UV-Indien Network ground-based measurements : comparisons with satellite and model estimates of UV radiation over the Western Indian Ocean

As part of the UV-Indien Network, 9 ground-based stations have been equipped with one spectroradiometer, radiometers and all-sky cameras. These stations are homogeneously distributed in 5 countries of the Western Indian Ocean region (Comoros, France, Madagascar, Mauritius and Seychelles), a part of the world where almost no measurements have been made so far. The main scientific objectives of this network are to study the annual and inter-annual variability of the ultraviolet (UV) radiation in this area, to validate the output of numerical models and satellite estimates of ground-based UV measurements, 5 and to monitor UV radiation in the context of climate change and projected ozone depletion in this region. The first results are presented here for the oldest stations (Antananarivo, Anse Quitor, Mahé and Saint-Denis). Ground-based measurements of UV index (UVI) are compared against satellite estimates (Ozone Monitoring Instrument (OMI), the TROPOspheric Monitoring Instrument (TROPOMI), the Global Ozone Monitoring Experiment (GOME) and model forecasts of UVI (Tropospheric Emission Monitoring Internet Service (TEMIS) and Copernicus Atmospheric Monitoring Service (CAMS). The median relative 10 differences between satellite or model estimates and ground-based measurements of clear-sky UVI range between – 34.5 % and 15.8 %. Under clear skies, the smallest UVI median difference between the satellites or model estimates and the measurements of ground-based instruments is found to be 0.02 (TROPOMI), 0.04 (OMI), -0.1 (CAMS) and -0.4 (CAMS) at St-Denis, Antananarivo, Anse Quitor and Mahé respectively. The cloud fraction and UVI diurnal profile are calculated for these four stations. The mean UVI values at local solar noon range between 10 (Antananarivo, Anse Quitor and Saint-Denis) and 14 at 15 Mahé. The mean UVIs in clear-sky conditions are higher than mean UVI in all-sky conditions, although it can still be noted that UVI maxima are higher for all-sky conditions than for clear sky conditions. This is the result of UVI enhancement induced by clouds, observed at these four stations. The greatest increase in UV radiation under cloudy conditions was observed at the 1 https://doi.org/10.5194/essd-2021-55 O pe n A cc es s Earth System Science Data D icu ssio n s Preprint. Discussion started: 3 March 2021 c © Author(s) 2021. CC BY 4.0 License.

Since April 2020, all stations for UVR measurements (one station at Rodrigues, two stations at Réunion, one station at Mahé, three stations at Madagascar, on station Juan de Nova and one station at Moroni) are operational, as are 4 stations for cloud cover measurements (see Table 1 and Figure 1).
All stations are now equipped with a Kipp & Zonen UVS-E-T broadband radiometer. Reunion Island is also equipped with 95 a Bentham UV spectroradiometer. The raw UV measurements obtained by the radiometers are reprocessed considering the calibrations and TOC measured simultaneously. The broadband radiometers will be calibrated every 2 years at the Moufia super site (University of Reunion Island, St Denis) using the Bentham UV spectrometer, itself calibrated every 3 months (Brogniez et al., 2016). The radiometer located at Saint-Denis was previously calibrated at Davos but is now calibrated using the Bentham measurements (Table 3). The TOC correction uses the total ozone measurements made with the UV-Visible 100 spectrometer SAOZ (Pastel et al., 2014;Toihir et al., 2015) on a smaller mesh size (due to the high cost of the instrument and also to the lower variability of ozone). For sites without local total ozone measurements, the TOC given by the OMI satellite is used. Currently, only sites in Reunion and Mahé are equipped with a SAOZ and there are plans to install a new one in Moroni (Ngazidja Island) in 2020 (UN Environment Programme under the Vienna Convention, funding acquired). The level 2 data (final data) obtained are archived on an ftp server accessible to all partners. Finally, cloud cover measurements are carried out 105 using the wide-angle total sky imager, SkyCamVision (SCV), manufactured by Reuniwatts (http://www.reuniwatt.com/). The camera acquires hemispherical images between 380 and 440 nm every minute. Cloud fractions (CF) are calculated following a cloud segmentation algorithm. (Liandrat et al., 2017;Breiman, 2001;Long et al., 2006). All the data can be easily downloaded (open data).
The scientific objectives of this network are the following:

110
-The study of the annual and inter-annual variability of UVR over the whole Western Indian Ocean basin.
-Validation of numerical model outputs for UVR forecasting. To compute the UVI at the surface, the Global Forecast Model uses a radiative transfer model (RTM) or a look-up table 125 generated by an RTM. Multiple parameters are required, such as TOC, aerosols, CF and SZA. To determine the UVI at the surface, satellite UV estimates are calculated using a combination of RTM calculations and measurements. In order to compare ground UVI measurements with the UVI product from satellites or forecast modelling, we have gathered together the multiple datasets presented in Table 2.
The Bentham spectrometer (called UVI-BENTHAM hereafter) and KippZonen Radiometer (called UVI-RADIO hereafter) 130 are the ground-based instruments part of the UV-Indien. These instruments will be compared with satellite surface UV products (TROPOMI, OMUVBG and GOME-2) and forecast model products (CAMS and TEMIS).
The TROPOMI instrument is onboard the Sentinel-5 Precursor (S5P) polar-orbiting satellite launched on 13 October 2017 as part of the EU Copernicus programme. TROPOMI surface UV radiation products include irradiances with daily integrals at four different wavelengths, and dose rates with daily doses for erythema (CIE Standard, 1998) and vitamin D synthesis (Bouillon 135 et al., 2006) action spectra. All parameters are calculated for overpass time, solar noon time, and for clear-sky conditions (no clouds, no aerosols). The TROPOMI UV algorithm (Lindfors et al., 2018) is based on two pre-computed look-up tables (LUTs): The first is used to retrieve the cloud optical depth from the measured 354 nm reflectance. This cloud optical depth, the total ozone column from the TROPOMI level 2 total ozone column product (Garane et al., 2019), the surface pressure, the surface albedo and the SZA are then used as input to the second LUT from which the UV irradiances and dose rates are 140 retrieved. A post-correction for the effect of absorbing aerosols (Arola et al., 2009) is applied to the irradiances. The ground resolution for the UV products is 7.2x3.5 km 2 (5.6x3.5 km 2 since 6 August 2019) at nadir. Only the estimates corresponding to the time of the overpass are chosen here. The UVI estimated in clear-sky conditions and in all-sky conditions will be used in this study and will both be called UVI-TROPOMI for ease of reading. UVI-TROPOMI computed for clear-sky conditions https://doi.org/10.5194/essd-2021-55 will be compared against UVI-RADIO or UVI-BENTHAM measured in clear-sky conditions and UVI-TROPOMI computed for all-sky conditions will be compared against UVI-RADIO or UVI-BENTHAM measured in clear-sky conditions. OMUVBG is a product derived from the Ozone Monitoring Instrument (Levelt et al. (2006), Tanskanen et al. (2007), Arola et al. (2009). It is based on the Total Ozone Mapping Spectrometer (TOMS) algorithm to retrieve TOC. TOC, measured with OMI, along with climatological albedo, ozone and temperature profile, elevation and SZA are used as input to an RTM to compute a first estimation of the UVI under clear skies. The measured reflectance at 360nm made by the OMI is used to estimate 150 a cloud modification factor (CMF). The CMF represents the attenuation of UV radiation by clouds and non-absorbing aerosols.
The clear-sky UVI computed previously is multiplied by the CMF to obtain an estimation of UVI in all-sky conditions (called UVI-OMUVBG hereafter). This estimate corresponds to the satellite overpass time. Absorbing aerosols are only corrected if the aerosol index is higher than 0.5 and if the measured reflectance at 360 nm is lower than 0.15 (Tanskanen et al., 2006). This could lead to overestimation for regions affected by absorbing aerosols (Tanskanen et al., 2007). UVI-OMUVBG computed for 155 clear-sky conditions will be compared against UVI-RADIO or UVI-BENTHAM measured in clear-sky conditions and UVI-OMUVBG computed for all-sky conditions will be compared against UVI-RADIO or UVI-BENTHAM measured in clear-sky conditions.
The offline surface UV is a product derived from the measurements of the GOME-2 instruments on-board the METOP-B and METOP-C satellites. The offline surface UV contains multiple variables related to UV radiation and human health: UVI, 160 integrated UVB and UVA, and daily doses derived from different biological weighting functions (erythema, DNA damage, plant damage and vitamin-D synthesis). TOC and cloud measurements provided by the AC SAF Total Ozone product and the Advanced Very High Resolution Radiometer (AVHRR-3) reflectances are used with an RTM to compute the offline surface UV product (Kujanpää and Kalakoski, 2015). The product is on a regular 0.5 x 0.5°grid and UVI is computed at local solar noon (called UVI-GOME hereafter).

165
The Integrated Forecasting System (IFS) of the Copernicus Atmosphere Monitoring Service (CAMS) has been providing UV forecasts since 2012 (Bozzo A., 2015). More precisely, it provides the spectral UV with a spectral resolution of 5 nm and the UVI by integrating the spectral UV according to the erythema action spectrum (CIE Standard, 1998). The UV irradiances and UVI are produced in clear-sky and in all-sky conditions (hereafter called UVI-CAMS). For UVI-CAMS, new forecasts are available every 12 hours: the model output has a time step of 3 hours. From the first forecast of the day, which 170 starts and is initialize at 00:00:00, we take the first, second, third and fourth timesteps of the model (00:00:00, 03:00:00:, 06:000:00 and 09:00:00 respectively) and, from the second forecast, which starts at 12:00:00, we take the first, second, third and fourth timesteps (12:00:00, 15:00:00:, 18:000:00 and 21:00:00 respectively). The horizontal resolution was approximately 80 km prior to 2016-06-21 and approximately 40 km afterwards. Following its continuous development, the CAMS model has been upgraded every year with components that significantly change the UVI forecasts. For instance, improved handling of 175 cloudiness (implemented on 2017-01-24) and a new extraterrestial UV spectrum (implemented on 2017-09-26) have resulted in significant improvements (W. Wandji, 2018). As these changes only affect the data following the upgrade, the CAMS UV forecast is not a homogeneous time series, but should be considered as an evolving UV product with which measurements can be compared. More information on the IFS and CAMS-IFS models is available at https://atmosphere.copernicus.eu/node/326.

Calibration
The UV-Indien project is deploying broadband radiometers at various sites in the Indian Ocean. In order to recalibrate these radiometers with a frequency of about 2 years, we plan to regularly reposition the radiometers on the Moufia super measurement site (University of Reunion Island -20.902°S, 55.485°E), and thus co-locate with the BENTHAM spectrometer, for a period of 195 3 or 4 months. For the recent recalibration of the UVS-E-T 15-0124, located in St-Denis, the absolute differences (AD) of the UVI, the relative differences (RD) between the UVI measured by the radiometer and the UVI measured by BENTHAM were calculated. These differences are calculated per SZA band (+-5°) and for clear-sky conditions. We noted that the differences depend on the SZA. As the SZA increases, the absolute values of the UVI tend to decrease and, therefore the absolute difference also tends to decrease. Nonetheless, the RD increases as the SZA increases. This behaviour is observed for all radiometers 200 currently compared to the Bentham (Table 3). While the UVS-E-T radiometer tends to overestimate UVI between 0°and 50°o f SZA, the SUV-E radiometer underestimates the UVI. The underestimation is greatest at high SZA. The SUV-E radiometers were installed colocally during the austral winter, when the SZA takes higher values. The differences with respect to the BENTHAM spectroradiometer were used as a reference to recalibrate the KZ Radiometers (UVI-RADIO). UVI-BENTHAM is cosine corrected, its temperature is stabilized and the instrument is regularly calibrated against standard 1000 W lamps, 205 whose origin can be traced to the National Institute of Standards and Technology (Brogniez et al., 2016). We compared both measurements again for all-sky conditions. We noted that the ADs between each radiometer and the BENTHAM were less than 0.5 . RDs were less than 5 to 10% except for the UVS-E-T 15-0124 around 25°of SZA, where the differences could reach about 11%. There was still a SZA dependency: as SZA increased, RD tended to increase, but the magnitude of this effect decreased after recalibration. Note that the radiometers installed in Mahé, Anse Quitor, and Antananarivo are still under the 210 constructor's calibration and have not yet been compared to BENTHAM. These radiometers should be recalibrated during the coming year. Table 4 presents the different radiometers and their current locations, along with the date of the next calibration.

Comparison of UV-OI Ground Based Measurements, Satellite Estimates and Modelling Estimates of UVI.
Ground-based UVI measurements from the UV-Indien Network were compared with satellite estimates and model UVI estimates. To do this, for each station we took the nearest model grid point or satellite pixel. Then, for each satellite or model 215 estimates computed at this point, we looked for the closest UVI-BENTHAM or UVI-RADIO measurement. Finally, for each couple of measurements we checked the following requirements: the time difference between the estimate and the reference must be less than 5 minutes and the SZA difference must be less than 5 degrees. For these comparisons, the UVI-BENTHAM was also scaled up and linearly interpolated at a resolution of 5 min. For data from the OMUVBG and TROPOMI product, we selected UVI estimates located within 10 km away of the corresponding ground-based station. For CAMS, we took the closest   The correlation between the BENTHAM at St-Denis and the other measurements is shown in Figure 3. The clear-sky measurements made by BENTHAM have been distinguished and are shown as blue crosses while the measurements for all-sky conditions are shown as red circles. The distribution of the data is represented on the histogram to the right and at the top of each sub-figure. All data sets are well correlated with the UVI-BENTHAM under clear sky conditions; Pearson correlation coefficients are greater than 0.9, except for OMUVBG (0.55) at Saint-Denis. Correlation results for the radiometers at St-Denis, 235 Antananarivo, Mahé and Anse Quitor stations are available in the appendix ( Figure A1, A3, A5 and A7 respectively). Correlation coefficients are greater than 0.8 at all stations except Anse Quitor (for TEMIS and OMUVBG), Mahé (for OMUVBG, GOME and TEMIS) and Saint-Denis (OMUVBG). On average, for all stations combined, OMUVBG is the product with the lowest correlation coefficients while CAMS shows the highest correlation coefficients. Mahé is the station that is least well represented station by most models, with correlation coefficients between 0.19 and 0.9.

240
The Absolute and Relative Differences between satellite product or forecast model and UVI-BENTHAM at St-Denis are shown in Figure 4 and Table  dian absolute differences decrease but a UVI overestimation can still be observed for CAMS (+0.21) and TEMIS (+0.27). The smallest median difference is obtained for GOME (-0.01). UVI-RADIO is expected to be aligned with UVI-BENTHAM since it has been recalibrated with the UVI-BENTHAM measurements, so it will not be discussed further here. The mean of the absolute differences under all-sky conditions ranges from 0.34 +-2.54 (OMUVBG) to +2.31 +-3.26 (TEMIS). Under clear-sky  This is mainly due to the spatial resolution and clouds representation. The satellite pixel or model grid point is representative of a 5.6 x 3.7 km region for the better defined satellite (TROPOMI) or a 0.5 deg x 0.5 deg region ( 55 x 55 km) (GOME, OMI and TEMIS). Thus the cloud cover considered is representative of this entire region but it is not necessarily that directly observed above the ground-based instruments. In addition, the pixels and grid points selected are not perfectly centred on the ground instruments. Finally, the four study sites are located in the tropics, present non-uniform topographic conditions and are very 260 close to the sea (with the exception of Antananarivo). These conditions favour the rapid development of clouds and complicate the estimation of cloud cover over the site by the satellite (Lakkala et al., 2020). Thus the clear-sky conditions observed on the ground may not be the same as those observed by satellites or models. This will induce discrepancy between UVI derived from satellite and modelling and UVI observed at the ground.
For the other stations (Appendix A1 to A8), all satellites and models underestimate the surface UVI (UVI-RADIO) except 265 CAMS at Antananarivo, where UVI-CAMS is just above UVI-RADIO with a mean absolute difference of +0.1. In clear-sky conditions, the MED-AD vary between -0.01 for CAMS in Antananarivo and -6.0 for GOME in Mahé. In all-sky conditions, the MED-AD vary between 0.05 for CAMS in Antananarivo and -4.5 for GOME in Mahé. In clear-sky conditions, the MED-RD vary between -0.5 % for CAMS in Antananarivo and -34.5 % for TROPOMI in Mahé. In all-sky conditions, the MED-RD vary between 15.8 % for CAMS in St-Denis and -32.7 % for TROPOMI in Mahé In clear-sky conditions, the M-AD vary 270 between 0.1 for CAMS in Antananarivo and -5.8 for GOME in Mahé. In all-sky conditions, the M-AD vary between 0.2for CAMS in Antananarivo and -3.5 for TROPOMI and GOME in Mahé. In clear-sky conditions, the M-RD vary between -10.9 % for CAMS in Antananarivo and 56.15% for TEMIS in Anse Quitor. In all-sky conditions, the M-RD vary between -25,0% for TROPOMI in Mahé and 26.4 % for TEMIS in Mahé.
Although TROPOMI has large relative differences, especially in Mahé, it is also the product with the most consistent differ- compared ranges from 72 to 253. Mahé station is also closer to the equator than the other stations (at about -4 South) and is strongly influence by the convection in this region.
Outliers and large differences could be due to numerous issues. The satellite measurements and the modelled UVI do not have spatial resolutions that can accurately represent the sky conditions just above the ground-based instruments. The satellite 285 or model grid points used in this study are either the closest grid point to the station or the average of four grid points encircling the station location. Nonetheless, for satellite estimates or model results, spatial resolution can be as high as 50km (TEMIS or GOME) or as low as 3.7km (TROPOMI). Also, cloud conditions, albedo or altitude can vary strongly over distances smaller than 100km. These differences can affect the quality of the model forecast or of the inversion algorithm used by the satellite.
Differences between a satellite product or model forecast and radiometers measurements could also be explained by a drift in 290 the radiometer calibration. The recalibration of radiometers in Antananarivo, Anse Quitor and Mahé is planned for early 2021.
Nevertheless, UVI-BENTHAM, which provides high quality data and is recalibrated every 3 months, still shows a MED-RD between -2.5% (TROPOMI) and 11.3% (CAMS), and a MED-AD between -0.2 (TROPOMI) and 0. lower throughout the year, which induces a higher UVI during the year. For clear-sky conditions, the mean UVI is higher than for all-sky conditions at all stations. In addition, the first and third quartiles are closer to the mean in clear-sky conditions. This is due to the impact of cloud attenuation on UVI variability.
It can be seen that the UVI maxima can be higher than 20 for Anse Quitor, Antananarivo and Mahé. For St-Denis, the 310 UVI maxima can reach 16. The highest UVIs are observed at the Mahé station with maxima of up to 25. At all stations, the UVI maxima are higher for all-sky conditions than clear-sky conditions. These maximas of UVI, in all-sky conditions, can be 1 to 4 higher than maxima in clear-sky conditions. This is due to the enhancement of UVI by fractional cloud cover, which can produce multiple scattering, thus enhancing the UVI on the surface. This phenomenon was previously observed at https://doi.org/10.5194/essd-2021-55 enhancement, by about 2 to 3. Antananarivo is at a higher altitude than the other three stations, at about 1.3 km asl. The city is also located inland and suffers from heavy pollution. Therefore, both aerosols and altitude can be expected to have a significant impact on the variability of surface UVIs.
In Antananarivo, the peaks of UVI mean and maxima are aligned for both clear-sky and all-sky conditions. In St-Denis the peak of mean UVI occurs later in the day for clear-sky conditions than for to all-sky conditions. In Anse Quitor, the peak of UVI is early in the day. This is due to the time zone used at Anse Quitor. Rodrigues Island, where Anse Quitor is located, shares the same time zone as St-Denis (Reunion island) or Mahé (Seychelles) but is farther east by about 8 degrees of longitude. In Mahé, the peaks of mean UVI, in all-sky and clear-sky conditions, are aligned. This is not the case for the maxima, which occur 1 to 2 hours earlier in all-sky conditions than in clear-sky conditions. The position of the UVI peaks can be explained by 325 the variability of the cloud cover as will be shown in the next section.
The average monthly climatology of the daily UVI maximum and the daily UVI at solar noon were also calculated. Although the datasets do not cover a sufficient period to be climatologically significant, the annual variability and the differences between UVI and UVI daily maxima at solar noon present interesting results. For all stations and all months of the year, the maximum daily UVI (UVI DMAX ) is higher than the average UVI at solar noon (UVI SNOON ). These climatologies are based on data sets for all types of cloud cover. Thus, the cloud cover will introduce a bias in the result. UVI DMAX include maxima of UVI during cloud enhancements, which could occur earlier or later than the local 335 solar noon, as discussed in the previous result ( Figure 5). UVI DMAX also include maxima of UVI during clear-sky days, which will usually occurs at local solar noon. UVI SNOON will include only UVI at local solar noon with or without cloud cover. This is why UVI DMAX is always higher than UVI SNOON in all-sky conditions. The differences range between 1 during austral winter at St-Denis and 6 in austral summer at Mahé. The corresponding SZA differences ranges from 0 to 6°.
Looking at the few complete clear-sky days revealed a UVI SNOON and a UVI DMAX that were almost equivalent. However, as 340 there were not enough clear sky measurements per month and per SZA bin, these results are not represented here.

Cloud measurements
Early results from the camera are able to describe the mean diurnal cloud cycles over each station. Figure 7a shows the mean diurnal cycle of the cloud fraction (CF) over Antananarivo for the entire year (black curve), along with the distribution of the first and third quartiles (blue shaded area). Seasonal means of CF are also presented for December, January and February 7b. Figure 7c represents the difference of UVI maxima between clear-sky UVI and all-sky UVI for the same period. The UVI distribution is also available in Figure 6d higher than the annual mean while JJA shows CF lower than the annual mean. The CF diurnal cycle alone is probably not the only factor involved in triggering UVI enhancement: cloud distributions and type of clouds probably play a role in UVI enhancement (Sabburg and Wong, 2000;Calbó et al., 2005).  (Figure 8b and 8d). A glance at the difference between the maxima of UVI in clear-sky conditions and the maxima of UVI in all-sky conditions (Figure 7c), shows that almost all the highest differences are observed for the DJF season. These differences can reach -2.4. The JJA season also shows large differences during the morning but, after 10 am, the differences are lower than in DJF season most of the time. 10 am is also the hour when the CF starts to rise more rapidly in JJA (Figure 7a). This result could indicate than UVI enhancement 370 occurs less frequently above a certain CF threshold.
The cameras of the UV-Indien network present promising results for studying cloud variability and its impact on UV radiation in the Indian Ocean region. CF alone is probably not sufficient to give an understanding UVI enhancement. Cloud types, types of cloud distribution and solar occultation by clouds also need to be considered. Moreover, the UV Indien Network is very young, so there is not yet enough data to conduct significant climatological studies. The TROPOMI surface UV radiation product used in this study is available at https://nsdc.fmi.fi/data/data_s5puv.php The AC SAF (GOME/METOP) data can be downloaded through the website at https://acsaf.org The TEMIS data can be downloaded through the website at http://www.temis.nl/uvradiation/UVindex.html The OMUVBG data can be downloaded through the website at https://disc.gsfc.nasa.gov/datasets/OMUVBG_003 The CAMS data can be downloaded through the website at https://apps.ecmwf.int 390

Conclusions
The ground-based spectroradiometer and radiometer of the UV-Indien network were compared with satellites estimates and the estimates of the UVI model. The correlation coefficient between the satellite or model estimates and the ground-based measurements was greater than 0.9 at all stations except Mahé and for all datasets except OMUVBG. In all-sky conditions, 395 the largest UVI median absolute difference between the satellite or model estimates and the ground-based instruments was -0.4 (TROPOMI), -1.4(TROPOMI), -2.3(GOME) and -6.0(GOME) at St-Denis, Antananarivo (), Anse Quitor and Mahé respectively. In clear-sky conditions, the largest UVI median absolute difference between the satellites or model estimates and the ground-based instruments was 1.1(TEMIS), -1.3(TROPOMI), -1.7(GOME) and -4.6 (GOME) at St-Denis, Antananarivo, Anse Quitor and Mahé respectively. In all-sky conditions, the smallest UVI median difference between the satellites or model  At St-Denis, satellite and model estimates were usually found to overestimate UVI compared to ground-based instruments, while at Antananarivo, Anse Qutor and Mahé, satellite and model were found to underestimate ground-based measurements of UVI.
The largest discrepancies were observed at the Mahé station. Many reasons can be evoked to explain these differences. The satellite or model resolution may not be able to accurately determine the sky conditions above the station. As Mahé is under the influence of the sea-breeze, the formation of clouds is frequent late in the morning. From the ground, these clouds could be seen to induce irradiance enhancements due to multiple scattering at their edges. From the satellite's point of view, only strong backscattering would be observed and an attenuation of the clouds would therefore be applied to the UV product. This is also consistent with the time of the satellite estimates (solar-noon or overpass), which is late in the morning when cloud formation is frequent which would induce backscattering. UVI enhancement occurs also frequently during this time ( Figure 5). This two 415 phenomenons could explain both the observed high UVI and the discrepancy with satellite and model estimates. A drift in the radiometer calibration could also explain part of the difference. Mahé also presents a low number of clear-sky comparison points since the K&Z radiometer temporal resolution is 5 min and there are only 16% of clear-sky days. St-Denis and Anse Quitor are mountainous islands also under the influence of trade winds, where there is also frequent cloud formation in the late morning. A fractional cloud sky cover could induce UVI enhancements measured by the ground-based instruments while 420 satellites and model could apply attenuation of the measured UVI instead. These two stations are located at about 20 degrees south, while Mahé is located at about 5 degrees south. Therefore, the maxima of UVI are higher at Mahé and the resulting enhancement of UVI should also be stronger. To investigate the impact of clouds on UVI, all-sky cameras have recently been deployed at each station of the UV Indien network. When a significant number of data have been be acquired, another study will address the subject of clouds and UV in the tropics. 425 We also analysed the UVI variability at these different sites. Mean UVI at local solar noon ranges between 10 (Antananarivo, Anse Quitor, St-Denis) and about 14 (Mahé, Seychelles). While the mean UVI in clear-sky conditions are higher than mean UVI in all-sky conditions, the UVI maxima in all-sky conditions are higher than in clear-sky conditions. This phenomenon of UVI enhancement by clouds can be observed at the four stations studied here. Measurements of the diurnal cloud fraction cycle have also been studied, enabling different daily profiles to be distinguished depending on the station and the season.

430
However, these data are still recent and there are not enough measurements available at present to give a significant statistical description. The highest UVI enhancement is observed at Mahé where there is a 4 to 5 UVI unit difference between maxima of UVI in all-sky conditions compared to clear-sky conditions. When the number of cloud fraction measurements is significant throughout the year and the time period covered by the camera is longer, it will be possible to combine Cloud Fraction and UVI data to better understand and quantify the UVI enhancement by clouds. The UVI data at the other 4 sites of UV Indien Network