The comparison between the WVMR ground-based lidar measurements and the IASI operational products took place in the frame of WaLiNeAs between October 2022 and January 2023. The WVMR vertical profiles derived from the IASI spaceborne spectrometer were compared with those measured by eight Raman lidars located along the western Mediterranean coast of Europe as shown in Fig. 1. Table 1 shows the geographical location of each site and their identification number used throughout the article. The main characteristics of the eight Raman lidars are summarised in Appendix A alongside the calibration process for each Raman lidar.

The technology used for each lidar is similar, as is the data inversion procedure. Useful references are given in Table 1. The maximum uncertainty lies with the contributions of bias and standard deviation. The bias is mainly due to the lidar calibration, aerosol correction and temperature dependency of the backscatter cross section of the H₂O Raman channel (Whiteman, 2003), whereas the standard deviation is due to shot noise. For this study, we will only consider measurements with a statistical error of less than 0.4 g kg⁻¹. In the majority of cases, biases remain less than 5 %. As explained in Appendix B, they are strongly dependent on Raman lidar technology. Note that the Raman lidar of Barcelona and Ajaccio use a rotational channel on dinitrogen and dioxygen (N₂ + O₂) whereas the others lidar use a roto–vibrational channel on N₂.

Autumn 2022 was characterised by unusual weather conditions in the western Mediterranean basin, including persistent positive temperature anomalies and widespread rainfall deficits. This followed an exceptionally hot summer, the effects of which continued into autumn through enhanced atmosphere–ocean interactions. Indeed, temperatures were well above seasonal values (by ∼ 3 °C), particularly in September and October. Significant positive temperature anomalies were observed over southern Europe and the western Mediterranean, reflecting the persistence of warm air masses and atmospheric blocking patterns. These conditions were reinforced by exceptionally high sea surface temperatures of the western Mediterranean Sea, resulting from heat accumulation during the summer of 2022. Compared to climatology, anomalies of several degrees (∼ 4 °C) were observed (Guinaldo et al., 2023).

Despite this thermodynamic context being favourable to instability, atmospheric circulation on the synoptic scale was dominated by recurrent anticyclonic conditions over Western Europe and the Mediterranean basin. This limited the intrusion of low-pressure systems and the establishment of unstable southerly flows, reducing the frequency and intensity of the Mediterranean weather systems typically observed in autumn (Blanchet et al., 2021). Consequently, autumn total precipitation amounts were significantly below normal by ∼ 24 %.

Despite the lower troposphere being drier than normal, two contrasting periods were observed. These periods are evident in the temporal syntheses derived from IASI measurements, as illustrated in Fig. 2, which depicts the precipitable water content over the western Mediterranean basin. Figure 2a shows the wetter period, while Fig. 2b shows the drier period after 5 November 2022. The temporal evolution of lidar, spaceborne and modelling data in the Supplement will highlight this. Precipitable water content is homogeneous across the western basin, but there is a significant variation between the two periods, with a ∼ 30 % decrease between the beginning of November and the rest of the year.

The lidars were calibrated individually using radiosondes and/or ground-based weather stations using similar approaches. This raises the question of how to standardize these different calibration procedures. To assess this, we compared each lidar's data over its entire operating period with reanalyses of ERA5. ERA5 is constrained by multiple sources of vertical profiles, including radiosondes, the IAGOS (In-Service Aircraft for a Global Observing System) network and IASI observations above 2–3 km (Hersbach et al., 2020). For this reason, ERA5 most closely mimics the hourly radiosonde observations that would have been available near the various lidar stations of the WaLiNeAs network. As an example, we present the comparison between all radiosondes and ERA5 for the period November–December 2022 at the meteorological station of Nîmes (Météo-France), located 40 km north of Grau-du-Roi, in the Supplement (Fig. S1). The relationship is linear with a slope of 0.99 and an intercept of 0.054 g kg⁻¹, a strong correlation of 0.99 and a low mean bias of +0.02 g kg⁻¹. The standard deviation is around 0.36 g kg⁻¹. Furthermore, as shown in Laly and Chazette (2025), for the sites of the Grau-du-Roi (RL4) and Cannes (RL7), the ERA5 pixels around a central pixel are strongly correlated (COR > 0.97), and exhibit MBs with an absolute value below 0.14 g kg⁻¹.

ERA5 data may nevertheless be biased. Biases in ERA5 arise from various sources: (i) inadequate accounting for instrumental errors during the data assimilation process; (ii) assumptions involved in the simulation of satellite radiances; (iii) deficiencies in the parameterizations of convection, evaporation, and turbulent exchanges; (iv) limited spatial resolution; (v) imperfections in the representation of the planetary boundary layer (PBL); and (vi) uncertainties in the representation of clouds and precipitation. Unlike ERA5 data, after correction for the baseline and sky radiance, the ratio between the water vapor and dinitrogen Raman channels tends toward zero in the absence of atmospheric water vapor. For this reason, the lidar calibration relies solely on the slope derived from the comparison between the WVMRs retrieved from the Raman lidar and those from ERA5. The intercept is therefore assumed to arise exclusively from biases in the reanalysis data. Note that, in the absence of atmospheric water vapor, no Raman backscattering is produced by water molecules, and the signal recorded on the H₂O–Raman channel corresponds solely to the background noise associated with the acquisition baseline and sky radiance. These two contributions are subtracted from the signal for each lidar used in this study.

For all lidars, we therefore sought a calibration yielding a slope of 1 in the comparison with ERA5. This required applying a site-specific correction factor. As shown in Table 2, the correction factors are lower than 10 %, ranging from 0.93 to 1.1 with values mostly close to 1. This attests to the quality of the original calibrations carried out independently at each station. More or less significant biases between lidar and ERA5 remain, as shown in Figs. S2 to S9. In some cases, these may be associated with imperfect calculation for the temperature dependency of the N₂-Raman channel, molecular and particulate contributions in the lidar data. All these biases contribute to an apparent moistening of the water vapor profile and are cumulative.

The same process has been used to correct all lidar signals for molecular transmission, and the residual error is below 0.5 %. The contributions of aerosols and temperature to the bias are discussed in Appendix B. The contribution of aerosols remains small, either because aerosol optical thickness (AOTs) values are mainly below 0.1, or because the aerosols are primarily desert dust with Ångström exponents below 0.5. Note that RL2 and RL8 use a rotational Raman channel for oxygen and nitrogen in addition to the vibro-rotational Raman channel for water vapour. Because the wavelength difference is larger in this configuration, the effect of aerosols is more pronounced. Failure to correct for aerosols leads to an overestimation of the WVMR. As shown in Appendix B, for a typical AOT of 0.1 (at 355 nm) and an Ångström exponent of 1.5 (corresponding to pollution aerosols), the overestimation could reach approximately +1 % for the use of N₂-channel and +2 % for the use of N₂ + O₂-channel and decreases progressively with decreasing distance to the lidar. It is less than +1 % (between +1.3 % and +5 %) for dust aerosol mixed and above the marine boundary layer for the altitudes lower than 1.5 km a.m.s.l. (above 1.5 km a.m.s.l.). In this study, temperature variations with altitude lead to a maximum bias of less than +2.7 %.

Another potential source of bias is imperfect correction of the overlap factor in the lower layers. The minimum range of WVMR data is ensured to be below 200 m (see Table 1) because the N₂– and H₂O–Raman channels share the same optical path. The remaining divergences are mainly related to the near range due to the angular acceptance of the interference filters. Overall, biases related to the instrument itself and to the data inversion procedure remain below or on the order of ±1 % for all lidars. This justifies performing only a slope correction with respect to ERA5, given the good coefficients of determination obtained (between 0.71 and 0.91).

Note that the intercept values reported in Table 2, representing systematic biases, are the result of multiple contributions that can partially cancel each other. Nevertheless, biases with an absolute value greater than 0.1 g kg⁻¹ are considered more likely attributable to modelling rather than Raman lidar measurements.

Comparisons between LIDAR measurements and IASI overpasses were performed by considering the nine pixels surrounding each site (Fig. 3), corresponding to an area of 75 km × 75 km. Due to variations in horizontal wind speed, the 9 pixels do not provide the same level of representation at different altitudes. During the period under consideration, wind speeds were below 20 m s⁻¹, ensuring that observations above the sites were accurately captured by at least one of the pixels. Furthermore, variability in vertical humidity profiles across these IASI pixels was low, averaging less than 0.2 g kg⁻¹. Additionally, as the stations are situated close to the coast, at least 4 of these pixels consistently provide data on the lower troposphere. The lidar data were averaged over a 1 h period, using profiles within ±30 min of the satellite overpass time. The signal-to-noise ratio is significantly improved by this averaging, with uncertainties of less than 0.4 g kg⁻¹ being achieved in the low troposphere. The number of IASI pixels contributing to the comparison varies with altitude, as the surrounding topography must be taken into account. Within the perimeter defined by the 9 pixels, the local summits generally remain below 500 m a.m.s.l., ensuring minimal topographic effects in the low troposphere including the PBL.

The statistical indicators used to evaluate the consistency of the IASI-derived WVMR (riasi) with respect to lidar measurements (rlidar) are the mean bias (MB), the root mean square difference (RMSD), the root mean square error (RMSE) and the (Pearson) correlation (COR). These metrics are commonly used to assess model performance (Boylan and Russell, 2006; Tombette et al., 2008) and given for each altitude z by: 1MB(z)=rlidart,z-riasit,z2RMSD(z)=rlidart,z-riasit,z23RMSE(z)=RMSD(z)2-MB(z)24COR(z)=rlidart,z-rlidart,z⋅riasit,z-riasit,zrlidart,z-rlidart,z2⋅riasit,z-riasit,z2 where is the temporal averaging. The altitude averaged MB (MB̃) and RMSE (RMSẼ) are then derived using the relationships: 5MB̃=1M⋅∑i=1MMBi(z)RMSẼ=1M⋅∑i=1MRMSDi(z)2-MB̃2 were M is the number of lidar samplings i along the altitude.

The CORs from different altitudes are combined using the Fisher F transformation (Fouladi and Steiger, 2008). This gives the following equation for each altitude z: 6F(z)=12⋅log⁡1+COR(z)1-COR(z) which leads to the COR averaged on the altitudes (COR̃) expressed by: 7COR̃=e2⋅F‾-1e2⋅F‾+1 with F‾ = ∫zazbN(z)-3⋅F(z)⋅dz∫zazbN(z)-3⋅dz, where N(z) represents the number of coincidences between Raman lidar and IASI at an altitude z, within the altitude range [za, zb]. N(z)-3 is an approximation used when the number of samples is less than or around 50 (Stuart and Ord, 1991). Note that the same approach is applied when lidar and ERA5 are compared.

Table 3 summarises the statistical analysis for: (i) the PBL (0.2–1.5 km a.m.s.l.), (ii) the transition layer to the free troposphere (1.5–3 km a.m.s.l.), (iii) the free troposphere (3–5 km a.m.s.l.), and (iv) all three layers combined. It is worth noting that the altitude of the top of the PBL varies over time. The first two layers in Table 3 may therefore include air-mass recirculation associated with the sea-land breeze cycle and circulation patterns that are very specific to the measurement site.

A synthesis of the behaviour of all lidar systems has been obtained by combining statistical profiles obtained from individual lidar systems (Fig. 4). The resulting synthesis across all lidar stations shows that the vertical profiles of the three statistical parameters exhibit relatively smooth continuity between adjacent layers, highlighting that the variability of the PBL height has on average a moderate influence. This is more noticeable at sites expose to strong winds, such as the Rhône delta and Barcelona. At these sites, a decrease in correlation is observed at the interface between the PBL and the free troposphere.

In general, there is a significant correlation between lidar and IASI measurements, mainly ranging from 0.7 to 0.8 (Fig. 4a). This strong correlation is also evident between lidar measurements and ERA5 reanalyses. However, at the Barcelona (RL2) and Coursan (RL3) sites weaker correlations are observed (between ∼ 0.5 and 0.7). For Coursan, the reduced correlations are less significant, given the much lower number of coincidences and the reduced range of the lidar, mainly due to frequent cloud cover. The slightly lower correlations values determined for Barcelona (RL2) can be possibly explained by the location of this lidar station. It is located in an urban environment at the foothill of a 500 m-high mountain range, and 5 km from the sea. Pérez et al. (2004) have shown that, in such conditions, the atmospheric vertical structures are complex, making comparisons with reference profiles (e.g., radiosondes) problematic. In most cases, the correlations between lidar and ERA5 are slightly higher than those between lidars and IASI for all layers. This may be attributed both to the assimilation of additional observational datasets in ERA5, beyond IASI, and to the lower sensitivity of IASI within the PBL. Nevertheless, WVMR profiles derived from IASI properly represent atmospheric variability occurring from one satellite overpass to another.

The RMSEs values are fairly consistent when comparing the different lidars and remain mostly between 1 and 2 g kg⁻¹, regardless of the layer considered (Table 3). It should be remembered that only lidar data with a statistical error of less than 0.4 g kg⁻¹ were considered in this study. The RMSEs shown in both Table 3 and Fig. 4b are significantly higher and reflect the statistical variability between the lidar and IASI or ERA5 data. These values provide a clear indication of the uncertainty in the vertical WVMR profiles derived from IASI or ERA5 in the first 5 km of the atmosphere, where most of the water vapour is present. Note that differences between the single lidar and IASI or ERA5 profiles reverberate more on RMSE, which make use of absolute differences (root-mean square), than on bias, which accounts for the sign of the differences, most part of those cancelling out. One possible source of the high RMSE values may be the displacement of air masses between IASI or ERA5 pixels, resulting in spatial heterogeneity between pixels.

Interpreting biases is more difficult because of their diverse origins, which are not always identifiable. As explained in Appendix A and B, biases related to lidar measurements are mainly associated with (i) the reference used for calibration, (ii) the differential transmission due to aerosols, and (iii) the temperature dependence of the Raman backscatter cross section (Appendix B). The lidars were individually calibrated using in situ measurements (e.g. weather stations or radiosondes) and subsequently recalibrated by adjusting the calibration slope using coincident ERA5 data for each site (see Sect. 3.1). The biases relative to IASI and ERA5 are presented in Table 3 after accounting for the biases related to temperature and aerosols affecting the lidar products. The corresponding vertical profiles are provided in the Supplement (Figs. S18b to S25b) and in Fig. 4b; however, the contributions of aerosols and temperature to the bias are not considered in these figures. In the synthesis, the bias values relative to IASI are higher than those relative to ERA5 for all altitude ranges. The difference ranges from 0.19 to 0.26 g kg⁻¹. For the individual sites, the vast majority of biases (lidar–IASI or lidar–ERA5) are positive, indicating a systematic underestimation of the water vapour content by IASI and to a lesser extent by ERA5.

Few comparative studies in the literature exist between IASI and lidar measurements, and the same is true for comparisons between ERA5 reanalyses and lidar measurements. Following the ChArMEx and HyMeX programmes, Chazette et al. (2014a) compared WVMR measurements derived from the WALI lidar, which was located in the Balearic island of Menorca (Spain), with IASI measurements following nighttime overpasses by METOP-A. The RMSE calculated for September–October 2012 was ∼ 1.06 g kg⁻¹ between 0 and 5 km a.m.s.l., which is consistent with the results presented in Table 3. The MB was around 0.5 g kg⁻¹, slightly higher than the value found in this study, while the COR (∼ 0.66) was also lower. A key difference lies in the fact that the lidar measurements from Menorca were acquired over open sea and included heavy precipitation events (HPEs), whereas the WaLiNeAs measurements were taken at coastal sites and were not affected by HPEs. A mutual bias value of 0.26 g kg⁻¹ was reported by De Rosa et al. (2020) when comparing the Raman lidar BASIL and IASI in the altitude interval 2–15 km (RMSE = 0.30 g kg⁻¹). More recently, Laly and Chazette (2025) investigated the potential impact of atmospheric water vapour Raman lidar measurements by comparing them with WVMR vertical profiles obtained from ERA5 data. This work was also carried out as part of the WaLiNeAs project and included the Grau-du-Roi (RL4) and Cannes (RL7) sites during the dry period. In the first 5 km of the atmosphere, they found an RMSE of 1.05 g kg⁻¹ for the Cannes (RL7) site, which is comparable to the results in Table 3 (0.97 g kg⁻¹). For this site, the MB also remained fairly similar, with a minimal value of ∼ +0.4 g kg⁻¹ compared to +0.25 g kg⁻¹ after accounting for aerosol and temperature contributions. The COR correlations are also similar (0.89 compared to 0.86). At the Grau-du-Roi site (RL4), the RMSE was slightly higher, at 1.57 g kg⁻¹ compared to 1.02 g kg⁻¹ in this study. However, the MB and COR were comparable, with values being higher and on the order of +1 g kg⁻¹ and 0.86, respectively.

Based on the study presented here, larger dispersion of results might have been expected for the two stations (RL4 and RL5) located in the Rhône Valley. In the Rhône Valley, there is often a process of recirculation, whereby the Mistral wind blows in altitude and the sea breeze blows lower down. The latter can penetrate more than 100 km inland. In addition, the lower layers in the Camargue can be very humid due to the abundance of liquid water areas. There is also a steep humidity gradient between the Mediterranean coast and inland areas, associated with various local phenomena linked to the Rhône delta's specific topography. The statistical results presented here show that the temporal variability of WVMR over the first 5 km is accurately represented by both IASI measurements and ERA5 reanalyses.

As previously explained, during the WaLiNeAs field campaign, there was a wetter period in October, followed by an exceptionally dry period from early November onwards. Most of the measurements were taken during the drier period. The wetter period was less sampled, except in Barcelona (RL2), Toulon (RL6) and Ajaccio (RL8). At the Barcelona station (Fig. S11), more than two-thirds of the measurements were taken during the wetter period, which may be subjected to greater variability than the drier period. This may also explain the slightly weaker correlations observed for this station (Table 3 and Fig. S19a) in addition to its specific location. Note that it is more difficult to draw conclusions about the samples in the free troposphere, as there are fewer of them during the wetter period when there was heavy cloud cover.

A synthesis was therefore carried out between the Toulon and Ajaccio sites in order to potentially highlight the differences between the wet and dry periods. The results are presented in Fig. 5 and Table 4. During the wet period, there was approximately 30 % more water vapour in the lower troposphere. The statistical parameters are comparable for both periods, with an overall IASI bias of 0.25 g kg⁻¹ (0.18 g kg⁻¹) for the wet (dry) period. In the case of ERA5, this bias is very small (≲ 0.02 g kg⁻¹). The RMSE values remain significantly high, exceeding 1 g kg⁻¹, while the COR values (∼ 0.7) indicate a satisfactory representation of the temporal variability of the WVMR vertical profiles for both the wet and dry periods.

It is interesting to note that METOP/IASI overpasses the western Mediterranean twice a day, at 08:00–10:00 and 19:00–21:00 local time. The second time slot corresponds to early nighttime conditions during the field campaign period. However, it is still early enough in the night for the PBL height to change significantly, particularly near the coast. Therefore, no notable differences were observed between the two time slots, apart from a decrease in samples and statistical representativeness when the two slots are split.