Validating the water vapour content from a reanalysis product and a regional climate model over Europe based on GNSS observations

regional climate model over Europe based on GNSS observations Julie Berckmans1,2, Roeland Van Malderen1, Eric Pottiaux3, Rosa Pacione4, and Rafiq Hamdi1 1Royal Meteorological Institute, Ringlaan 3, 1180 Brussels, Belgium 2Centre of Excellence Plants and Ecosystems (PLECO), University of Antwerp, Belgium 3Royal Observatory of Belgium, Ringlaan 3, 1180 Brussels, Belgium 4e-GEOS S.p.A. ASI/CGS Matera, Italy Correspondence to: Julie Berckmans (julie.berckmans@vito.be)

The significance of the differences between ERA-Interim/ALARO-SURFEX and the GNSS observations was assessed by applying the Kolmogorov-Smirnov (K-S) test. This test determines whether the datasets have the same (continuous) distribution. As integrated water vapour follows a non-normal distribution (Foster et al., 2006), the advantage of using this test is that it does not make any assumption on the distribution of the data and is non-parametric (von Storch and Zwiers, 1999). The 10 significance was tested for the monthly values over the entire simulated 19-yr period at the 5% significance level.

Dataset comparison
The IWV monthly values (all stations together) show a correlation coefficient of 0.99 and 0.98 for ERA-Interim and ALARO-SURFEX respectively with the IWV derived from GNSS (Fig. 2) IWV bias) at the low IWV range, and an increasing IWV underestimation (increasing negative biases) in the middle and upper range of IWV Values (see Table 1).
The linear regression slopes of ERA-Interim IWV vs. GNSS IWV are increasing for the different IWV ranges considered in Table 1), and is larger than 1 for the upper range (above 25 kg m -2 ). This is due to an increasing number of ERA-Interim IWV outliers above the 1:1 line with increasing IWV values. For all IWV ranges, ERA-Interim has a positive IWV bias w.r.t. GNSS, with similar values. The linear regression slopes of ALARO-SURFEX IWV vs. GNSS IWV are lower than 1 for all IWV ranges 10 ( Table 1). Despite smaller (but negative) biases by ALARO-SURFEX for all IWV ranges, the standard deviations are larger for both middle and upper range (below 10 kg m -2 and 10-25 kg m -2 ). This indicates a larger variability of ALARO-SURFEX w.r.t. GNSS around the 1:1 line than the ERA-Interim w.r.t. GNSS. This is due to the lack of assimilated ground-based observations by ALARO-SURFEX (Ning et al., 2013).

15
The mean GNSS-based IWV over the 19-yr period is 16.31 kg m -2 and is higher for ERA-Interim with 16.58 kg m -2 and lower with ALARO-SURFEX with 16.22 kg m -2 (Fig. 3a). Both ERA-Interim and ALARO-SURFEX are able to reproduce the yearly IWV variability. For all data sets, the inter-annual variability is very small compared to the intra-annual variability (Fig.   3a, Table 2). The mean inter-annual variability is 1.20, 1.17 and 1.16 kg m -2 for GNSS, ERA-Interim and ALARO-SURFEX respectively, whereas the mean intra-annual variability is 5.57, 5.52 and 5.39 kg m -2 for the corresponding datasets. The IWV overestimation of ERA-Interim w.r.t. GNSS is a persistent feature for all the years, whereas ALARO-SURFEX has a positive IWV bias w.r.t. GNSS at the beginning of the year and a negative IWV bias in the summer periods (Fig. 3b).  Interim vs. GNSS IWV differences is larger for the first years of the time series and is less pronounced afterwards. A possible explanation is the increased data assimilation in ERA-Interim in the more recent years. On the other hand, the ERA-Interim IWV biases w.r.t. GNSS seem to be rather constant in time. This is in contrast to the ALARO-GNSS IWV biases, which seem to decrease in time (drifting) (Fig. 3b). The seasonal cycle of ALARO-SURFEX exists for all the years, with a peak of overestimated IWV in spring.
The overall IWV seasonal cycle, averaged over all the GNSS station locations, is shown for the three datasets in Fig. 4a. The nearest to the GNSS stations shows a wet bias of 5-38% (Fig. 5a) compared to the 0.22 • ECA&D E-OBS dataset (Haylock et al., 2008). E-OBS is a daily high-resolution gridded observational dataset, which consists of the daily mean temperature and the daily accumulated precipitation. The most recent version v14.0 was selected on the 0.22 • rotated pole grid, corresponding to a 25 km horizontal resolution in Europe.
The overall wet bias is associated with an overall cold bias of -0.5 • C to -1.8 • C when averaging the 2 m temperature measurements over the grid boxes that contain the GNSS station locations (Fig. 5b). These findings are in agreement with the Both precipitation and temperature follow a clear seasonal cycle with strong biases in winter that increase in early spring and subsequently decrease in summer, when they are minimal, followed by gradually increasing biases again in autumn. The seasonal cycle of precipitation bias coincides well with the seasonal cycle of IWV bias for autumn, winter and spring (Fig. 4b). The largest difference between the ALARO-SURFEX IWVs and the GNSS-based IWVs occur in summer (Fig. 4b). On the contrary, the smallest temperature and precipitation bias appears in summer. This can be explained by a feedback between the land surface and the atmosphere that is strongest during summer (Seneviratne et al., 2010). Even though the temperature biases are smallest in summer, the means are still negative with values of -0.45 • C, -0.63 • C and -1.06 • C for June, July and August respectively. The lower temperatures in summer as compared to the observations are likely to lead to lower evapotranspiration 5 rates. This result is in agreement with the findings by Ning et al. (2013).
The water vapour-evapotranspiration feedback is less dominant in the winter, when large-scale stratiform systems determine the weather and less interaction exists between the land surface and the atmosphere (Koster et al., 2000). Moreover, the precipitation is largely overestimated in winter (Fig. 5a) which corresponds to a larger overestimation of IWV (Fig. 4b). During autumn and spring, the previously explained land surface-atmosphere interaction exists for a selected number of stations. More 10 specifically, it exists for 6 stations in autumn and 4 stations in spring (Fig. 5, Fig. 4b), as the region covering our stations hosts different climate regimes. Therefore, the coupling between the temperature and the IWV depends on the coupling strength between the land surface and the atmosphere.  In winter, a large number of stations (about 80%) show smaller differences between ALARO-SURFEX and GNSS than between ERA-Interim and GNSS (Fig. 6b). In total, 81 stations present IWV differences between -0.5 m -2 and +0.5 m -2 . This results in a mean difference of only 0.03 kg m -2 compared to 0.34 kg m -2 by using ERA-Interim. The mean difference is smallest for stations in flat areas (< 100 m) with -0.04 kg m -2 , and increases for stations at higher altitudes (> 100 m & < 1000 m) with 0.07 kg m -2 . At highest altitudes (> 1000 m) the IWV difference is much larger with 0.34 kg m -2 (Table 3). Therefore, ALARO-SURFEX is more sensitive than ERA-Interim to the station height for modelling the appropriate IWV. However, the variability (indicated by the standard deviations) is similar for the different altitudes.
Similarly as for the winter, ERA-Interim overestimates IWV for 69% of the GNSS stations in summer (Fig. 6c). This results

5
in an average value of 0.22 kg m -2 which is lower than the winter averaged value by ERA-Interim. Furthermore, 73 stations give small differences of -0.5 kg m -2 to +0.5 kg m -2 between ERA-Interim and the observations. Smallest IWV differences occur for the GNSS stations located in flat areas with an average of 0.16 kg m -2 (< 100 m) (Table 3). For stations at higher altitudes (> 100 m & < 1000 m) the mean IWV difference increases to 0.34 kg m -2 and is -0.6 kg m -2 for stations at altitudes above 1000 m (Table 3). This deterioration of the IWV difference with highest altitudes of the GNSS stations could not be 10 distinguished for winter.
In contrast to the general overestimation of ERA-Interim, ALARO-SURFEX underestimates IWV in summer for 76% of the stations (Fig. 6d). The geographical distribution of the ALARO-SURFEX biases with GNSS is different in summer and winter.
Large negative IWV differences occur for GNSS stations located in flat areas (< 100 m) with an average of -0.69 kg m -2 and 15 large positive IWV differences exist for GNSS stations located at high altitudes (> 1000 m) with an average of 1.28 kg m -2 (Table 3). The IWV is best represented by the stations between 100 m and 1000 m of altitude.
In summary, we find that the dependence of the model-GNSS IWV bias on the altitude of the GNSS station is the strongest in summer and also the strongest in ALARO-SURFEX. The IWV biases increase with increasing altitude for the stations, both for summer and winter, except in winter with ERA-Interim. For stations with height differences larger than 500 m, IWV 20 differences from 0.75 kg m -2 up to 1.71 kg m -2 are observed except for ALARO-SURFEX in winter.

Hourly variability
The IWV varies during the day due to changes in the solar radiation. Therefore, it is of interest to investigate the capability of the model to capture this daily variations. The IWV is observed on a high temporal resolution by the GNSS stations, hence it is a valuable technique for validating the ability of regional climate models to represent the diurnal cycle (Ning et al., 2013;Wang et al., 2007). It is not possible to investigate the exact comparison of the amplitude of the diurnal cycle, as a phase shift could be present (Ning et al., 2013) and because this shift is not detectable as we only archived ERA-Interim data every 6 hrs.
So, hereafter we investigate the seasonal variations of the diurnal cycle.
The observed IWV diurnal cycle is the strongest in summer with an amplitude of approx. 1 kg m -2 and the weakest in winter 5 with an amplitude of less than 0.5 kg m -2 (Fig. 7). For all seasons, the observed IWV peaks in the evening at 18 UTC and is lowest at 06 UTC (Fig. 7). These characteristics of the IWV diurnal cycle are captured similarly by the three datasets. Different factors contribute to this diurnal variation, the most important factor being the temperature increase during the day that drives the evaporation due to increased water vapour holding capacity, which increases the IWV. During the night, condensation takes place, which causes cooling and a decrease of the IWV (Ortiz de Galisteo et al., 2011). The diurnal variability is controlled by 10 varying factors such as precipitation and wind speed (Wang et al., 2007;Wang and Zhang, 2009).
The IWV diurnal differences between ERA-Interim or ALARO-SURFEX and the GNSS-based IWVs are examined (Fig.   8), again for the different seasons separately. ERA-Interim overestimates IWV for all seasons and for all hours, except for 06 UTC and 12 UTC in summer (Fig. 8a). During winter, ERA-Interim presents largest differences of 3-4% that are consistent from midnight till evening. For spring, summer and autumn, the differences are largest at 00 and 18 UTC (around 2 to 3%), In contrast to ERA-Interim, the representation of the IWV diurnal variations by ALARO-SURFEX is now highly different between spring and winter on the one hand and summer and autumn on the other hand (Fig. 8b). During summer and autumn, the IWV is underestimated by the model at all time, with increasing differences from midnight (when they are close to zero) towards noon and decreasing again in the afternoon. We believe that the strong underestimation at 12 UTC is due to the mechanism of an evaporation-temperature interaction that is most pronounced during the day due to incoming solar radiation.
The same diurnal variations of the ALARO-GNSS biases are found for spring and winter, but are offset with 1 to 2%, so that the best agreement is now found for the observations at noon, and the worst at midnight. So ALARO-SURFEX simulates IWV better at 00 UTC for summer and autumn at 12 UTC for spring and winter (Fig. 8b).

5
In contrast to summer, ALARO-SURFEX is very good in representing the IWV daily cycle in winter and outperforms ERA-Interim (Fig. 8a,b). Moreover, it performs well for spring with similar values than ERA-Interim. A possible explanation might be that the water vapour during winter and spring is more controlled by large-scale stratiform systems, that are forced by ERA-Interim. Therefore, the behaviour of ERA-Interim and ALARO-SURFEX in winter and spring is more comparable than in summer and autumn. Moreover, it seems that ALARO-SURFEX further improves the representation of the IWV daily cycle This study explored the potential of the ERA-Interim reanalysis and the Regional Climate Model (RCM) ALARO-SURFEX in simulating the atmospheric water vapour content derived from ground-based GNSS observations. The reprocessing efforts made within EUREF during the course of the the COST Action ES1206 (GNSS4SWEC)  For all datasets, the intra-annual variability was much higher than the inter-annual variability by a factor of 4 to 5. Both ERA-Interim and ALARO-SURFEX were capable of reproducing the observed yearly IWV cycle. However, ERA-Interim overestimated IWV for most years with an average of 0.27 kg m -2 , with a decreasing seasonality in the more recent years, possibly due to the increased amount of data assimilated in ERA-Interim along the years. ALARO-SURFEX demonstrated a 15 strong seasonal effect with overestimated IWV values in spring and underestimated IWV values for the summer periods.
The variability of the monthly differences was 25% higher by ALARO-SURFEX than by ERA-Interim, possibly due to the fact that no observations were assimilated by ALARO-SURFEX. The significant underestimation of IWV in summer by ALARO-SURFEX was related to the modelled precipitation and temperature bias. An overall cold and dry bias in the summer led to lower evaporation rates and thus an underestimation of the IWV by ALARO-SURFEX. This mechanism was most pro-20 nounced in summer as land surface-atmosphere feedbacks are strongest in summer.
No clear latitudinal or longitudinal dependence of the IWV biases could be detected. Such effect is probably hampered by the seasonal variation of the biases. The dependence of the IWV on the altitude of the GNSS station was strongest in summer and strongest for ALARO-SURFEX. The IWVs were corrected for a possible height difference between the surface model 25 height and the actual height of the station. For stations with large height differences between the reanalysis or model grid box and the GNSS station, the IWV differences were highest of all stations.
The IWV peaks in the evening and reaches its minimum in the morning. The diurnal cycle has the largest amplitude in summer and was even enhanced by ERA-Interim. The diurnal cycle of ERA-Interim was most comparable to ALARO-SURFEX in winter and spring. However, ALARO-SURFEX outperformed ERA-Interim in winter with smaller IWV differences. ALARO-30