The Tropospheric Water Vapour Radiometer (TROWARA) is a microwave radiometer that has been retrieving IWV and integrated liquid water (ILW) since November 1994 in Bern, Switzerland (46.95∘ N, 7.44∘ E; 575 m above sea level, a.s.l.). It measures the thermal microwave emission at the frequencies of 21.39, 22.24 and 31.5 GHz with a time resolution of several seconds and an elevation angle of 40∘. The measured signal is used to infer the atmospheric opacity, using the Rayleigh–Jeans approximation of the radiative transfer equation as described in and .

The opacity linearly depends on the water content in the atmosphere and can therefore be used to derive IWV and ILW : 1τi=ai+biIWV+ciILW, where τi is the opacity of the ith frequency channel of the radiometer. The coefficients ai and bi are statistically derived from nearby radiosonde measurements and fine-tuned with clear-sky measurements . The coefficient ci is the Rayleigh mass absorption coefficient of liquid water.

The initial instrument setup and measurement principle is presented in . To improve the measurement stability and data availability, the instrument was upgraded in 2002 and 2004 and a new radiometer model was developed . Furthermore, it was moved into an indoor laboratory in November 2002, which made it possible to measure IWV even during light-rain conditions . However, to maintain consistency with the measurements before 2002, data observed during rainy conditions were excluded in the present study as soon as the ILW exceeded 0.5 mm or rain was detected by the collocated weather station . We use hourly IWV data from the STARTWAVE database (http://www.iapmw.unibe.ch/research/projects/STARTWAVE/, last access: 29 September 2020) which were derived from the opacities at 21.39 and 31.5 GHz. Before 2008, we use TROWARA data in which data gaps were filled with data derived from a collocated radiometer as described by and . Furthermore, change points in TROWARA data due to instrumental changes have been detected and corrected by a careful comparison of the TROWARA time series before 2008 with simultaneous measurements from other techniques . No instrumental changes have been performed in recent years. We therefore presume that the data are well homogenized and suitable for trend estimation.

A ground-based solar Fourier transform infrared (FTIR) spectrometer is located at the high-altitude observatory Jungfraujoch in Switzerland (46.55∘ N, 7.98∘ E; 3580 m a.s.l.). Water vapour information is retrieved from absorption in the solar spectrum at three spectral intervals within 11.7 and 11.9 µm. The optimized IWV retrieval for FTIR spectrometry is described by , and instrumental details are given in . FTIR measurements at Jungfraujoch provide water vapour data since 1984. For consistency with our study period, we use data only from 1995 to 2018. In this period, two FTIR instruments were installed at Jungfraujoch, with overlapping measurements from 1995 to 2001. showed that the bias between both instruments is negligible. We therefore compute monthly means of a merged time series including both instruments. FTIR measurements are weather dependent (cloud-free conditions are required) and thus provide irregularly sampled data at Jungfraujoch, with on average eight measurement days per month in our study period. This sparse sampling can be problematic when calculating monthly means. We therefore apply the resampling method proposed by when calculating monthly means of FTIR-derived IWV. For this, the background IWV data are determined by fitting a seasonal model to daily IWV means. The seasonal model is given by a mean IWV0, the first two seasonal harmonics with periods Tn=365.25/n, and the fit coefficients an and bn: 2IWV(t)=IWV0+∑n=12an⋅sin⁡2πTn⋅t+bn⋅cos⁡2πTn⋅t. This seasonal model is fitted to the 15th of each month using a window length of 2 years. Due to the sparsity of the FTIR data, the model fit to each month provides a more robust estimate compared to the statistical monthly means, which might be based on only 1 or 2 d of observations at the beginning or end of a month that are not necessarily representative as a monthly mean. The measurement uncertainties of the obtained monthly mean values are derived from the covariance matrix of the model fit. Furthermore, we also tested a seasonal model with higher seasonal harmonics. However, due to the sparse FTIR measurements it appeared not to be useful to improve the obtained monthly mean IWV estimates.

The signal of GNSS satellites is delayed when passing through the atmosphere. This so-called zenith total delay (ZTD) can be used to infer information about the atmospheric water vapour content. Various studies explain the method to derive IWV from the measured ZTD e.g.. We briefly summarize the procedure that we used in our study. The ZTD can be written as the sum of (i) the zenith hydrostatic delay (ZHD) due to refraction by the dry atmosphere and (ii) the zenith wet delay (ZWD) due to refraction by water vapour : 3ZTD=ZHD+ZWD. The ZHD (in metres) is calculated from the surface pressure at each GNSS station as proposed by : 4ZHD=(2.2768±0.024)×10-3psf(λ,H), with surface pressure ps in hectopascals. The dependency of the gravitational acceleration on latitude and altitude is considered in the function f : 5f(λ,H)=1-0.00266cos2λπ180-0.00028H, where λ is the station latitude in degrees and H is the station altitude in kilometres. With the measured ZTD and the calculated ZHD, we obtain the ZWD (Eq. ), which can then be used to infer information about the IWV in millimetres. It is calculated according to with 6IWV=κZWD1ρH2O, where ρH2O is the density of liquid water (ρH2O=1000 kg m-3). The factor κ is given by 71κ=Rvk3Tm+k2′10-6, with the constants k3 and k2′ as derived by from (k3=(3.776±0.004)×105 K2 hPa-1 and k2′=17±10 K hPa-1). The required estimate of the mean atmospheric temperature Tm is linearly approximated from the surface temperature Ts (damped with the daily mean) as proposed by (Tm=70.2K+0.72Ts). Another possibility would be to estimate Tm from reanalysis data. However, GNSS estimates would then depend on reanalyses, which would make validation of GNSS with reanalyses problematic. Furthermore, showed that the use of reanalyses temperature and pressure data can lead to a bias in IWV compared to the use of surface measurements, especially in mountainous regions in Germany. We therefore follow their recommendation to use the Bevis approximation derived from surface temperature. The pressure ps and the surface temperature Ts at the GNSS station are vertically interpolated from pressure and temperature measurements at the closest meteorological station, assuming hydrostatic equilibrium and an adiabatic lapse rate of 6.5 K km-1.

We use hourly ZTD data from the Automated GNSS Network for Switzerland (AGNES), containing 41 antennas (at 31 locations), as well as data from a few stations that are part of the COGEAR network (https://mpg.igp.ethz.ch/research/geomonitoring/cogear-gnss-monitoring.html, last access: 29 September 2020) and from two additional stations in Bern. The AGNES network was established in 2001 , and it is maintained by the Swiss Federal Office of Topography (swisstopo). A monitor web page shows the current status of all stations . In 2008, most of the antennas and receivers were enhanced from GPS only to GPS and GLONASS (Russian global navigation satellite system). Since spring 2015, AGNES has been a multi-GNSS network also using data from Galileo (European global navigation satellite system) and BeiDou (Chinese navigation satellite system). All European GNSS data were reprocessed in 2014 within the second EUREF (International Association of Geodesy Reference Frame Sub-Commission for Europe) Permanent Network (EPN) reprocessing campaign as described in . In the present study, only the reprocessed ZTD products of swisstopo are used .

The stations used in our study are shown in Fig.  and listed in Table . We only use stations that provide measurements for more than 10 years. At some GNSS stations, a new antenna and receiver were installed at the same or nearby location, replacing the older ones after an overlapping measurement period. An antenna change often leads to a small height difference, which can lead to a jump in the ZTD time series. It is therefore important to decide how to handle such instrumental changes for trend analyses. In cases of antenna and receiver replacements, we merged these stations to a single time series by calculating the mean value for overlapping periods. They are marked by “_M” (for “merged”) in their station abbreviation (Table ), and a potential jump was considered in the trend estimation (see Sect. ). At nine stations, new multi-GNSS receivers and antennas were installed at an additional location nearby, but the old GPS-only receivers and antennas are still operating. Swisstopo installed such twin stations to ensure a best possible long-term consistency. Simply replacing antennas at all stations would not guarantee continuous time series, even if the phase centres of the antennas were individually calibrated. Furthermore, no calibrations have been available for the tracked satellite systems Galileo and BeiDou until today. In the case of twin stations, we only used the old, continuous GPS-only station, because the stability is better suited for trend calculations than merged time series with potential data jumps.

IWV, relative humidity (RH) and temperature data from two reanalysis products are used in the present study, the ERA5 and the MERRA-2 reanalyses. The Modern-Era Retrospective Analysis for Research and Applications, version 2 (MERRA-2), is an atmospheric reanalysis from NASA's Global Modeling and Assimilation Office (GMAO), described in . The MERRA-2 product used in the present study for IWV data contains monthly means of vertically integrated values of water vapour ( (GMAO), 2015) with a grid resolution of 0.5∘ latitude × 0.625∘ longitude. The ERA5 reanalysis is the latest atmospheric reanalysis from the European Centre for Medium Range Weather Forecasts (ECMWF) . In the present study, we use an ERA5 product providing integrated water vapour and another product providing RH and temperature profiles , both with a grid resolution of 0.25∘ latitude × 0.25∘ longitude ( (C3S), 2017). Reanalysis models assume a smooth topography that can deviate from the real topography, especially in mountainous regions . For validation of reanalysis data with specific station data (e.g. GNSS), the reanalysis IWV value would need to be corrected for altitude differences as proposed by for example or . For linear trends, however, such a linear correction is not relevant. We therefore use uncorrected reanalysis data, which might lead to some differences in IWV when comparing reanalysis IWV directly with IWV measured from the radiometer or at a GNSS station.

When using reanalysis data for trend estimates, one has to keep in mind their limitations. Due to changes in observing systems of the assimilated data, the use of reanalyses for trend studies has been debated e.g.. The recent reanalysis products contain some improvements in handling possible steps in assimilated data. For example, the bias correction of assimilated data in ERA5 has been extended to more observation systems and MERRA-2 reduced certain biases in water cycle data . Nevertheless, future studies have to assess whether these improvements affect the reliability of reanalysis data for trend estimates. We exclude MERRA-2 lower-tropospheric-temperature trends in our study because we found unexpected large trends in some Alpine grids. They seem to be related to a bias in tropospheric temperature in some grids after 2017, but further investigations would be required to understand the origin of the observed trends. In this study, we therefore concentrate on ERA5 data for the temperature-related analyses in Sects.  and .

The trend model is able to consider jumps in the time series by assuming a bias for a given subset of the data. For this, a fully correlated block is added to the part of Sy that corresponds to the biased subset. For each subset, the block in Sy is set to the square of the estimated bias uncertainty of this block. The block with the most data points (longest block) is set as a reference block in which no bias is assumed. This possibility of bias fitting in the trend estimation has been presented in and is mathematically explained in . The method has been applied for example by to consider a data jump after retrieval changes in a satellite product. It is also described in , in which it has been applied on ozone data to consider data irregularities in a time series due to instrumental anomalies.

Trends of IWV for the different data sets around Bern are shown in Fig.  and Table . IWV measured by the radiometer TROWARA increased significantly by 4.8 % per decade from 1995 to 2018. This trend value is similar to the bias-corrected trends from GNSS stations in Lausanne (EPFL), Huttwil (HUTT), Lucerne (LUZE), Neuchâtel (NEUC) and Wabern next to Bern (WAB1), which all report trends around 5 % per decade (Fig.  and Table ). We observe a slightly larger trend in Payerne (PAYE, 7.0 % per decade). The GNSS station in Bern, located on the roof of the university building of exact sciences (EXWI), shows a trend of nearly zero (0.1 % per decade). Unfortunately, the site EXWI has not been operation since September 2017. Reanalysis IWV at Bern increases significantly by 3.7 % per decade for MERRA-2 and by 2.3 % per decade for ERA5 data, both for the period from 1995 to 2018. With the exception of Payerne, all GNSS trends are not significantly different from zero at the 95 % confidence interval. The larger GNSS trend uncertainties compared to TROWARA and reanalysis trends are mainly due to the bias correction, which adds some uncertainty to the trend estimates. Furthermore, all GNSS trends result from a shorter time period than TROWARA and reanalysis trends (see Table ), which also increases the trend uncertainty and may lead to some trend differences. For comparison, the GNSS trends without bias correction are also shown in Table . They are generally smaller than the bias-corrected trends, suggesting that GNSS trends are mostly underestimated when biases are not accounted for. Furthermore, their uncertainties are smaller, reflecting the additional uncertainty when biases are considered.

In brief, most of the GNSS stations around Bern report positive trends of approximately 5 % per decade. However, two of the GNSS stations around Bern (EXWI and PAYE) report different trends. The near-zero trend at the EXWI station is less reliable than the other trends because the EXWI station is not part of the AGNES network and therefore does not fulfil the same quality requirements. The large GNSS trend in Payerne results from the bias correction. If the bias correction in the trend fit (as described in Sect. ) is not applied, the trend in Payerne is only 2 % per decade (0.32 mm per decade), whereas it increases to 7.0 % per decade (1.09 mm per decade) when accounting for antenna changes. found IWV trends in Payerne from GNSS measurements of 0.8 mm per decade, which lie between our corrected and uncorrected trends. This suggests that the instrumental changes in Payerne play an important role but might be overcorrected in our case. The recent study by reports similar IWV trends in Payerne using nighttime radiosonde measurements (6.36 % per decade) and even larger trends using clear-night lidar data (8.85 % per decade) in the period from 2009 to 2019, suggesting that IWV in Payerne was strongly increasing, especially in recent years. However, comparing their trend results with ours has to be done with care, because their trend time period is short and the lidar trends might contain a clear-sky bias.

The trend from the TROWARA radiometer of 4.8 % per decade (0.72 mm per decade) slightly differs from the TROWARA trends reported by and . It is larger than TROWARA's 1996 to 2007 trend of 3.9 % per decade (0.56 mm per decade) . found no significant TROWARA trend for the period 1994 to 2009, which suggests that our larger IWV trends are mainly due to a strong IWV increase in the last decade. This is also confirmed by , who observed larger trends for recent years (1.5 mm per decade for 2004 to 2015). However, care has to be taken when comparing these TROWARA trends of different trend period lengths.

To summarize, IWV trends around Bern from TROWARA and GNSS data generally lie around 5 % per decade, whereas reanalysis trends for the Bern grid are slightly smaller.

To examine the relationship between IWV trends and changing temperature, we present the theoretical change in water vapour in the atmosphere due to observed changes in temperature (Fig. c). For this, we determined the temperature-dependent change in saturation vapour pressure for the time period 1995 to 2018. The saturation vapour pressure es describes the equilibrium pressure of water between the condensed and the vapour phase. It increases rapidly with increasing temperature . In cases where the water vapour pressure e is smaller than es, the available water is in the vapour phase, whereas for e≥es it condenses. With increasing temperature, es increases, which leads to an increase in e for a given relative humidity (RH). Changes in es can therefore directly be compared to changes in the amount of water vapour, assuming that RH remains constant : 11RH=ees≈constant. A change in es is then directly reflected in a change in e and therefore in IWV: 12deses≈dee=dIWVIWV.

The fractional change in es for a given change in temperature can be approximated by the Clausius–Clapeyron equation: 13deses≈LvRvT2dT, where Lv is the latent heat of evaporation (Lv=2.5×106Jkg-1), Rv is the gas constant for water vapour (Rv=461JK-1kg-1), dT is the change in temperature and T is the actual temperature. To obtain the tropospheric temperature change dT, we derived the temperature trend (1995 to 2018) from ERA5 temperature profiles, averaged below 500 hPa. This limit was chosen because around 95 % of IWV resides below 500 hPa for the averaged ERA5 profiles in our study period. The resulting temperature trend (in kelvin per decade) is then used for dT in Eq. () to determine the change in es in percent per decade. For the actual temperature T we used the mean of ERA5 temperature profiles below 500 hPa for the same time period.

The fractional changes in ERA5 es for the Bern grid for different months are shown in Fig. c. These temperature-induced changes in es agree generally well with the observed trends in IWV. They agree especially well with TROWARA and reanalysis trends in spring (March and April), late summer and autumn (July to November), but they agree less in the winter months and in May and June. Furthermore, they agree less with GNSS trends from September to March. Generally, the good agreement between the change in es and the IWV trends indicates that observed IWV changes around Bern can mostly be explained by temperature changes. However, the changes in es do not confirm our observed IWV winter trends, especially in January and February. This discrepancy can be related to changes in RH, which was assumed to be constant (Eq. ). Indeed, our trends of ERA5 RH for the Bern grid (Fig. c) show that RH was not constant in those months, especially in winter but also in May and June. Even though the RH trends are not significantly different from zero, these results suggest that assuming RH to be constant may not be valid during all seasons, especially in winter. This makes the attribution of IWV trends to changes in temperature more challenging. Furthermore, other factors than temperature might be responsible for IWV changes in winter, such as changes in dynamical patterns and the horizontal transport of humid air. Indeed, showed that evaporation of surface water plays a minor role in winter, with a latent heat flux that is 6 to 7 times smaller than in summer in Bern, suggesting that, in winter, horizontal transport of humid air is more important than evaporation.

We conclude that IWV in Bern changes as expected from temperature changes in early spring, late summer and autumn, but other processes might also be responsible for IWV changes, especially in winter.

The IWV trends at the Jungfraujoch station from FTIR and fully sampled GNSS data are presented in Fig. . The GNSS antenna update has been considered in the trend estimate as described in Sect. . We observe IWV trends of 0.08 mm per decade (2.6 % per decade) for GNSS and 0.04 mm per decade (1.8 % per decade) for FTIR. However, both trends are insignificant. The difference between both trends can partly be explained by the dry sampling bias of the FTIR spectrometer, which measures only during clear-sky day conditions. Indeed, the absolute GNSS trend is comparable with the FTIR trend when we use GNSS data coincident with FTIR measurements, with 0.05 mm per decade (not shown). Our IWV trends at Jungfraujoch are similar to the trend by , who reported insignificant FTIR trends at the same station of 0.08 mm per decade in the time period 1996 to 2008. In contrast to these results, found larger trends at Jungfraujoch that were significantly different from zero. They decided not to use GNSS IWV data from Jungfraujoch due to the high IWV variability and the missing calibration of the antenna before the replacement in 2016. Therefore, they derived their trends from IWV data based on a parameterization from surface temperature and relative humidity measurements. However, they admit that this approximation is prone to large uncertainties , which might explain parts of the differences compared to our trends.

The GNSS data generally report positive IWV trends throughout Switzerland (Fig. ). Using data for the whole year (Fig. a), 50 % of the stations show trends between 2.3 % per decade and 5.1 % per decade (0.27 and 0.74 mm per decade). The trends of all stations range between 0.1 % per decade and 7.2 % per decade (0.01 mm per decade and 1.09 mm per decade), with exception of three stations that show negative trends (ANDE, HOHT and MART_M). The mean trend value of all GNSS stations is 3.6 % per decade (0.49 mm per decade), and the median is 4.4 % per decade (0.57 mm per decade).

Only three stations (9 % of all stations) show negative IWV trends and none of them is significantly different from zero at the 95 % confidence interval. Significant positive trends are reported at 17 % of the stations (six stations), being generally stations with long time series and lying mostly in western and south-eastern Switzerland. Most significant trends are observed in summer (Fig. d), with significant positive trends at five stations. In winter, only two north-eastern Swiss station trends are significant (Fig. b). In spring (Fig. c) and autumn (Fig. e), none of the IWV trends are significantly different from zero. Autumn trends tend to be negative, especially in the south-western part (Rhône valley in the canton of Valais), but they are all insignificant.

Our trend range covered by all GNSS stations is consistent with results from , who observed in Sweden and Finland IWV trends between -0.2 and 1 mm per decade. However, they concluded that their study period was too short (10 years) to obtain stable trends. Our trends also lie within the range of trends observed in Germany by . Their trends vary even more between different stations, with trends ranging between -1.5 and 2.3 mm per decade. Note, however, that both studies use different trend period lengths than in our study, which makes trend comparisons difficult. The recent study by reports IWV trends from GNSS data at three Swiss stations for the period 2001 to 2015. Using Sen's slope trend method, they found positive all-sky trends in Davos (0.89 mm per decade), Locarno (0.42 mm per decade) and Payerne (0.80 mm per decade).

Our GNSS trends for these stations are slightly different (Davos: 0.71 mm per decade, Locarno: 0.72 mm per decade, Payerne: 1.09 mm per decade), which might be due to the three additional years in our analysis, but also due to our bias correction in the trend model. Furthermore, our GNSS-derived ZTD data were reprocessed until 2014 (see Sect. ), whereas still used the old GNSS data.

Note that most bias-corrected GNSS trends are larger than the uncorrected trends. This suggests that earlier GNSS data overestimate IWV compared to recent measurements. A possible explanation might be the enhancement from GPS-only to multi-GNSS antennas that was performed on AGNES stations in spring 2015. In our example case (Fig. b), this update improved GNSS IWV measurements compared to ERA5 data, suggesting that IWV was overestimated by GNSS in earlier years. This overestimation would then lead to a smaller trend, whereas the trend would increase when the jump is corrected in the trend estimation.

The altitude dependence of the GNSS trends is shown in Fig. . We observe that most of the stations that show significant positive trends lie at higher altitudes. Indeed, 83 % of the stations showing significant trends lie at altitudes above 850 m a.s.l., whereas less than half of the stations lie above 850 m. This is consistent with the expectation of that the rate of warming is larger at higher altitudes. Due to the direct link between temperature and water vapour content, an increased warming at higher altitudes would lead to larger IWV trends. The increasing significance with altitude provides some observational evidence for this suggestion. However, the altitude dependence is less visible in absolute trends (not shown), which indicates that, due to less IWV at higher altitudes, these trends are more sensitive to changes when calculating trends in percent. Also, the IWV trends of the six stations with highest altitudes (>1650 m a.s.l.) are not significantly different from zero.

We conclude that Swiss GNSS stations generally show positive IWV trends, with a mean value of 3.6 % per decade (0.49 mm per decade) and a tendency for more significant percentage trends at higher altitudes.

Reanalysis trends of IWV for Switzerland are presented in Figs.  and . The trends are on average 2.6 % per decade (0.35 mm per decade) for ERA5 (Fig. a) and 3.6 % per decade (0.52 mm per decade) for MERRA-2 (Fig. a). Both reanalysis trends show only small spatial variability. The seasonal trends are positive, with the largest values in summer (Figs. d and d). This is consistent with our observed GNSS trends, which are mostly positive in summer. The smallest and partly negative reanalysis trends are observed in winter (Figs. b and b), which contrasts with our GNSS trends that showed the smallest (but insignificant) trends in autumn and not in winter. In spring and autumn, the reanalysis trends are spatially more variable. Both data sets report slightly larger autumn trends in south-eastern Switzerland and northern Italy (Figs. e and e). In spring, ERA5 shows larger IWV trends in south-western Switzerland.

The averaged MERRA-2 trend agrees with our averaged GNSS trend (both 3.6 % per decade), which is slightly larger than the averaged ERA5 trend (2.6 % per decade). However, the reanalyses do not resolve small-scale variability, which can explain the differences compared to some GNSS station trends. Furthermore, the GNSS point measurements are generally more variable than the gridded reanalyses data. also observed larger differences in mountainous regions between GNSS-derived IWV and reanalyses data in Germany. Our mean ERA5 trend for Switzerland of 0.35 mm per decade is consistent with IWV trends from ERA-Interim in Germany reported by (0.34 mm per decade). The MERRA-2 trends are generally slightly larger than the ERA5 trends. also found larger IWV trends for MERRA-2 compared to ERA-Interim reanalysis trends on a global scale, especially in summer.

To determine the relationship between temperature changes and IWV trends for the whole of Switzerland, we present changes in saturation vapour pressure es derived from ERA5 temperature changes below 500 hPa (as described in Sect. ). The fractional change in ERA5 es, which corresponds to the change in IWV (Eq. ), is presented in Fig. . The averaged changes in ERA5 es of 2.9 % per decade are similar to our ERA5 IWV trends described before (2.6 % per decade), which indicates that IWV is on average following the temperature change as expected from the Clausius–Clapeyron equation. The ERA5 es changes are spatially more uniform than the ERA5 IWV trends but agree well in all seasons, except in winter (Figs. b and b). ERA5 es is decreasing in winter, whereas ERA5 IWV winter trends are increasing. These conflicting results indicate that factors other than temperature might dominate IWV changes in winter, as already discussed in Sect. . Furthermore, it indicates that the assumption of constant relative humidity might not be valid in winter. This is confirmed by the ERA5 RH trends (Fig. ), which are around zero for the whole of Switzerland in all seasons but slightly positive in winter. Even though these positive winter RH trends are not significantly different from zero, they raise the question of whether it is justified to assume RH to be constant.

The partly negative winter changes in es are surprising because they result from a decrease in reanalysis winter temperature. Such a decrease in winter temperature is not consistent with long-term temperature observations in Switzerland, which report a temperature increase also in winter . This difference is due to our short study period. A few cold winters in the past 15 years have hidden the overall positive temperature trend when looking only at the relatively short period from 1995 to 2018 (, Federal Institute for Meteorology and Climatology, 2019).

To summarize, the ERA5 IWV trends follow on average the changes expected from temperature changes. The reanalysis IWV trends generally agree well with GNSS trends in Switzerland, but the spatial trend variability is not resolved by the reanalyses. Local measurements of IWV such as microwave radiometer, FTIR or GNSS measurements are therefore crucial to monitor changes in IWV, especially in mountainous regions such as Switzerland.