Due to their high resolution, data sets of the A-sites required multiple pre-processing steps to deal with missing data points and outliers mainly. Small gaps ( < = 3 h) were filled using a linear interpolation. Larger gaps ( > 3 h) were filled by the corresponding hourly mean of the month, when available. At A2, a long interruption of part of the monitoring system caused the unavailability of P _DC,arr , G _POA and T _rear for the month of January 2023 and could not be replaced. The uptime for these variables was approximately 82% over the analysed period. Otherwise, the uptime reached 98.2% at A2 and 95.2% at A1. For M1, the obtained AC-power timeseries (15-min resolution) did not contain any irregularities.

Measurements are aggregated to monthly, seasonal and annual time resolutions. The adopted definition of the seasons is taken from the Swiss Energy Act (Energiegesetz, 2016) which splits the year in two semesters: summer from April 1 to September 30 and winter from October 1 to March 31. This choice is made to provide results that directly fit the legal framework of the analysed power plants. Only complete monitored seasons (i.e. 6 months long) are kept for the analysis. Note that monthly aggregates of additive variables are affected by the differences in day counts across months. For the purpose of the present analysis, normalisation was however not necessary. Further details on the data sets and their pre-processing are provided on the Envidat repository (Section 5).

The conditions of plane-of-array irradiation are evaluated using the reference yield, Y _r , which is the equivalent number of hours of test-condition irradiation (G _POA = G _0,POA ): Y r = H POA G 0 , P O A h (2) where H _POA [W/m ²] is the time integrated irradiation in the plane-of-array. The final yield that normalises the available energy produced E _AC is then: Y f = E A C P 0 k W h k W p (3) Y _f is the equivalent number of hours during which the PV system produced at the rated power P ₀.

The performance ratio (PR) corresponds to the fraction the actual produced energy represents with respect to the energy an ideal PV cell would produce in the actual irradiation conditions. It is a popular ratio at the plant level as it aggregates in a single number all losses occurring between the available radiative resources (Y _r ) and the actual produced power (Y _f ). This includes losses due to wiring, temperature, mismatch, clipping, etc. The norm IEC 61724-1 defines PR for bifacial modules (IEC, 2021): P R = Y f / Y r + ϕ ∗ Y r rear (4)

In the case of monofacial modules, Y r rear = 0 and PR = Y _f /Y _r .

To account for temperature conditions, Y _r can be corrected for cell temperatures deviating from STC (Kratochvil et al., 2004): Y r , T = 1 + γ T cell − 25 ∗ Y r h (5) where, γ is the tabulated power related temperature coefficient provided in %/°C, T _cell is the cell temperature and is derived from the backsheet module temperature using the Sandia model for an open-rack and glass-cell-glass module type (Kratochvil et al., 2004).

The capture thermal losses (L _cT ) can then be computed: L c T = Y r − Y r , T h (6)

Bifacial modules are well adapted to high albedo environments (e.g., Gu et al., 2020; Ganesan et al., 2023). Reported system bifacial gains in such environments range from 20% to 30% (e.g., Wang et al., 2020; Hayibo et al., 2022). Optical and system bifacial gains are defined as follows: B G opt = G POA rear / G POA (7) B G sys = E A C , b i − E A C , m o n o / E A C , m o n o (8)

The computation of BG _sys requires concurrent mono- and bifacial measurements which is only available at A1.

The primary interest is the comparison in seasonal production between Alpine-PV and PV installed in the midlands. Figure 4 shows the average and standard deviation of the annual final yields (Y _f ) measured between 2018 and 2022 at A1 and M1. The best performing module is A1-b60 with mean values of annual and winter yield reaching 1853 kWh/kWp and 889 kWh/kWp respectively. The best performing monofacial module is A1-m60 with mean values of 1547 kWh/kWp and 743 kWh/kWp respectively. Y _f,M1 reached 1090 kWh/kWp and is coherent with the typical annual production of Swiss PV systems (Hostettler, 2022). The minimum and maximum annual ratios between regions (Y _f,A /Y _f,M , Table 4) are 1.21 and 1.70 for A1-m90 and A1-b60 respectively. In winter, this ratio strongly increases up to 3.14 for A1-b60. In summer, performances at M1 increase and the same inter-region ratios decrease to 1.19 and 0.74.

The winter energy fractions are all close to 50% except for A1-m30 (40%). The maximum is reached by A1-m90 with 55%. It is much lower at M1 with only 26% (165 kWh/kWp).

Overall, regardless of their type and tilt, all alpine setups have performed better both in terms of annual yield and winter fraction.

Comparing now the different setups at A1, bifacial modules show higher final yields than their monofacial counterparts. The system bifacial ratios (BG _sys ) are 0.19 and 0.28 for tilts of 60° and 90° respectively. Such higher yields are expected for bifacial modules with respect to monofacial ones, especially in high-albedo ( > 0.7) environments (Burnham et al., 2019). Regarding the inclination of the modules, the yield results suggest that: (i) Steeper bifacial tilts get lower annual yields, but a higher winter fraction on average. (ii) The tilt that maximizes the annual yield is around 60° for both types of modules.

Figure 5 shows the mean monthly yields. It allows the identification of intra-seasonal patterns between winter (larger tilt angles) and summer (smaller tilt angles) optimized configurations, as well as between bifacial and monofacial modules.

The peak of production happens in March for winter optimized monofacial configurations and in April for the bifacial counterparts. Bifacial modules performances are well above the monofacial ones from November to June. This period corresponds to the presence of snow cover on the terrain at this elevation. This confirms the additional gain of bifacial modules in high-albedo environments.

The advantage of Alpine setups optimised for low sun elevation angles and a high-albedo environment, remains until May included. This explains the balance between the winter and summer fractions observed before (Figure 4). From June to August, both monofacial and bifacial configurations show a deficit with respect to M1.

The summer optimised setup at A1 (A1-m30) shows a flatter curve. The relative lower winter yields is compensated by higher ones in summer and leads to an annual yield that remains well above M1.

It has already been observed that bifacial modules showed significantly higher final yields than their monofacial counterpart. Using pairs of monofacial and bifacial modules with the same geometrical setup, the bifacial gains are evaluated. In the following, two assumptions are made: (i) The front side of the monofacial and bifacial modules pairs are considered identical. (ii) The irradiation point measurements are translated to the entire surface of the module, although partial shading due to structural elements and anisotropical reflections may modify the real irradiation of the modules on both sides.

Figure 6 presents the seasonal front-rear splits of reference yields (2) and final yields (3), for the three available pairs of monofacial and bifacial modules. The corresponding bifacial yields are provided in Table 4. Annually, the system bifacial gains range from 0.19 to 0.28 for A1-b60 and A1-b90 respectively. The annual optical gains are more than doubled, ranging from 0.43 to 0.47 for the same modules.

Annual reference yields (Figure 6A) are similar for all setups around 2300 h. The rear-side summer splits increase with tilt from 16% (A1-b60) to 19% (A1-b90). In winter the same splits all decrease to 13%–14%. Annually, approx. 30% of the total irradiation comes from the rear-side for all setups. The corresponding final yield splits (Figure 7B) show how much of the reference yields actually get converted into useable electricity. The ratio of the two yields is the performance ratio and is detailed further below.

It is here legitimate to question the difference in the final yields: although all setups have similar reference yields, the final yield significantly drops for A1-b90, in the summer. Low incident angles have a strong impact on PV modules efficiency (King et al., 2015) as compared to the pyranometers and may be the primary cause for this difference, in a season of high sun elevation angles. Rear-side splits show a stronger relative drop compared to the front-side. The losses on the rear-side are thus higher than for the front-side. For instance, the bifacial ratio was not accounted for in the computation of Y _r and is thus one component of the apparent losses that apply. Its correction is rather simple and will be applied on the performance ratio below. In addition, it is expected that rear-sides of bifacial modules show a higher mismatch especially in snowy environments that lead to a stronger anisotropy of the reflected irradiation (Wang et al., 2020).

The mean annual performance ratios are between 77% and 90%. Higher values ( > 85%) are obtained for the monofacial modules. This is expected because of the bifacial ratio and mismatch losses which are larger and more consistent on the rear side of the modules (Wang et al., 2020).

In order to isolate losses, PR can be further corrected for temperature. In this article however, it is computed indirectly through the capture thermal losses L _c,T , also reported in Table 4. Mean annual gains of 1%–2% due to temperature are observed for bifacial modules. Monofacial modules show smaller gain and even a 1% loss for A1-m70. Note here that the point measurement provided by the temperature sensors (T _bs ) is taken as the mean value for the whole module. In addition to the bias of the model used to derive T _cell (Kratochvil et al., 2004), the uncertainty sources in the computation of L _c,T are numerous. The obtained results however show some coherence with Haeberlin (2004) (1.5%).