The paper describes the UVC III campaign for calibrating and intercomparing solar UV radiometers, which was held in Davos, Switzerland, from June to August 2022, involving filter radiometers and the portable reference spectroradiometers QASUME and QASUMEII.  However, the focus is on incremental improvements of the radiative transfer modeling tool (UVIOS2), which was used to forecast the UV index (UVI) with inputs from satellite, reanalysis, and ground-based sources.

Comparisons with the reference QASUME UVI measurements were used to demonstrate overall good performance of the model for clear skies, i.e., when the sun was not covered by clouds.  However, much larger differences were found with instantaneous and daily UVI measurements, which were explained by cloud modeling challenges (Fig.5). Under cloud-free skies enhanced aerosol absorption, i.e., low single scattering albedo (SSA), might have explained model overestimation (Fig. 3 and 4), but there were no SSA measurements in UV to confirm this hypothesis.  

There is very brief mention of comparisons between QASUME and filter radiometers in section 3.3 and Figure 10 (previously published) shows that the results mainly depend on application of the consistent calibration factors (PMOD/WRC). This section needs to be either expanded or removed.

UVI references are incomplete.

The paper may be suitable for publication after improving quality of the figures and completeness of the text and addressing technical questions described below.

RT modeling approach.

More details are needed describing extraterrestrial solar irradiance source, e.g., spectral smoothing applied, comparison with the state-of-the-art satellite TSIS-1 hybrid solar reference spectrum [Coddington, et al., https://doi.org/10.1029/2022EA002637 ].

The aerosols are included into the cloudless LUT (Tables 1, 2). This is different to OMI and TROPOMI satellite UVI retrievals, where aerosol and cloud effects are parameterized as a separate scattering (Cc) and absorbing (Ca) correction factors, UV = Ca(SZA, AAOD)*Cc(SZA,COT,…)*UVclear (SZA,TOC,…) [Arola et al., 2021 https://doi.org/10.5194/amt-14-4947-2021]. This explicit absorbing aerosol correction based on aerosol absorption optical depth (AAOD) would be especially important for North Africa and Middle East sites affected by desert dust, e.g. Roshan et al., Atmosphere 2020, 11, 96; doi:10.3390/atmos11010096.

Using aerosol optical thickness in UV (e.g., 340nm or 380nm) would be more appropriate as inputs to UVIOS2 model, because extrapolating visible AE would result in systematic overestimation of AOD in UV, e.g., see Fig 1 in Eck, et al., ‘‘Wavelength dependence of the optical depth of biomass burning, urban and desert dust aerosols,’’ J. Geophys. Res. 104, 31333–31350, 1999.

Using cloud optical thickness in UV would be more accurate, e..g., Krotkov,et al., "Satellite estimation of spectral surface UV irradiance 2. Effects of homogeneous clouds and snow", J. Geophys. Res., http://doi.wiley.com/10.1029/2000JD900721

Measurements:

High mountain site is not ideal for the absolute hemispherical irradiance measurements due to horizon obstruction by mountains. Provide mountain elevation at the measurements site as function of the azimuth (in Figure 1) and estimate horizon blockage correction, which needs to be applied to the model and/or measurements.

Clarify the difference between “clear-sky” (i.e., sun not blocked by clouds [line 275]) and “cloudless” (i.e., “clear sky”, [line 180]) conditions. Provide separate comparisons statistics for completely cloud-free periods.

 Describe correction for a non-lambertian angular response of the QASUME and radiometers involved into the UVC III campaign.

Technical comments:

Figure 1: It would be useful to add a panoramic photo of the site and angular horizon elevation table for the observation site at PMOD. Calculate the correction factor in UVIOS2 to account for the horizon blockage effect at different SZAs.

Figures 2, 8-9: Add year in X-axis. Use logarithmic Y-scale.  Symbols are difficult to see. Use different and larger symbols and line styles.

55 future climatic changes – climate changes

73-74. limited by the finite width of the satellite pixel – reword

74 weakness of satellite sensors – need clarification

77-78: Copernicus Atmospheric Monitoring Service (CAMS) – Atmosphere 

100 information of the public – information to the public

117 reconstructed UVI series  - reconstruct

118 The UVIOS (UV-Index Operating System) nowcasting system that its basic features have been already described … - reword sentence

126 summarized as follows – use colon :

147 data were used as a reference

168 serves as a reference

197 atlas plus modtan extraterrestrial spectrum – What was spectral resolution of ETS? Was a spectral smoothing and Sun-Earth distance correction applied?  Compare with the TSIS-1 HRRS [Coddington et al., https://doi.org/10.1029/2022EA002637]

201 The US standard atmosphere (Anderson et al., 1986) was used – This model was not developed for a mountainous Davos site.

202 the surface albedo was set to 0.05 – this may not be representative for N. Africa or Middle East sites.

205 A correction for the effect of altitude, assuming an increase of 5% per km – There is a strong spectral dependence of the UV increase with altitude ~5% at 330nm to ~10% at 290nm, e.g., see Fig. 7 in Krotkov et al., JGR, http://doi.wiley.com/10.1029/98JD00233

230 Analyses of different AERONET datasets shows – show

231 around a typical [value]

232 Given that ASY generally increases? with wavelength - ASY should decrease with wavelength

241-244 Table 2: If input parameters are the same (SSA, ASY, surface albedo) they do not need to be included in the table.

245: Re-word the sentence.

267 Level 2 AERONET retrievals were not used because they are not available yet. – They are available with a longer latency and could be used for reanalysis.

268 nearly real time  - near real time

275 For the analysis, measurements were classified as clear-sky (i.e., sun was not fully or partially covered by clouds) – This classification is not consistent with the “clear-sky” assumption in UVIOS2 model, where “clear-sky” is defined as “cloudless” conditions (line 182). This leads to inconsistencies in “clear sky” model to measurement comparison results.

287: Under clear-skies – This case includes scattered clouds not blocking the sun. It would be useful to show a separate comparison for the cloud-free periods in Fig 2.

287-288: Remove “both”

Figure 2: Add Year in X-axis. Symbols are difficult to discriminate. Use different and larger symbols and different line styles. It would be useful to show cloud-free periods using different symbols.

Calculating average UVI ratio between DOY 190 and 200 would result in positive bias, while the bias is negative between DOY 200 and 210. Is there an explanation?

295 Figure 2 shows that using highly accurate inputs for TOC, AOD at 500 nm, and AE does not result in a noticeable improvement in the accuracy of the modeled average  clear-sky UVI.  StDev decrease by less than 10% by using GB inputs

302 Differences in AOD are in all cases within ± 0.1 - There are larger differences in Fig. A1

304-305 on average, TEMIS slightly underestimates TOC – TEMIS TOC is higher than Brewer TOC in Fig. A2

309: differences in AOD – Use Brewer measured AOD.

313: ranging from values smaller than 0.8 (during e.g., events of dust or biomass burning aerosols – These events are not typical for Davos location. Please, provide evidence if such events did occur during  UVC-III campaign.

Figure 3. Why show a hypothetical case with SSA=0.8 which is not representative for UVC-III campaign?

327 which denotes that the SSA – which means that the SSA

Figure 4. – Suggest moving this figure to supplement. You can use AERONET SSA retrievals on days 197-199.

354-355: Although we have not corrected the modeled UVI for the effect of limited horizon – This horizon correction should be important for Davos site. Quantify this effect using horizon elevation angle as a function of the azimuthal angle.

Figure 8: analysis of the outliers will be useful.

Figure 9. The campaign average difference is close to zero, but there are certain periods (i.e., 200-210) with larger  differences.  Again, analysis of the largest outliers would increase the value of the comparisons.

390-400: Section 3.3 is too short. The results in Figure 10 are not discussed. Expand or remove this section.

394: when the PMOD/WRC calibration – explain the difference between USER and POD/WRC calibration. Explain if  radiometers were calibrated for the non-lambertian angular response (cosine correction)?

405 Figure 10: Text in the figure is difficult to read. Try to increase the size of the text or move the text to the caption. 

426-427: when solar disc is occluded, we do not know the exact COT. – Clarify this sentence.

436. shows the significance of systematic and accurate calibration of such instruments. – This is true regardless of the model performance …

437 discussed in previous studies – add reference to Fioletov, et al., (2004) “UV index climatology over North America from ground-based and satellite estimates”,  J. Geophys. Res., 109,  D22308, http://doi.wiley.com/10.1029/2004JD004820

440 associated to the assumptions – with the assumptions

451 not available (e.g., Bais et al., 2019). – add these references:

Krotkov, et al., “Aerosol UV absorption experiment (2002- 04): 2. Absorption optical thickness, refractive index, and single scattering albedo”, Opt. Eng., 44(4), 041005, http://doi.org/10.1117/1.1886819 , 2005,

Corr, Chelsia, et al., “Retrieval of aerosol single scattering albedo at ultraviolet wavelengths at the T1 site during MILAGRO (2009)”, Atmos. Chem. Phys., 9, 5813–5827, http://doi.org/10.5194/acp-9-5813-2009

Mok, J., et al., “Impacts of atmospheric brown carbon on surface UV and ozone in the Amazon Basin”, Sci. Rep. (2016); https://doi.org/10.1038/srep36940  

Mok, J., et al., “Comparisons of spectral aerosol absorption in Seoul, South Korea”, Atmos. Meas. Tech., 11, 2295-2311, https://doi.org/10.5194/amt-11-2295-2018

Go, et al., “Ground-based retrievals of aerosol column absorption in the UV spectral region and their implications for GEMS measurements”. Remote Sensing of Environment, 245, 2020, 111759,  https://doi.org/10.1016/j.rse.2020.111759

GENERAL COMMENTS

The manuscript by Fountoulakis et al. presents the results of a modelling system for estimating the UV Index at the surface (UVIOS2) during an international comparison campaign of broadband UV radiometers in Davos. The system is described, and the results are illustrated for both (i) clear-sky days and (ii) all-sky conditions. The factors influencing the comparison are discussed.

In my opinion, the research topic is important and within the scope of the journal, and the results are in good agreement with the reference instrument. However, the description of the methodology and the presentation of the results could be improved. Hence, I recommend publication after the authors address the issues outlined below.

SPECIFIC COMMENTS

• The study evaluates the performance of the UVIOS2 system during a summer campaign (June–August). The authors should be cautious not to extrapolate these results to seasons when snow is present near the measurement area, as surface albedo determination and the estimation of cloud properties from satellite data could be problematic. Additionally, at higher altitudes and latitudes, snow or ice may persist even in summer. A discussion on this topic would be beneficial. A brief reference is given in lines 423–424 ("we expect that the assumption of homogeneity in the satellite pixel would still be the main uncertainty factor"), but the basis for this expectation is unclear. Could the authors provide a bibliographic reference? Moreover, depending on the SEVIRI channel used for retrieving the COT, is there sufficient contrast between snow and clouds?
• Role of broadband radiometers in the manuscript: The authors state in the abstract that "Radiometers that were not properly maintained and/or calibrated were found to provide UVI measurements with uncertainty that was comparable to the uncertainty of the UVIOS2 estimates, which highlights the significance of systematic maintenance and calibration of the UV radiometers." They further state (lines 434–436) that "The uncertainty in the UVIOS2 forecasts was found to be comparable to the measurements of filter radiometers when they were not properly calibrated, but ~3 times larger compared to the measurements of accurately calibrated radiometers, which shows the significance of systematic and accurate calibration of such instruments." However, is a model really needed to affirm the importance of systematic and accurate calibration of UV instruments? The comparison of the model accuracy with that of the least reliable broadband radiometers is not particularly insightful. For example, should the conclusion be that UVIOS2 estimates are only as accurate as an uncalibrated or poorly maintained radiometer? In summary, are considerations regarding broadband radiometers necessary in this paper?
• Model inputs require a more detailed description:

1. Extraterrestrial spectrum: A more recent extraterrestrial spectrum (QASUMEFTS) exists compared to "Atlas plus MODTRAN." Given that QASUMEFTS was obtained using the same instrument employed here as a reference, why was it not used for the calculations? What is the expected deviation resulting from this choice?
2. Cross-sections: These should be presented in a separate table. Are the ozone cross-sections used in the model consistent with Brewer retrievals?
3. Total Ozone Column (TOC): How was the TOC from the Brewer instrument obtained? Additionally, why was TEMIS ozone used instead of CAMS, given that other parameters are derived from CAMS?
4. Cloud properties: What is the spatial resolution of the COT and CMFUV products? This information is crucial for understanding the influence of shadows/3D effects and parallax errors in cloud property determination over complex terrain. Furthermore, which SEVIRI channels were used for retrieving cloud optical properties? Do they enable reliable discrimination between clouds and surface snow?
5. Altitude correction for the station (line 270): How was it determined? Is it just the 5% factor per km mentioned a few lines above?

• Results for all-sky simulations: It would be beneficial to compare CMFs in addition to UV Indices to exclude the trivial correlation arising from the diurnal SZA cycle.
• Classification of clear/cloudy skies: The classification, based on measured direct irradiance (lines 275–279), appears simplistic. For instance, the situation shown in the upper-right image of Fig. 5 does not seem to represent clear-sky conditions (as defined by the model). The authors could refine their classification to better account for scattered clouds and their effects on surface UV radiation.
• Figures: Three different types of plots are used: (i) ratios (e.g., Figs. 2–3), (ii) absolute UVI values (e.g., Figs. 6–7), and (iii) absolute differences (Figs. 8–9). Can these be more homogenised for consistency?
• Other services, such as TEMIS and CAMS, already provide UV Index estimates, at least on a European scale. The authors should clarify the advantages offered by UVIOS2, such as improvements in spatial and temporal resolution.
• Spectral validation: Have the authors attempted to compare the spectral output of UVIOS2 with spectra measured by QASUME-II? This could provide additional insights into the influence of various factors.   

TECHNICAL REMARKS

- Lines 20–21: "The UVIOS2... depending on the inputs" is too general and can be removed.

- Lines 76–84: The distinction between nowcasting and forecasting is unclear. If all inputs are forecasted (as suggested in lines 76–77), what differentiates nowcasting from forecasting?

- Line 118: "system that its basic features" → Incorrect grammar; rephrase.

- Line 134: Remove the typo "as Heading 2."

- Lines 136–139: The stated considerations apply to all radiative transfer models, not specifically to UVIOS2. The description of UVIOS2 should begin with details that are unique to this system.

- Figure 1: What does "% altitude" represent on the colour scale?

- Section 2.2 (QASUME-II vs QASUME): The uncertainty of QASUME is discussed in Sect. 2.2, but QASUME-II is used in the study. Why was QASUME not used, and why is the uncertainty of QASUME-II not discussed?

- Table 2: A similar table for all-sky UV determination might be useful, with different headings.

- Figure 2: The y-axis label should clarify that the ratio represents simulated vs measured values. In addition, would plotting the ratio against solar zenith angle provide useful insights?

- Figure 3: "ratio BG" should be replaced with "UVIOS2 / QASUME" or a similar term to clearly define the ratio. Also, why are the plots not ordered by increasing SSA?

- Lines 330–331: If the aerosol is primarily secondary, could hygroscopicity have influenced the correlation between AOD and SSA?

- Line 342: Cloud-induced irradiance enhancements near the solar disk typically last for a short time. What do short-term variations in the ratio indicate about this effect?

- Lines 355–356: The treatment of limited horizon effects in the model is crucial due to the site complex orography. This information should be presented earlier. Additionally, provide an approximate (cosine-weighted) fraction of the sky obstructed by mountains.

- Figure 10: If retained, specify the UVIOS2 configuration used (in the caption).

- Lines 408–409: The past tense may be more appropriate in the conclusions.

- Table 3: It is unusual to introduce new results in the conclusions. Would it be better placed in the results section?

- Line 434: The term "UVIOS2 forecasts" is ambiguous here, as—if I understand correctly—clouds are not forecasted.