PV Capacity Testing for Complex Sites: Multiple Orientations, Bifacial Modules, and GHI-Only Sensors

The ASTM E2848 regression takes a single irradiance value as its primary input:

$P = E_{POA} \times (a_{1} + a_{2} E_{POA} + a_{3} T_{a} + a_{4} v)$

That works cleanly on a single-orientation, monofacial site with a pyranometer in the plane of the array. Plenty of real sites are not that, and three deviations come up constantly:

Multiple array orientations, common on commercial rooftops and behind-the-meter sites, where there is no single plane of array.
Bifacial modules, where a meaningful share of the energy arrives at the back of the panel and the standard has nothing to say about it.
Limited sensors, where putting a POA pyranometer on every orientation costs more than the test is worth.

None of these require departing from E2848. All three are handled by computing the right $E_{POA}$ before the regression runs. This article covers each, and — since real sites often present all three at once — the order they have to be applied in.

Multiple Array Orientations

When a site has several orientations, each with its own POA sensor, the measurements must be combined into one representative irradiance for the whole site.

Module Area Weighting

The standard recommendation is a module area-weighted average:

$E_{POA} = \frac{\sum _{i} E _{POA, i} \times N _{i}}{N}$

where:

$E_{POA}$ is the weighted average POA irradiance
$E_{POA, i}$ is the measured POA irradiance of the i-th array
$N_{i}$ is the number of modules in the i-th array
$N$ is the total number of modules at the site

Two arrays with different POA sensors — 850 W/m² across nine modules and 930 W/m² across four — combined by module count into a single weighted average of 874 W/m² Weighting by module count pulls the site value toward the larger array: 874 W/m², not the 890 W/m² a simple average would give.

Weighting by STC Rating

If the site mixes modules with different STC (Standard Test Conditions) ratings — common on sites built in phases, or repowered — the weighting must account for rating as well as count:

$E_{POA} = \frac{\sum _{i} E _{POA, i} \times N _{i} \times ST C _{i}}{\sum _{i} N _{i} \times ST C _{i}}$

where $ST C_{i}$ is the STC rating of the modules in the i-th array.

This is the general form: module area weighting is the special case where every $ST C_{i}$ is equal. HelioTest implements the STC-weighted method for that reason, and it reduces to the simpler result automatically on a uniform site.

Whichever form applies, it must be applied identically to the measured and the modeled datasets. Weighting the measured irradiance by module count while the model reports a simple average is a subtle asymmetry that will bias the capacity ratio without producing any obvious symptom.

Bifacial Modules

A tilted bifacial array receiving direct sunlight on the front face and ground-reflected light on the rear face The rear face collects ground-reflected light the standard's single irradiance term has no place for.

The published version of ASTM E2848 does not address bifacial systems. Established practice is to replace the front-side POA irradiance in the regression with a total effective irradiance that folds in the rear contribution:

$E_{TOTAL} = E_{POA,front} + E_{POA,rear} \times ϕ \times (1 - SSF)$

where:

$E_{POA,front}$ is the front-side POA irradiance
$E_{POA,rear}$ is the rear-side POA irradiance
$ϕ$ is the module's bifaciality factor, typically 70–95 %
$SSF$ is the rear structural shading factor — the production loss caused by the mounting structure shading the back of the module, typically 3–10 %

$E_{TOTAL}$ then substitutes for $E_{POA}$ in the regression, and the test proceeds exactly as it would on a monofacial site. Separate regressions are still fitted to measured and modeled data, evaluated at the same reporting conditions, and compared to produce the Capacity Test Ratio.

Working with PVsyst

In a PVsyst export, the GlobBak column is rear POA irradiance that already accounts for structural shading. For the modeled dataset, the calculation therefore simplifies to

$E_{TOTAL} = E_{POA,front} + Gl o b B ak \times ϕ$

Applying $(1 - SSF)$ a second time here is a double-count, and it is an easy one to make because the measured-side formula does need it. It shows up as a modeled capacity that is quietly too low, which flatters the test result.

Sites With Only GHI Sensors

On small and mid-size installations with several orientations, a pyranometer per plane can be prohibitively expensive — often a significant fraction of the cost of the test itself. E2848 can still be run by measuring global horizontal irradiance and transposing it to each plane.

What You Need to Measure

Monofacial: at least one GHI sensor.
Bifacial: an additional rear-facing (downward) GHI sensor.
Recommended in both cases: a DHI sensor. Diffuse measurement materially improves transposition accuracy, and on a site where transposition is already introducing uncertainty, it is usually worth the cost.

Transposing GHI to POA

Sub-hourly GHI data is imported into an established modeling tool — PVsyst, NREL's System Advisor Model, or pvlib — which applies a transposition model such as Perez-Ineichen or Hay to compute POA irradiance for each array orientation.

A single GHI measurement passed through a transposition model to produce three separate POA irradiance values, one per array orientation One horizontal measurement, transposed separately onto each plane. The weighted average from the previous section is then applied to these outputs.

Use the same transposition model on both sides of the comparison. PVsyst uses Perez-Ineichen by default, so HelioTest uses Perez-Ineichen too; transposing the measured data with one model while the baseline was built with another introduces a systematic offset that has nothing to do with plant performance. Where DHI measurements are available, HelioTest uses them to improve the transposition.

Rear Irradiance From a GHI Pair

For a bifacial site instrumented with front and rear GHI sensors rather than rear POA sensors, the rear plane-of-array irradiance is estimated by scaling the front POA by the measured GHI ratio:

$E_{POA,rear} = E_{POA,front} \times \frac{G H I _{rear}}{G H I _{front}} \times ϕ \times (1 - SSF)$

This applies to the measured data only. On the modeled side, PVsyst supplies front POA (GlobInc) and rear POA (GlobBak) directly, and as noted above GlobBak already includes structural shading:

$E_{POA,rear} = Gl o b B ak \times ϕ$

In both cases the total is then

$E_{TOTAL} = E_{POA,front} + E_{POA,rear}$

Putting All Three Together

On a site that is multi-orientation, bifacial, and GHI-instrumented, the order of operations matters:

Transpose measured GHI to a front POA irradiance for each array orientation.
Weight those per-orientation values into a single site-representative front POA, using the STC-weighted average.
Add the rear contribution to produce the total effective irradiance.
Regress measured and modeled data separately against that value, evaluate both at the reporting conditions, and take the ratio.

The tempting shortcut here is to skip step 2 altogether: average the array orientations into one representative tilt and azimuth, transpose the GHI once onto that single plane, and call the result the site POA. One transposition instead of three, and no weighting to do.

It is wrong because transposition is nonlinear in orientation. The POA irradiance at the average orientation is not the average of the POA irradiances at each orientation, and the gap between them widens with the angular spread between the arrays — precisely the situation that made a weighted average necessary in the first place. A site with arrays at due east and due west is the extreme case: their mean azimuth is due south, an orientation on which the site has no modules at all.

The weighting has to be applied to the transposed outputs, never to the orientations feeding into them.

Is a Dedicated POA Sensor Worth It?

If one orientation dominates the site, installing a real POA pyranometer for that array and transposing only the remainder can be a sound economic trade-off: most of the site's output is then measured directly, and transposition uncertainty is confined to the minority of the capacity.

This is worth pricing out on sites where a single orientation carries more than roughly two-thirds of the capacity. Note that HelioTest does not currently support mixing a measured POA with transposed POAs within one test — it is transposition for all orientations or measured POA for all of them.

How HelioTest Handles This

You configure each sub-array's orientation, module count, STC rating, and — for bifacial systems — the bifaciality factor and structural shading factor. From there the platform applies the STC-weighted average, the transposition, and the bifacial correction consistently to both the measured and modeled datasets, which removes the asymmetries described throughout this article as a class rather than one at a time.

There are no spreadsheet formulas to maintain, and the weighting and transposition parameters are recorded with the test, so a reviewer can see exactly what was applied.