-rhee]

Thank you @williamhobbs !

[kurt-rhee]

I just gave the paper a read and I now understand why height may not affect the model outputs, but I still don't understand some of the other parameters.

For example, here are the results of a bifacial simulation using df.describe() using a pitch hardcoded to 5m (first), 10m (second) and 100m (third). I would have guessed that the difference in poa_global would be higher as pitch varies.

        poa_global  poa_direct  poa_diffuse  poa_ground_diffuse  poa_sky_diffuse  shaded_fraction
count  5635.000000      5635.0  5635.000000         5635.000000      5635.000000          10045.0
mean     55.859816         0.0    55.859816           49.457470         6.402346              0.0
std      41.411760         0.0    41.411760           42.728887         4.985205              0.0
min       1.089290         0.0     1.089290            1.086769         0.000707              0.0
25%      15.268487         0.0    15.268487            5.638188         1.543563              0.0
50%      50.793646         0.0    50.793646           39.560659         6.509789              0.0
75%      96.330675         0.0    96.330675           91.694704        10.446087              0.0
max     118.685759         0.0   118.685759          118.684016        29.162632              0.0

        poa_global  poa_direct  poa_diffuse  poa_ground_diffuse  poa_sky_diffuse  shaded_fraction
count  5635.000000      5635.0  5635.000000         5635.000000      5635.000000          10045.0
mean     56.356139         0.0    56.356139           49.953793         6.402346              0.0
std      41.245335         0.0    41.245335           42.513648         4.985205              0.0
min       1.088969         0.0     1.088969            1.086448         0.000707              0.0
25%      16.317507         0.0    16.317507            6.379204         1.543563              0.0
50%      50.979821         0.0    50.979821           39.758618         6.509789              0.0
75%      96.533843         0.0    96.533843           91.897872        10.446087              0.0
max     118.675837         0.0   118.675837          118.674094        29.162632              0.0

        poa_global  poa_direct  poa_diffuse  poa_ground_diffuse  poa_sky_diffuse  shaded_fraction
count  5635.000000      5635.0  5635.000000         5635.000000      5635.000000          10045.0
mean     56.509388         0.0    56.509388           50.107042         6.402346              0.0
std      41.546577         0.0    41.546577           42.835093         4.985205              0.0
min       1.096453         0.0     1.096453            1.093933         0.000707              0.0
25%      15.937848         0.0    15.937848            6.089170         1.543563              0.0
50%      51.526062         0.0    51.526062           40.402832         6.509789              0.0
75%      96.998577         0.0    96.998577           92.362607        10.446087              0.0
max     120.533110         0.0   120.533110          120.368342        29.162632              0.0

[cwhanse]

That function returns POA irradiance on one face, not the sum of front and back. The face is implied by surface_tilt. So I suspect you are seeing only front side irradiance.

What do you get if you use infinite_sheds.get_irradiance?

Suggestions welcome for improving those docstrings.

[kurt-rhee]

I do hope it is a simple configuration issue.

I have been using the _backside function to reverse the surface tilt and azimuth. I notice that if I don't use the _backside function I get much higher values that I expect are representative of the frontside irradiance.

I do get similar results when I use get_irradiance without the _backside function as get_irradiance_poa with the _backside function.

            # Reverse surface angles to get the rear-side orientation
            (
                bifacial_inputs["rear_surface_tilt"],
                bifacial_inputs["rear_surface_azimuth"],
            ) = _backside(
                tilt=bifacial_inputs["surface_tilt"],
                surface_azimuth=bifacial_inputs["surface_azimuth"],
            )

            # Define columns that must be passed as scalar floats to the pvlib function
            # These will be our grouping keys.
            grouping_cols = ["racking_controls_gcr", "pile_height", "pitch"]

            results_list: list[pd.DataFrame] = []

            # Loop over each unique combination of scalar inputs
            for _, group_df in bifacial_inputs.groupby(grouping_cols):
                gcr = group_df["racking_controls_gcr"].iloc[0]
                height = group_df["pile_height"].iloc[0]
                pitch = group_df["pitch"].iloc[0]

                # Call the pvlib function for this subset of data
                rear_irradiance_group = get_irradiance_poa(
                    surface_tilt=group_df["rear_surface_tilt"],
                    surface_azimuth=group_df["rear_surface_azimuth"],
                    solar_zenith=group_df["apparent_zenith"],
                    solar_azimuth=group_df["azimuth"],
                    gcr=gcr,
                    height=height,
                    pitch=100,
                    ghi=group_df["ghi"],
                    dhi=group_df["dhi"],
                    dni=group_df["dni"],
                    albedo=ALBEDO,
                    model="hay_davies",
                    dni_extra=group_df["dni_extra"],
                    iam=1.0,
                    npoints=6,
                    vectorize=True,
                )
                results_list.append(rear_irradiance_group)  # type: ignore

[cwhanse]

Hmm. I can't see a disconnect in the tilt and azimuth part, but there's this:

                    pitch=height,

Intended?

[kurt-rhee]

Not intended, I'll edit the example above.

[kandersolar]

• height: not sure without digging into the math, but it seems plausible that it ends up canceling out in all the averaging that infinite_sheds does.
• pitch: make sure you also change the gcr input accordingly. The code below shows how changing pitch (and gcr) produces the expected results. Note that for pitch of 100m but a normal gcr and height, much of the module surface will be underground...
• model, dni_extra: this is because it has to be model='haydavies', not model='hay_davies'. Maybe the function should be changed to raise an error for invalid model strings.
• iam: this is applied only to direct irradiance, which is zero on the rear-side of a tracker.

import pvlib
import pandas as pd
import matplotlib.pyplot as plt

times = pd.date_range("2019-06-01", "2019-06-02", freq="5min", tz="Etc/GMT+5")
location = pvlib.location.Location(40, -80)
sp = location.get_solarposition(times)
cs = location.get_clearsky(times, solar_position=sp)
tr = pvlib.tracking.singleaxis(sp.zenith, sp.azimuth, backtrack=True, gcr=0.4)

dniet = pvlib.irradiance.get_extra_radiation(times)
albedo = 0.2

tilt_rear, azimuth_rear = pvlib.bifacial.infinite_sheds._backside(tr['surface_tilt'],
                                                                  tr['surface_azimuth'])

collector_width = 2  # meters, constant
height = 3  # meters, constant

for pitch in [3, 5, 10]:
    gcr_calc = collector_width / pitch
    irrad = pvlib.bifacial.infinite_sheds.get_irradiance_poa(
        surface_tilt=tilt_rear,
        surface_azimuth=azimuth_rear,
        solar_zenith=sp.zenith,
        solar_azimuth=sp.azimuth,
        gcr=gcr_calc,
        height=height,
        pitch=pitch,
        ghi=cs.ghi,
        dhi=cs.dhi,
        dni=cs.dni,
        albedo=albedo,
        dni_extra=dniet,
        model='haydavies',
    )
    
    irrad['poa_global'].plot(label=f'{pitch=}')

plt.legend()
plt.ylabel('Rear-side irradiance [W/m2]')