The mean and standard deviation of IWV differences (ERA-Interim minus GPS) for all 120 stations over the 16-year period are shown in Figs. 2 to 5. Figure 2 shows the results as a function of station latitude. The general tendency is depicted by the fitted polynomials (the outlying stations, defined beyond the dotted red lines will be discussed in Sect. 4). The different plots show a clear dependence of the results on latitude. The mean difference (Fig. 2a, c) is positive at northern and southern extratropical latitudes (30–80^∘ N and 30–60^∘ S), while it is negative in the intertropical band (30^∘ S–30^∘ N). This result is consistent with the results of Schröder et al. (2016), who compared ERA-Interim to satellite data. The alternation of positive and negative differences is most likely due to biases in the ERA-Interim reanalysis reflecting the difference in moisture information entering the reanalysis over ocean (mainly microwave satellite data) and land (mainly radiosonde and infrared satellite data) (Dee et al., 2011). Indeed, the tropical GPS stations used here are mostly representative of oceanic areas, while the extratropical GPS stations are mainly continental. Similar biases in ERA-Interim were also highlighted by Trenberth et al. (2011), and Parracho et al. (2018), in comparison to other atmospheric reanalyses. The biases remain small, however (below ±1 kg m⁻² or ±10 %). The absolute standard deviation of IWV differences (Fig. 2b) also shows a latitudinal variation with two peaks, around 30^∘ S and 30^∘ N, and dips around the Equator and towards the poles. The equatorial dip is more marked in the relative standard deviation plot (Fig. 2d) because the mean IWV is the largest at these altitudes (∼40 kg m⁻²; see the dashed blue line in Fig. 2c). The enhanced discrepancy between ERA-Interim and GPS daily IWV estimates in the subtropics coincide quite well with the highest day-to-day variability in both hemispheres (see the superposed blue lines in Fig. 2b, d). This strong day-to-day variability is mainly due to the moisture transport associated with the extratropical cyclones in the Northern and Southern Hemisphere storm tracks (Chang et al., 2002; Pfahl et al., 2014). It is not uncommon to observe IWV variations exceeding 20 kg m⁻² d⁻¹ at GPS sites located in the storm track (Bock et al., 2005, 2016). Increased discrepancy between ERA-Interim and GPS at those sites can be due to the imperfect spatiotemporal location of such large moisture variations in the reanalysis or to a representativeness difference between the GPS observations and the reanalysis. No systematic increase in GPS formal error was found in these situations; i.e. the discrepancy is not due to GPS errors.

Figure 2(a, c) Mean and (b, d) standard deviation of daily IWV difference (ERAI minus GNSS) for 120 global stations as a function of station latitude. Panels (a) and (b) are in kg m⁻²; panels (c) and (d) are in percent of GNSS IWV. The dashed black lines show polynomial fits of orders 5 and 9 for the mean difference and the standard deviation, respectively. The dashed blue lines show polynomial fits of order 7 for (b) the standard deviation of dIWV∕dt (kg m⁻² d⁻¹); (c) the mean IWV (kg m⁻²); (d) the relative standard deviation of dIWV∕dt (% d⁻¹) computed from GPS IWV data. The dotted red lines show the range-check limits used to detect outlying sites (named stations).

Download

Figure 3Similar to Fig. 2 but plotted as a function of GPS station altitude. The dashed black lines show linear fits.

Download

Figure 3 shows the mean and standard deviation of IWV differences as a function of altitude of the GPS stations. The mean differences (Fig. 3a, c) show no dependence on altitude, meaning that the method of computation of GPS IWV (from ERA-Interim P_s and T_m estimates) and ERA-Interim IWV (from pressure levels) is highly consistent throughout a large altitude range. The standard deviation (Fig. 3b) shows no dependence on altitude either but the relative standard deviation (Fig. 3d) does. The fitted straight line in Fig. 3d shows that this statistic is increasing quite fast as a function of altitude. This tendency can be explained by larger representativeness errors in the reanalysis humidity field as a function of altitude as also found by Waller et al. (2014) in the Met Office high-resolution UK variable resolution model.

Figure 4 shows the standard deviation of IWV differences, σ_Δ, as a function of a few other parameters which give further insight into possible reasons for the discrepancy between GPS and ERA-Interim. Figure 4a and b indicate that, apart from the outliers, there is a moderate tendency for increased discrepancy with increased mean IWV (i.e. warmer and moister climate) and increased IWV variability (including the seasonal variations). The slope of the tendency is actually steeper at the lower IWV bound (mean IWV < 25 kg m⁻²) corresponding to midlatitude and high-latitude sites, while it vanishes at the upper bound, corresponding to intertropical sites (mean IWV ≥ 25 kg m⁻²). The standard deviation of IWV differences reaches a nearly constant level of σ_Δ≈2 kg m⁻² throughout the Equator and the intertropical band. Figure 4c shows that there is a strong tendency for increased discrepancy with increased spatiotemporal variability around the GPS site measured by μ_R (see Sect. 2.3). This interrelation is actually the strongest among all the tested relations between σ_Δ and other statistics. It indicates that representativeness differences are a major source of discrepancy between GPS and ERA-Interim IWV estimates. Finally, Fig. 4d shows that there is only a small tendency for increased discrepancy with increased GPS formal errors.

Figure 4Standard deviation of daily IWV difference (ERAI minus GNSS) for 120 global stations, as a function of (a) mean GPS IWV; (b) standard deviation of GPS IWV; (c) mean spatial variability of ERAI IWV from the four grid points surrounding the GPS sites used as representativeness statistics (see text); (d) formal error of GPS IWV estimates. Only three outlying stations are named on these plots for clarity.

Download

Figure 5 shows that time averaging is a means of reducing the representativeness differences, as smaller-scale local features captured by the GPS point observations get smoothed out. The mean differences (Fig. 5a, c) are not impacted by the averaging, as expected. The standard deviation of differences (Fig. 5b, d), on the other hand, decreases for the monthly averages, both in absolute and relative units, at all sites. The median standard deviation of the daily IWV differences (ERA-Interim minus GPS) is 1.2 kg m⁻², while the value for the monthly series is 0.51 kg m⁻². The reduction of standard deviation due to averaging is 2.35, which is smaller than the value of $\sqrt{\mathrm{30}}=\mathrm{5.48}$ that one would expect with independent normally distributed data (when averaging over a mean month of 30 d). This inconsistency can be due to the serial correlation in the IWV differences revealing a dependence of the IWV differences upon the meteorological situation. This point might be further investigated by, e.g. separating the IWV differences in different weather regimes. Another means of reducing the discrepancy due to representativeness differences is to use a reanalysis with higher spatial resolution and improved physics representing the smaller-scale atmospheric processes. We compared, for instance, daily GPS IWV data to the Applications of Research to Operations at Mesoscale (AROME) West-Mediterranean operational analysis of Météo-France (this model has a horizontal resolution of 2.5 km × 2.5 km) and found a median standard deviation of difference of 0.81 kg m⁻² over a period of 2 months; we used the GPS and AROME data from the Hydrological cycle in the Mediterranean experiment (HyMeX) Special Observing Period, 5 September–6 November 2012, described in Bock et al. (2016). The median standard deviation of GPS-ERAI differences over 12 stations in the same domain amounts to 0.98 kg m⁻², so there is clear benefit of higher resolution and more modern physics.

Figure 5Mean vs. standard deviation of IWV difference (ERAI minus GNSS) for (a, b) daily values and (c, d) monthly values. The median values of mean and standard deviation over all 120 stations are 0.47 and 1.2 kg m⁻² (3.1 % and 8.3 %) for the daily results, and 0.47 and 0.51 kg m⁻² (3.1 % and 3.8 %) for the monthly results, respectively. The dotted red lines show the range-check limits used to detect outlying sites (named stations) in the case of the daily data.

Download

Since representativeness differences impose a strong limitation on the agreement between GPS and reanalysis, one may wonder if the horizontal interpolation from the four surrounding ERA-Interim grid points does not further enhance the differences by mixing information from the different grid points. We investigated this question by computing the statistics for each of the four surrounding grid points. Figure 6 shows the results in comparison to the results obtained with the bilinearly interpolated IWV values. The comparison of the mean values (Fig. 6a and b) emphasizes large variations in the biases at some stations which will be further discussed in Sect. 4. The slight shift of the ensemble of results below the 1:1 line is reflecting the fact that a majority of sites exhibit small positive bias (0.47 kg m⁻² on average) as already noticed in Fig. 2a, c, which is not due to representativeness differences but rather to the type of assimilated data (see Sect. 3). The comparison of standard deviations (Fig. 6c and d) shows unambiguously that, at almost all sites, the results for the bilinearly interpolated IWV values are better than for any one of the four surrounding grid points (almost all results sit above the 1:1 line). This conclusion holds for 112 out of 120 stations for the absolute standard deviation (Fig. 6c) and 111 out of 120 stations for the relative standard deviation (Fig. 6d). It indicates that the temporal variability represented by the bilinearly interpolated ERA-Interim IWV data matches best the temporal variability observed by the GPS (i.e. better than from the nearest grid point in the horizontal or in the vertical dimension). When monthly IWV data are compared (not shown), the conclusions are similar, though the number of sites of improved results drops to 71 out of 120 (both for absolute and relative standard deviations). The drop confirms again that the representativeness differences can be reduced by the temporal averaging.

Figure 6Scatter plots of (a, b) mean and (c, d) standard deviation of daily IWV difference (ERAI minus GPS) when ERAI IWV is bilinearly interpolated from four surrounding grid points (x axis) versus the spread of the mean (a, b) or standard deviations (c, d) for the four surrounding grid points (y axis). The spread is plotted as vertical error bars from the minimum to maximum values. In panels (c) and (d), vertical bars extending below the 1:1 line indicate sites where at least one of the surrounding grid points is in better agreement with GPS than the bilinearly interpolated values; the corresponding stations are named and indicated by a black dot. The dotted red lines show the range check limits as in previous figures. In panels (a) and (b), some of the sites with statistics outside the limits indicated by the dotted red lines are named as well.

Download

In the previous section, we have seen that the general agreement between GPS and ERA-Interim is limited by representativeness differences which are enhanced in regions of strong temporal variability (Fig. 2b, d), at higher altitude (where mainly the relative standard deviation of differences is impacted; Fig. 3d), and at sites where the mean spatial variability at the four surrounding ERA-Interim grid points is large (Fig. 4c). The standard deviation of differences, σ_Δ, is actually well predicted by our representativeness error statistic, μ_R, with a linear correlation coefficient of r(σ_Δμ_R)=0.73. This strong correlation suggests that the outlying sites, i.e. sites with the largest discrepancy, may have enhanced representativeness errors (Fig. 4c). To investigate this idea, we will analyse in more detail these sites here. First, let us define range limits for each of the four statistics of differences to separate the acceptable sites (i.e. those satisfying the following conditions) from the outliers:

The values of the limits were determined from visual inspection of Figs. 2, 3, and 5, and shown as the dotted red lines in these figures. The choice for the limits is subjective because we think that the results are very unpredictable when analysing a global network (due to the variety of climates, equipment, and reanalysis performance). So it is necessary to visually inspect the results and determine the limits beyond which the results do not look “normal”. We believe this approach is quite robust thanks to the combination of several representations of the results such as those shown in Figs. 2–6 (i.e. function of latitude, altitude, mean vs. SD scatter plots, etc.). This methodology can be safely applied to other datasets. The reason why we chose non-symmetric limits with respect to zero for the mean differences is because the distribution is not centred on zero. The result is a detection of 15 outlying sites, some of which exceed the limits in more than one test: three sites have excessive absolute bias (CFAG, KIT3, and SANT); nine sites have excessive relative bias (CFAG, COSO, DAV1, KIT3, MAW1, MCM4, POL2, SANT, and SYOG); eight sites have excessive standard deviation of differences (AREQ, BLYT, CFAG, DHLG, IISC, KIT3, LONG, and SANT); and nine sites have excessive relative standard deviation of differences (AREQ, CFAG, KIT3, MAW1, MCM4, MKEA, POL2, SANT, and SYOG). Three sites have statistics exceeding the limits in all four tests (CFAG, KIT3, and SANT). Two of these sites (CFAG and KIT3) are also characterized among the largest representativeness error statistics (Fig. 4c). The time series of IWV and IWV differences for four of the worst cases (CFAG, KIT3, MCM4, and SYOG) can be found in Fig. B2 of Parracho et al. (2018).

Figure 7 shows the values of the four comparison statistics for the 15 outlying cases for the bilinearly interpolated ERA-Interim values and also from the values at the four surrounding grid points (ordered by increasing distance to the GPS station). The results are grouped by region as outlying sites appear to form several clusters located in specific areas of the globe (see Fig. 1). In addition to the four statistics (Fig. 7a to d), we included the altitudes of the GPS stations, h_GPS, and of the four surrounding grid points (Fig. 7e). The above-chosen range limits are superposed in Fig. 7a to d, and a range limit for the altitudes is indicated as h_GPS±500 m (Bock et al., 2014).

Figure 7(a, b) Mean and (c, d) standard deviation of daily IWV difference (ERAI minus GNSS) for 15 outlying sites grouped by region: Andes (CFAG, SANT, AREQ), central Asia (KIT3, POL2), India (IISC), western USA (DHLG, BLYT, LONG, COSO), Hawaii (MKEA), and Antarctica (MCM4, SYOG, MAW1, DAV1). In panels (a) to (d), the black bars show results for the bilinearly interpolated ERA-Interim data, and the grey bars the results for the four surrounding grid points, ordered by increasing horizontal distance from the GPS station. Panel (e) shows the altitudes of the GPS stations (black bar) and the altitudes of the four surrounding grid points (grey bars). The dotted red lines show the acceptable range limits, similar to Fig. 2 for panels (a) to (d), and ±500 m around the GPS station's altitude in panel (e).

Download

AREQ, SANT, and CFAG are all located in the Andes Cordillera, with AREQ (h_GPS=2470 m) and SANT (h_GPS=696 m) on the western flank of the mountain range facing the sea, and CFAG (h_GPS=680 m) on its eastern flank. The local topography peaks above 3000, 4000, and 3000 m within a radius of 100 km from these three sites, respectively. The altitudes of the four surrounding ERA-Interim grid points are very variable (Fig. 7e), and for AREQ (SANT), all (some) of them are exceeding the altitude range limit. At AREQ, absolute and relative standard deviations of the interpolated data slightly exceed the limits, with σ_Δ=2.4 kg m⁻² and ${\mathit{\sigma }}_{\mathrm{\Delta }}^r=\mathrm{21}$ %, while the bias is almost zero. Moreover, most of the statistics at the four surrounding grid points exceed the range limits. There is thus a significant representativeness difference between the four grid points which is not surprising given the steep orography and the very different altitudes of the grid points. Three of the grid points are actually located more than 500 m higher than the GPS station. For these grid points, the validity of the lower pressure-level data can be questioned as the atmospheric variables are extrapolated far below the model's surface. The results at SANT have similar issues with biases again correlated with variations in the model topography. At both sites, issues with the GPS measurements were eliminated by verifying their consistency with collocated DORIS measurements (Bock et al., 2014). Compared to AREQ and SANT, CFAG has much worse results and gets actually the worst statistics of all 15 sites: μ_Δ=5.8 kg m⁻², ${\mathit{\mu }}_{\mathrm{\Delta }}^r=\mathrm{35}$ %, σ_Δ=3.7 kg m⁻², and ${\mathit{\sigma }}_{\mathrm{\Delta }}^r=\mathrm{22}$ %. Contrary to the previous sites, the results for the four grid points are very similar, though the biases vary slightly (from 5.9 to 4.1 kg m⁻²), which suggests that the discrepancy at this site may not be due that much to spatiotemporal variability in the IWV field. Problems with the GPS measurements cannot be excluded at this site and should be checked by comparison with independent observations.

Further insight into the nature of the discrepancies is given by inspection of the seasonal variation of the comparison statistics (Fig. 8) and of the atmospheric environment (Fig. 9). Figure 8 shows that at all three sites, the biases and standard deviations vary over the year, in relation with the variation of the mean IWV (μ_W, Fig. 9a) and the day-to-day variability (σ_W and σ_dW∕dt, Fig. 9b and c, respectively). Both the μ_Δ and σ_Δ are peaking when μ_W is peaking, during the austral summer months. The relative differences, ${\mathit{\mu }}_{\mathrm{\Delta }}^r$ and ${\mathit{\sigma }}_{\mathrm{\Delta }}^r$, and IWV variability, ${\mathit{\sigma }}_W^r$ and ${\mathit{\sigma }}_{\mathrm{d}W/\mathrm{d}t}^r$, are peaking in winter when the mean IWV is low. Inspection of μ_R (Fig. 9d) confirms the strong impact of spatiotemporal variability at all three sites but especially at AREQ, where it is the largest among all sites (peaking at μ_R=6.4 kg m⁻²). It is noticeable that at CFAG the yearly mean and the seasonal cycle of IWV in ERA-Interim are larger than observed by GPS (Fig. 9a), which suggests that a representativeness difference is most likely the explanation rather than GPS measurement issues evoked above.

Figure 8Seasonal variation of (a, b) mean and (c, d) standard deviation of daily IWV difference (ERAI minus GNSS) for 15 outlying sites. The grey bars show the statistics computed for each month (January to December, from left to right) over the 16-year period. The dotted red lines show the range check limits.

Download

Figure 9Seasonal variation of daily IWV data: (a) mean IWV; (b, e) absolute and relative standard deviation of IWV; (c, f) absolute and relative standard deviation of IWV derivative; (d, g) absolute and relative mean spatial variability of ERAI IWV from the four grid points surrounding the GPS sites. The grey (blue) bars show GPS (ERA-Interim) statistics computed for each month (January to December, from left to right) over the 16-year period.

Download

The next two sites, KIT3 (h_GPS=659 m) and POL2 (h_GPS=1755 m), are located in Uzbekistan and Kyrgyzstan, respectively, close to the Alai/Tien Shan mountain range. They both show large difference statistics, with μ_Δ, ${\mathit{\mu }}_{\mathrm{\Delta }}^r$, σ_Δ, and ${\mathit{\sigma }}_{\mathrm{\Delta }}^r$ exceeding the limits for KIT3 and ${\mathit{\mu }}_{\mathrm{\Delta }}^r$ and ${\mathit{\sigma }}_{\mathrm{\Delta }}^r$ for POL2 (Fig. 7a to d). Considering the individual grid points, they almost all also exceed the limits, with large variations both in the bias and standard deviation at KIT3, with somewhat smaller differences at POL2. These variations can again be related to variations in the grid point altitudes, some of which exceed the range limits (Fig. 7e). The difference statistics at these sites exhibit large seasonal variations, with μ_Δ, ${\mathit{\mu }}_{\mathrm{\Delta }}^r$, and σ_Δ peaking in boreal summer (Fig. 8) when μ_W and μ_R are peaking (Fig. 9). The representativeness error statistics peaks are particularly marked at these stations, with KIT3 showing the largest monthly values among all sites (${\mathit{\mu }}_{R,i}=\mathrm{8.8}$ kg m⁻² in August; Fig. 9d). At this site, the GPS measurements were also verified with collocated DORIS measurements (Bock et al., 2014), confirming that representativeness differences between ERA-Interim and GPS IWV data are the main reason for this discrepancy. Interestingly, it can be noticed that the peak in IWV during summer is significantly larger in ERA-Interim compared to GPS (Fig. 9a), suggesting excessive moisture transport into this region in the reanalysis, possibly connected with the too-smooth topography in the model.

The next five sites belong to two geographical regions: IISC in India, and DHLG, BLYT, LONG, and COSO in California, USA, which are all characterized by small discrepancies with only one statistic exceeding the range limits (σ_Δ for the first four, and ${\mathit{\mu }}_{\mathrm{\Delta }}^r$ for COSO). At all five sites, the variation of statistics among the four grid points are small (Fig. 7a to d), as are the variations of the altitudes (Fig. 7e). Station IISC shows a small seasonal variation in the bias and standard deviation (Fig. 8) which might be linked to the variation in IWV temporal variability (Fig. 9b, c, e, f) and spatiotemporal variability (Fig. 9d and g) that show peaks in spring and autumn, i.e. during transitions seasons between the summer monsoon (June to October) and the cooler winter season (December to March). It has been shown previously that monsoon transition periods are accompanied by strong spatial and temporal variability in IWV, which is difficult to represent in atmospheric reanalyses (Bock et al., 2008; Bock and Nuret, 2009; Meynadier et al., 2010; Means, 2013).

The four outlying Californian sites can be separated into two groups: DHLG, BLYT, and LONG, located south of the Sierra Nevada mountain range, in a region of moderate topography, and COSO located in the Basin and Range Province, a narrow valley at the southern exit of the Sierra Nevada. The higher altitude (1485 m) and more complex topographic environment of COSO enhances the representativeness differences. Interestingly, all four sites show a step-like variation of the mean IWV and variability (Fig. 9a, b, c) peaking in July–August–September associated with the North American monsoon (Adams and Comrie, 1997; Means, 2013). This feature contrasts with the Indian monsoon observed at IISC where variability was enhanced during the transition seasons and not during the monsoon. At DHLG and BLYT, the biases actually reverse signs in July–August (Fig. 8a, b) and the standard deviation peaks at σ_Δ>4 kg m⁻² (Fig. 8c). Figure 9b and c show that ERA-Interim underestimates IWV variability at these sites which suggests that GPS observations capture some small-scale moisture variability not represented in the reanalysis.

The next site, MKEA (h_GPS=3730 m), is located on the Mauna Kea volcano on the island of Hawaii. Due to small size of the emerged land area (approximately 10⁴ km²), the imprint of the island is almost inexistent in the reanalysis topography (Fig. 7e). Hence, it is not surprising that the comparison statistics are bad (although only ${\mathit{\sigma }}_{\mathrm{\Delta }}^r$ is exceeding the range limits). The relative IWV differences are huge (Fig. 9e and f) when computed with respect to the low GPS IWV content of this high-altitude site.

The last group of sites is located in eastern Antarctica (Fig. 1). Unfortunately, four of the five Antarctica sites used in this study suffer from large discrepancies. Three of them have two statistics (${\mathit{\mu }}_{\mathrm{\Delta }}^r$ and ${\mathit{\sigma }}_{\mathrm{\Delta }}^r)$ exceeding the range limits (Fig. 7b and d). MCM4 is the worst case and has the largest relative standard deviation among all 15 sites: ${\mathit{\sigma }}_{\mathrm{\Delta }}^r=\mathrm{32}$ %. This station is located in McMurdo Sound, an area with complex landscape, including local low mountain peaks, valleys and glacier corridors, and sea within a radius of 100 km. The other three stations are located close to the coastline backed to the main ice shelf with large surface elevation variations (up to 2000 m within a distance of 100 km). The grid points in ERA-Interim are associated at different altitudes with differences in representativeness leading to IWV biases (Fig. 7b). The marked seasonal variation of ${\mathit{\mu }}_{\mathrm{\Delta }}^r$ and ${\mathit{\sigma }}_{\mathrm{\Delta }}^r$ (Fig. 8b and d) also confirms a dependence of the IWV differences on the atmospheric state and especially on IWV variability, which is enhanced during the austral winter months (Fig. 9e and f). The winter variability is actually much underestimated in ERA-Interim as seen in Fig. 9e and f at MCM4, SYOG, and MAW1, and, quite surprisingly, the spatiotemporal variability, ${\mathit{\mu }}_R^r$, remains nearly constant in ERA-Interim (Fig. 9g). These differences point to an issue in ERA-Interim IWV contents in Antarctica, especially during austral winter, as also suggested by Parracho et al. (2018), who compared ERA-Interim to the NASA Modern Era Retrospective Analysis for Research and Applications version 2 (MERRA-2) reanalysis. These authors also pointed to some issues in the GPS measurements at MCM4 and SYOG between 2002 and 2006, as well as a break in the IWV series at all sites in Antarctica due to a discontinuity in the GPS processing. The IWV issues in ERA-Interim may be linked to the large surface air temperature biases of the reanalysis diagnosed by Bracegirdle and Marshall (2012), from coastal station observations which are related to its too-smooth orography. In addition, Xie et al. (2016), showed that the replicability of daily and annual variance of surface air temperature in this reanalysis decreases from the coast to the interior of the continent. These results also support the findings of Parracho et al. (2018) that the IWV variability and trends in ERA-Interim reanalysis are more realistic near the coast where in situ observations are assimilated than in the interior where the reanalysis mainly relies on satellite observations and short-term model forecasts. Representativeness differences between GPS and ERA-Interim in Antarctica are thus enhanced by deficiencies in the reanalysis.

GPS IWV data have the following DOI: global GPS IWV data at 120 stations of the permanent IGS network; https://doi.org/10.14768/06337394-73a9-407c-9997-0e380dac5591 (Bock, 2016). ERA-Interim data can be downloaded at https://www.ecmwf.int/en/forecasts/datasets/archive-datasets/reanalysis-datasets/era-interim (last access: October 2018; Dee et al., 2011). The results presented in the paper are also provided in the Supplement.

The supplement related to this article is available online at: https://doi.org/10.5194/acp-19-9453-2019-supplement.

OB prepared the GPS and ERA-Interim data, performed the comparisons, and wrote the paper. ACP contributed to the data analysis and discussion of results.

The authors declare that they have no conflict of interest.

This article is part of the special issue “Advanced Global Navigation Satellite Systems tropospheric products for monitoring severe weather events and climate (GNSS4SWEC) (AMT/ACP/ANGEO inter-journal SI)”. It is not associated with a conference.

This research has been supported by the Centre National de la Recherche Scientifique (grant no. LEFE/INSU projet VEGA).

This paper was edited by Roeland Van Malderen and reviewed by two anonymous referees.