The radiative balance between incoming solar radiation and outgoing longwave radiation is the primary regulator of Earth's climate. Changes in atmospheric components, such as aerosols and greenhouse gases which affect the radiative balance, have an impact on climate (McCormic and Ludwig, 1967; Twomey, 1974; Lohmann and Feichter, 2005; Boucher et al., 2013). As the dominant greenhouse gas, water vapor is considered to be the main feedback agent of the atmosphere's state (Held and Soden, 2000; Dessler et al., 2008). As its concentration mostly depends on the air temperature, climate models suggest an amplified initial warming effect in global warming scenarios (Boucher et al., 2013). This important feedback roughly doubles the amount of warming caused by carbon dioxide (Held and Soden, 2000; Soden et al., 2002, 2005). In addition it is also involved in most of the atmospheric processes, such as frontal generation systems (Van Baelen et al., 2011), cloud formation, atmospheric mixing, photochemical processes (McCormack et al., 2008) and aerosol hydration (Feingold et al., 2003; Estillore et al., 2016).

The validation of numerical weather forecast and climate models usually falls to the low spatial and temporal resolution of the observational parameters. Unlike other greenhouse gases, water vapor can be highly variable both in space and time, making it hard to simulate. To help address this issue more accurate and nested observational data are needed. Retrievals from space-borne passive sensors can provide some information but their vertical resolution is insufficient given the frequent occurrence of strong vertical gradients.

Based on the measuring technique, water vapor mixing ratios (WVMR) can be separated into two major categories. The first category considers in situ measurements of temperature and relative humidity which can be converted into WVMR. A plethora of weather stations provide ground-based WVMR over the globe. Nevertheless, such measurements are not representative for the whole atmosphere since their spatial availability is poor over remote areas (e.g., over oceans) and no vertical information is provided. Vertical mixing ratio profiles with high accuracy are delivered by means of radiosondes. Radiosondes (RSs) are a common and reliable in situ technique but they also lack temporal and spatial coverage as the number of sites is rather low and most of them perform very few soundings per day. Furthermore, the wind-driven drifting of the device can be misleading in terms of geographical location of the vertical information. In contrast, remote sensing techniques such as microwave radiometers (England et al., 1992; Reagan et al., 1995), differential absorption lidars (DIALs) (Bösenberg, 1998), photometers (Barreto et al., 2013) and Raman lidars (Whiteman et al., 1992; Ferrare et al., 1995; Turner et al., 2002; Whiteman, 2003; Leblanc et al., 2012; Navas-Guzmán et al., 2014; Foth et al., 2015) have been successfully adopted in water vapor studies. While microwave radiometers and photometers can accurately deliver the total precipitable water vapor (TPW), lidars (DIAL and Raman) are the only instruments available for high temporal and vertical resolution of continuous WVMR measurements. DIALs are able to provide accurate high-resolution profiles of water vapor, but their complex laser transmitter setup makes their WVMR automatization a difficult task. Raman lidars use a simpler setup than DIALs, although the majority of them are limited to nighttime performance due to collection of the intense daytime background light by the weak Raman-shifted channels. Nonetheless, Foth et al. (2017) proposed a methodology to retrieve water vapor mixing ratios during daytime by using a microwave radiometer and the Raman lidar profiles.

In the present paper, we calibrate Raman lidar WVMR profiles using in situ, satellite and model data. We have used RSs, retrievals from AIRS/AMSU instruments on board of Aqua satellite (Parkinson, 2003) and modeled WVMR from ALADIN/HIRLAM NWP model (Seity et al., 2011; Bengtsson et al., 2017). Our dataset concentrates on three field campaigns conducted during 2014 and 2015. Within this time frame, 723 radio soundings were performed in total, but not all of them were suitable for direct comparison with the lidar due to the daytime limitation or occurrence of low-level clouds. Furthermore, we derive the seasonal variation of WVMR at Kuopio site where the lidar instrument is permanently located, both from the lidar and model, thereby validating the accuracy of the model and the capabilities of the lidar under the demanding low-water-content conditions of Finland.

The outline of the paper is as follows. First, we give a short description of the instruments/model (Sect. 2) followed by the methodology used to calculate/extract the WVMR profiles from the various sources (Sect. 3). An overview of the existing calibration methods for the lidar is also given in Sect. 3. The calibration factors from the various reference instruments are calculated in Sect. 4. Section 4 also includes comparisons between the RSs and the various instruments and the model, under cloud-free conditions. Furthermore, Sect. 5 presents the seasonal variability of the WVMR at the Kuopio site in Finland and the seasonal discrepancies between the lidar and the model based on a 4-year period of lidar and model data. Our summary and concluding remarks are given in Sect. 6.

The site locations where the three field campaigns took place cover Finland from the southwest to the north: Hyytiälä (61.84∘ N, 24.29∘ E; 179 ma.s.l.) from 1 April to 29 September 2014, Kuopio (62.73∘ N, 27.54∘ E; 190 ma.s.l.) from 11 to 29 May 2015, and Pallas (67.99∘ N, 24.24∘ E; 345 ma.s.l.) from 22 September to 5 December 2015. Further information on the site locations and available instrumentation can be found in Hatakka et al. (2003) and Hirsikko et al. (2014).

The WVMR is defined as the ratio of the mass of water vapor to the mass of dry air in a given volume, calculated as wr=MWH2OMWDryAirNH2OrNDryAirr∼0.78MWH2OMWDryAirNH2OrNN2r, where MWDryAir and MWN2 are the molecular weights of water vapor and nitrogen. NH2Or, NN2r and NDryAirr denote the molecular number densities of the two atmospheric gases and dry air at altitude r. The 0.78 value stands for the fractional volume of nitrogen in the atmosphere. The latter expression is utilized in the water vapor Raman lidar technique, and is proportional to the water vapor mixing ratio.

For the calibration of the lidar signal, cloud-free nighttime atmosphere is considered. The lidar data were averaged over 30 min centered on the RS launch time. The RS vertical resolution was interpolated to the lidar's grid, where a 90 m unweighted sliding average was applied to smooth the signal. Only lidar signals between 0.5 and 3.5 km were retained for the calibration. This height limitation was used to minimize the inaccuracies in the WVMR values due to possibly different overlaps between the channels used for the water vapor calculation (see Sect. 4.1) and major drifts in the RS at higher altitudes. In unstable atmospheric conditions these two could result in measuring very different atmospheric layers showing temporal and spatial mismatching (Brocard et al., 2013) (see also Sect. 4.1). However, the upper height limit was lowered from 3.5 km when lidar signal-to-noise ratio (SNR) values were lower than 2. For the satellite data we performed a trajectory analysis, forward or backward depending on the overpass and the RS launch time, for each of the satellite footprints selecting the profile whose trajectory endpoint was closest to the site. This method was used because it produced better agreement between the satellite data and the radio soundings than the usage of the closest satellite pixel to the site. The trajectory analysis was performed using the HYSPLIT model (HYbrid Single Particle Lagrangian Integrated Trajectory; Draxler and Hess, 1998; Stein et al., 2015). Satellite data marked with quality assurance flag 2 are omitted (AIRS version 6, level 2; Kahn et al., 2012) while the maximum overpass difference between the on-site RS and the satellite was set to 6 h. For NWP model data the profile closest in time was used. In each campaign model's grid box was centered on the lidar site. A fixed intercept of 0 was used in the regression analysis in order not to introduce positive WVMR at heights where none should exist.

An example case (23 May 2015) demonstrating the calibration method with all the available reference instruments and the model is shown in Fig. 1. The satellite overpass is from 00:28 UTC on 24 May 2015. The closest satellite footprint was 14 km away from the site (white dot) and the selected WVMR for this case was based on the trajectory analysis which originated at the green dot shown in Fig. 1a. Figure 1b shows the range corrected signals at 1064 nm along with the RS launch times marked as vertical red lines. The RS was launched at 22:00 UTC, and averaged lidar signals between 21:45 and 22:15 UTC were used (white rectangle). Applying the slope method, we obtained calibration factors of 17.87±0.17, 17.43±0.13 and 17.44±0.16 gkg-1 for RS, AIRS/AMSU and ALADIN/HIRLAM, respectively (Fig. 1c). The uncertainty reported here indicates the standard error of the slope.

From the 630 soundings in Hyytiälä, 66 in Kuopio and 27 in Pallas, 23, 10 and 5 RSs are suitable for the calibration procedure based on background light and cloud conditions. Due to the high geographical latitude and the time of the campaigns in Hyytiälä and Kuopio (between May and September), Finland's background sky light is too intense for this technique to retrieve water vapor profiles from lidar observations. For example, over the 18-day campaign in Kuopio, daytime increased by 2 h. For all suitable aforementioned calibration cases factors from the various instruments/model were calculated and are summarized in Table 1 for Hyytiälä and Pallas and in Table 2 for Kuopio. The overall calibration factor computed has an associated uncertainty of <1 % for the RS launched on site and 2.8 % for the nearest RS (only for Kuopio) corresponding to a mean factor of 17.46±0.13 and 16.94±0.48 gkg-1, respectively. The uncertainty for the satellite and the model fluctuated at 7.4 and 3.9 % with a mean factor of 18.53±1.37 and 17.78±0.69 gkg-1, respectively. These accuracies comply with previous observational studies (Ferrare et al., 1995; Navas-Guzmán et al., 201l) yet we are aware of none publications regarding the calibration of lidar WVMR signals with the use of a model. We observe that the satellite-derived calibration factor diverges from the RS-derived one by about 6 %. This deviation is interpreted, depending on the water vapor amount, to as high as 0.4 gkg-1 offset for mixing ratios of about 8 and 0.1 gkg-1 offset for drier conditions when calibrating with the satellite. Such bias can have a large impact, for example in changing environments such as the tropopause, where the radiative forcing of surface climate is being calculated (Leblanc et al., 2012; Müller et al., 2016).

It can be seen in Tables 1 and 2 that there are a couple of cases where the satellite or/and the model have underestimated/overestimated the mixing ratios – hence the calibration factor. Consequently, individual cases can have a large effect on the calculated calibration factor. A more robust factor was retrieved by applying the regression method to all cases together (Fig. 2). Discrepancies are much lower when using this method as all techniques converge to the value calculated with the on-site RS, which is presumably the closest one to the true factor. If available, we suggest using this technique when calibrating lidar WVMR signals. We should mention here that the calculation of an overall factor is possible since in all three campaigns the lidar setup had the same configuration for the channels relating to water vapor.

In this study we employed water vapor profiles from radio soundings, retrievals from AIRS/AMSU instruments on board of Aqua satellite and for the first time the ALADIN/HIRLAM model to calibrate a Raman lidar with water vapor capabilities. The uncertainty of the calibration factor from the radio soundings is <1 and 2.8 % for the on-site and nearest RS, respectively. The calibration factor derived from the satellite had a discrepancy of 7.4 %, and from the model a 3.9 %. These results lay well within previous studies using the most common calibration technique – that of the regression method. If possible, we urge future lidar users to calculate the calibration factor applying a combined regression line to all their cases simultaneously, since individual cases can have an impact on the derived calibration factor. In our study, this impact was translated into the aforementioned averaged percentages, yet the goal of the lidar calibration is to determine a proper factor as close as possible to the real factor (assuming that of the RS) and this is better obtained with the proposed method.

For the period being studied, intercomparisons between the on-site RS and the rest showed that the model and the nearest available soundings are an effective alternative when no soundings are on site, resulting in the smallest deviations after the lidar. We should note here that the model represents the shape of atmospheric WVMR profiles more accurately in the lowest few kilometers than the nearest RS which is located 100 km away from the lidar site. However, care should be taken when using the nearest RS, as non-stable atmospheric conditions between the two sites may lead to inadequate vertical representation of water vapor; for example, when there is sufficient time lag between the RS and the lidar measurement, especially when PBL activity is not constant – e.g., between day and night. The highest discrepancies were observed when comparing on-site RS and satellite WVMR. These discrepancies resulted from the larger spatial and poorer vertical resolution of the satellite data. A maximum 6 h overpass difference between the on-site measurements and the satellite was allowed in this study. This time limit is usually valid for the nearest RS where the 18:00 UTC RS is used, for example, with the 00:00 UTC lidar measurements. As has been shown in Sect. 4.1, the 6 h delay is acceptable for nighttime comparisons but not when one of the observations is during the day. We also encourage future users to apply the trajectory method when selecting satellite WVMR profiles and not the closest footprint to that of the site of interest. We found better agreement between the RS, the lidar and the satellite when applying this technique.

In addition, 4-year water vapor data from the lidar and the model were adopted to study the water vapor mixing ratio seasonality at the Kuopio site in eastern Finland. The analysis showed that three humidity states exist in Kuopio: a wet one during summer months where water vapor values of 5.52±1.01 gkg-1 were observed within the lowest 2 km, a dry state during winter months with concentrations of 1.17±0.42 gkg-1 and a transition state during spring and autumn with intermediate values. Similar WV vertical structure to that of lidar was found when using modeled WVMR from ALADIN/HIRLAM. In general, the model simulates correctly WVMR for each season although some discrepancies were observed which are attributed partly to a combination of sparser vertical modeled bins and larger averaged area than that of the lidar. There seems to be a negative bias between the model and the lidar at higher altitudes.