the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 06 Nov 2025
PV power modelling using solar radiation from ground-based measurements and CAMS: Assessing the diffuse component related uncertainties leveraging the Global Solar Energy Estimator (GSEE)
Abstract. Accurate PV power production modelling requires precise knowledge of the distribution of solar irradiance among its direct and diffuse components. Since this information is rarely available, this requirement can be addressed through the use of diffuse fraction models. In this study, we try to quantify the errors in PV modelling when measurements of the diffuse solar irradiance are not available. For this purpose, we use total and diffuse solar irradiance data obtained from ground-based measurements of BSRN to simulate the PV electric output using GSEE. We have chosen five sites in Europe and North Africa, with different prevailing conditions, where BSRN measurements are available. GSEE incorporates an implementation of the BRL diffuse fraction model, along with a Climate Data Interface that enables simulations across different time scales. We evaluate the capability of BRL in providing accurate estimations of the diffuse fraction under diverse atmospheric conditions, with particular attention on the presence of clouds and aerosols and assess the extent to which its associated errors propagate to energy production modelling. Furthermore, we compare GSEE outputs when using CAMS radiation time-series as input instead of ground-based measurements, to quantify the impact of the CAMS radiation product uncertainties in PV modelling.
Status: open (until 12 Dec 2025)
- AC1: 'Comment on egusphere-2025-4320', Nikolaos Papadimitriou, 05 Dec 2025 reply
-
RC1: 'Comment on egusphere-2025-4320', Anonymous Referee #1, 05 Dec 2025
reply
The manuscript “PV power modelling using solar radiation from ground-based measurements and CAMS: Assessing the diffuse component related uncertainties leveraging the Global Solar Energy Estimator (GSEE)” by Nikolaos Papadimitriou et al. focuses on the exploitation of the impact of the partitioning of the global horizontal irradiance (GHI) in its direct and diffuse components on the PV power production by simulations with the widely used Global Solar Energy Estimator (GSEE) model. The solar irradiance, air temperature, aerosol optical depth input sources are the BSRN and AERONET measurements from 5 sites in Europe, North Atlantic Ocean and Sahara desert and the CAMS model. The diffuse fraction (DF), being rarely measured except in a limited number of sites, is estimated within the GSEE through the logistic Boland-Ridley-Lauret (BRL) model, based on the clearness index as the main parameter. The effects of different cloud cover and aerosol optical depth on DF, and the corresponding impact on the simulation of power production is evaluated for fixed and 2-axis tracking PV systems based on c-Si and CdTe technology. The effects on different timescales are also explored. Finally, an assessment of the financial impacts deriving from evaluating the DF of desert dust from the BRL model compared to the measured GHI components, for a hypothetic PV solar farm around Tamanrasset is provided.
The main conclusion is that the best agreement in DF estimation with BRL model and ground-based measurements is for cloud-free and very load to moderate aerosol, i.e. for the simplest atmospheric conditions to be modelled. The worst situation for a reliable power production estimation is for partially cloudy skies. In sites impacted by high dust load, the BRL underestimated the DF and the power generation is overestimated.
General comments
While it is clear that the GSEE is widely used and the BRL model is optimized for both the Northern and Southern Hemispheres, I suspect it is not the best model for analyzing the effects of aerosols on the GHI partition, as the clearness index is mostly influenced by cloudiness and the reader is not informed about how the aerosols are accounted for. The authors should address this aspect, which is a key point in the development of the work.
as well as discuss similar works, if any, dealing with this topic.
Despite the large amount of calculations done in this work, the results are not valued by an adequate discussion, both in qualitative and quantitative terms. As a general comment I suggest quantifying the results in the text (Results and Conclusions sections) and not leaving the values only in the supplementary material.
The authors do not cite in the Introduction any previous paper dealing with the estimation of PV power generation under different cloudiness and/or aerosol load conditions, nor compare any of their results with previous works. If a similar work is not found in literature, this aspect, which increases the importance of this study, should be emphasized both in the introduction and in the conclusions.
Overall, I recommend a major revision of the key points before publication.
Specific comments
Lines 67-72: since the use of an empirical model constitutes a relevant part of the work, I suggest to detail a little the description of these models and in particular the description of the BRL model, for example by mentioning here the variables that are used to derive the DF.
Line 71: some information about the BRL model should be provided, since the model is widely used in the paper. Is this the only model for DF estimation incorporated in the GSEE?
Table 3 with the libRadtran input parameters is somehow unclear. The SZA input is “with step 90°”: what do the authors mean? In addition, is the wavelength dependence of the surface albedo, SSA and gg accounted for? Finally, I suggest “integrated water vapor” instead of “water vapor”.
Lines 257-264 and Figure 2: the comparison of the DF from the BSRN measurements and the BRL model is tricky. Do the differences for SZA>60° increase because of the difficulty of the model in estimating the DF for high SZAs? The bottom of Figure 2 shows that this happens for cloud-free conditions above 70° (not 60° SZA as said in the text) and not for all sites: Izana do not show the SZA dependence. It appears also for the cleanest conditions, i.e. for AOD500≤0.05.
In my opinion this deserves a little bit more investigation, instead of simply limiting the comparison to SZAs below 60°, as in Figure 3.
Line 281: under overcast conditions the BRL DF takes a range of values, approximately from 0.6 to 1, while the BSRN DF is close to 1. This means that even for homogeneous sky conditions, isotropic radiation the BRL model is not capable of providing reliable DF estimates. The authors could also refer to the 3D variability of cloud properties, whose effect cannot be accounted for by a model like BRL.
Section 3.2: the dependence on latitude and altitude is not discussed, although presented in Figure 5.
The results show that for totally scattering aerosols (SSA=1) the BRL model underestimates the DF, while for absorbing aerosols (SSA=0.7) is overestimates the DF.
The authors may briefly discuss the results of the sensitivity analysis in terms of how the BRL logistic model treats the aerosol-radiation interactions. The model does not explicitly include the aerosol optical properties, but incorporates their effects into the clearness index, together with those exerted by the clouds, which are by no means larger. In this section, instead, the authors examine in detail how aerosols influence the partitioning of GHI into direct and diffuse components, and the results obtained with the BRL model are strongly biased, as expected. This is also a consequence of how the model was conceived, in particular of the data with which the relationship between the DF and the geometric, meteorological and atmospheric variables were determined.
Section 3.3: the authors should comment the results presenting the quantification of the differences in power production derived from using only GHI or GHI and DHI as GSEE model input. Not all sites and all atmospheric conditions have to be considered, but at least for two sites with different characteristics, such as Izana and Lindenberg, for fixed and 2-axis tracking systems.
Lines 393-395: the data gaps are in the GHI data and/or in the irradiance components? Can the author suggest a reason for these gaps? Does using a less stringent condition on the number of days per month (for example 15 days) allow for a less fragmented annual evolution?
Lines 396-401: I find it very useful to present the differences also in percentage, referred to the energy production obtained using GHI and DHI as a reference.
Section 3.5: more details about the CAMS data selection are needed. For example: the authors say that the CAMS solar radiation data are adapted to the investigated sites, but how is this done? By interpolation, by considering the nearest CAMS grid point? Moreover, the Izana site is excluded from the analysis because the altitude of the site is not directly comparable to the model vertical grid. For this reason, it is useful to know the CAMS 3D spatial grid and to add it in Section 2.3.
Another missing point is the quantification of the irradiance differences on power production, a qualitative discussion is not sufficient.
Finally, the authors should refer to previous papers, if any, dealing with the use of CAMS irradiance data for modelling PV potential power production.
Conclusions: should summarize the main results and report some numerical data, which otherwise are only relegated in the supplementary material. For example, the results may be evaluated for tow or three sites with different characteristics in terms of latitude and aerosol regimes. I suggest reporting the impact on the power generation not only as absolute values, but also as percent, to facilitate understanding. In addition, even if the results pertaining to the panels with CdTe technology are not different from those obtained for c-Si panels, they should be briefly recalled.
When discussing the assessment of the financial impacts of the desert dust effects at Tamanrasset, the authors should clearly address that the analysis considers the differences in diffuse function derived from the measurements and calculated by the model. The sentence in lines 474-476 “Therefore, site assessments that do not account for the impact of desert dust aerosols may overestimate financial performance….” may be misleading, as it may be interpreted as the assumption that desert dust is not accounted for in the model simulations. I suggest reformulating the sentence.
Technical corrections
Line 32: the BRL acronym should be made explicit here.
Lines 53-54: the sentence “For practical reasons, it is critical to extend solar irradiance forecasting to encompass methods linked to solar-based power generation” is unclear.
Line 76: add “In” before “Regions dominated by…”.
Line 84: put a period after “ease of use”.
Line 113: use “air” or “ambient” temperature instead of simply “temperature”.
Line 116: maybe “radiation” instead of “actinometric”.
Line 117: can the authors quantify how much is the uncertainty on using a fixed surface albedo of 0.3?
Line 131: is this reference correctly cited?
Line 137: change “detailed” with “in detail”.
Table 1: is it necessary? In my opinion the text description is exhaustive.
Line 169: is it “daily” or “hourly”?
Line 188: change “regarding” with “based on”.
Line 201: the “c” letter is a typo.
Lines 223-224: I suggest to include a sentence on how the solar radiation is estimated in CAMS and the 3D spatial resolution, an important information for the CAMS data selection operated for the analysis described in Section 3.5.
Line 237: due the authors mean “the difference between libRadtran and BRL calculations” with “aerosol-related discrepancy”?
Line 255: I suggest “As a first step” instead of “Initially”.
Line 261: I would talk about “trend” but “behavior”.
Line 285: I suggest “A decrease for increasing AOD” instead of “A negative trend”.
Line 287: add “compared to the other sites”.
Line 309: maybe “ground altitude”?
Line 318: delete “regarding SSA”.
Line 345: the authors should introduce here that the produced energy is evaluated both for fixed and two-axis tracking PV systems.
Line 343: add “from BSRN” after “only GHI”.
Line 464: express the power overestimation also in percent number and the site where this is observed (should be Tamanrasset). Is the 49.2 Wh/kWp/hour on hourly value? I could not find this number in the supplementary material tables.
Line 469: Table 4 is not present.
Line 471: check the presence of an arrow in the text.
Figure 2: the caption may be improved by changing the text as follows: “Difference between the diffuse fraction estimated by the ground-based measurements and by using the BRL model as a function of SZA under diverse atmospheric conditions: (top) classification with respect to cloudiness and (bottom) classification with respect to aerosol optical depth”.
The diffuse fraction derived from the BSRN measurements is called in different ways in Figures 2, 3, and 4.
Figure 6: bottom graphs. Why the data produced with only GHI for Izana seem to be cut around 900 Wh/kWp per hour?
Figures 6 and 7: the caption should explain that the upper plots are for fixed systems and the lower one for 2-axis tracking systems.
Table S2: the metrics for high and very high aerosol load are missing.
Citation: https://doi.org/10.5194/egusphere-2025-4320-RC1
Viewed
Since the preprint corresponding to this journal article was posted outside of Copernicus Publications, the preprint-related metrics are limited to HTML views.
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 110 | 0 | 1 | 111 | 0 | 0 |
- HTML: 110
- PDF: 0
- XML: 1
- Total: 111
- BibTeX: 0
- EndNote: 0
Viewed (geographical distribution)
Since the preprint corresponding to this journal article was posted outside of Copernicus Publications, the preprint-related metrics are limited to HTML views.
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
The supplementary material referenced in the manuscript was not included in the initial preprint import from SSRN.
The Supplementary Material is now provided here as an attachment so that all supporting content is available to readers and reviewers.