Detection of land-surface-induced atmospheric water vapor patterns

Finding observational evidence of land surface and atmosphere interactions is crucial for understanding the spatial and temporal evolution of the boundary layer, as well as for model evaluation, and in particular for large-eddy simulation (LES) models. In this study, the influence of a heterogeneous land surface on the spatial distribution of atmospheric water vapor is assessed. Ground-based remote sensing measurements from a scanning microwave radiometer (MWR) are used in a long-term study over 6 years to characterize spatial heterogeneities in integrated water vapor (IWV) during clear-sky conditions at the Jülich ObservatorY for Cloud Evolution (JOYCE). The resulting deviations from the mean of the scans reveal a seasonand direction-dependent IWV that is visible throughout the day. Comparisons with a satellite-derived spatial IWV distribution show good agreement for a selection of satellite overpasses during convective situations but no clear seasonal signal. With the help of a land use type classification and information on the topography, the main types of regions with a positive IWV deviation were determined to be agricultural fields and nearby open pit mines. Negative deviations occurred mainly above elevated forests and urban areas. In addition, high-resolution large-eddy simulations (LESs) are used to investigate changes in the water vapor and cloud fields for an altered land use input.

considered (50 • , 160 • , 180 • , 260 • ) and missing values are filled using a linear interpolation. For all scans, the derived values for LWP, IWV and q are air-mass corrected to account for the slant angle of the scanning MWR.

Doppler lidar and boundary-layer classification
As a pulsed lidar system, the Halo Photonics Streamline Doppler lidar (Pearson et al., 2009) provides range-resolved profile 5 measurements of radial Doppler velocity and backscattered signal. With a wavelength of 1.5 µm (near-IR) the instrument is sensitive to the backscatter of aerosols and clouds and is able to scan the full hemisphere. The maximum detectable range depends on the presence of atmospheric particles and the lowest reliable range is at 105 m. At JOYCE the system is set to a range resolution of 30 m and performs plan position indicator scans every 15 min to estimate wind speed and direction profiles based on the velocity-azimuth display (VAD) method using 36 beams at 75 • elevation. In addition the Doppler beam swing 10 (DBS) technique with three beams and range height indicator scans are scheduled every 5 min and 30 min, respectively. The remaining time, the instrument is staring zenith to derive the vertical velocity with high temporal resolution (1 s).
To study land surface atmosphere exchange processes it is crucial to know the turbulent state of the boundary-layer. Therefore an objective classification of the mixing sources presented by Manninen et al. (2018) is utilized to describe the turbulence 15 characteristics during MWR scans at JOYCE. The method is based on the combination of multiple Doppler lidar quantities including the dissipation rate of turbulent kinetic energy (TKE) derived from vertically pointing observations using the method presented in O' Connor et al. (2010). The TKE dissipation rate is based on the variance of the observed mean Doppler velocity and allows for a threshold based estimation of the convective boundary-layer (CBL) height. 20 The passive, imaging Moderate Resolution Imaging Spectroradiometer (MODIS) measures in 36 spectral bands ranging from 0.4 µm to 14.4 µm. Two MODIS instruments are currently airborne on NASA's sun-synchronous near-polar-orbiting Earth Observing System Terra and Aqua satellites. A full coverage of the globe is achieved in 1-2 days with an orbit height of 705 km and a scan rate of 20.3 rpm. The swath dimension of MODIS is 2330 km (cross track) and 10 km (along track at nadir). Within the 36 spectral bands, five channels in the 0.8-1.3 µm near-infrared spectral region can be used for water vapor remote sensing 25 (Gao and Kaufman, 2003). For IWV estimates the Level-2 (Collection 6.1) near-infrared retrieval (MODIS-NIR) with a 1 km spatial resolution is chosen. The retrieval by Gao and Kaufman (2003) is based on three channels at 0.936 µm, 0.940 µm and 0.905 µm for the water vapor absorption and at 0.865 µm and 1.24 µm to correct for atmospheric gaseous absorption. In order to derive the total vertical amount of water vapor, the reflected NIR solar radiation in the water vapor absorption channel is compared to the window channels yielding the atmospheric water vapor transmittance. The amount of water vapor is 30 then obtained from look-up tables derived from a line-by-line atmospheric transmittance code. Reliable estimates of the water vapor total column amount over land areas can only be inferred during daytime and for cloud free regions. Typical errors of the MODIS-NIR water vapor product range between 5-10%. Here, a height correction similar to Steinke et al. (2015) of the retrieved values is performed due to the variations of the horizontal and height distance to JOYCE per flight track of MODIS.

MODIS IWV
The height difference is corrected by assuming an exponential decrease of the humidity profile and by using the water vapor density obtained from measurements of temperature, humidity and pressure of a weather sensor attached to the MWR and the topography with a 200 m horizontal resolution. Furthermore, the IWV product was resampled to 100 m for calculating the mean values of several overpasses.

ERA5 data products
To distinguish between local influences and large scale features regarding the observed spatial pattern of IWV deviations, the reanalysis products of ERA5 with a 31 km horizontal resolution are analyzed (Copernicus Climate Change Service (C3S), 2017). Besides the u and v wind components at different pressure levels (1000 hPa, 700 hPa), also the direction of the IWV 10 transport (IWVT, in degree) is considered at a 3 h temporal resolution for the closest point to JOYCE. The vertical integral of water vapor flux, used to derive IWVT, is calculated utilizing the specific humidity and winds on model levels. The ERA5 IWV is selected at the closest output time to the MWR scans.

ICON-LEM
As a state-of-the-art atmospheric modeling system, the ICOsahedral Non-hydrostatic model ICON (Zängl et al., 2015)  advancing Climate Prediction (HD(CP) 2 ) project for improving moist processes in climate prediction models (Heinze et al., 2017). In this study, the ICON-LEM simulations are used to provide a spatial representation of the IWV field to compare with the measurements obtained from the scanning MWR and the MODIS-NIR water vapor product around JOYCE.

Land use classification and measurement site description
To be able to link atmospheric water vapor measurements with land surface properties, spatial land use information is needed. This is addressed by using a remote sensing-based regional crop map (Waldhoff et al., 2017) that was applied to a study area in Western Germany including the surrounding area of JOYCE. In this method, supervised multi-temporal remote sensing data of Sentinel-2, ancillary information and expert-knowledge on crops are combined in a Multi-Data Approach (MDA). The 5 classification is therefore able to differentiate between 44 vegetated, urban and water areas with a spatial resolution of 15 m.
The detailed and highly resolved classification is used to identify areas with a predominant land use type. Therefore the classified types are condensed into five main types, in particular agricultural areas and grass land, bare ground, urban areas, deciduous forest and water. These five groups are expected to have a significantly different behavior in terms of transpiration 10 and/or evaporation and therefore might cause atmospheric water vapor patterns that can be distinguished and related to the appropriate type. In Fig. 1 the simplified land use classification of a 12x13 km area centered around JOYCE is shown. The The artificially created pit mine dump hill Sophienhöhe is located in the northeast direction, which is up to 200 m higher than JOYCE and covered mainly by a deciduous forest. In the northern and southeastern part of the selected domain mostly agricultural sites can be identified. The main crop types between April and September are winter wheat and sugar beet, but also maize and potato. A common crop rotation is a two year cycle of sugar beet to winter wheat (Waldhoff et al., 2017). The 5 southwestern parts are mostly grass lands surrounding the Rur River, with its valley going from southeast to northwest. The pit mines (bare ground) with depressions down to 300 m below JOYCE are located to the east and southwest.

Long-term MWR scans and boundary-layer characteristics
In order to find patterns in the long-term water vapor scans at JOYCE, that can be related to local land surface characteristics, 10 the MWR scans are evaluated during meteorological conditions that are favorable for strong land surface atmosphere interactions. This excludes overcast situations and large scale advection of moist or dry air. The cloud detection is obtained by using the 31.4 GHz channel, which is within an atmospheric window. The signal from this channel is dominated by the presence of liquid water in case of clouds appearing in the instrument's field of view. During a single scan the maximum difference of the measured 31.4 GHz brightness temperature for each azimuth direction and the mean of the whole scan must be below 2 K, 15 since liquid water clouds are expected to cause a much higher difference. Furthermore the air-mass corrected LWP from the statistical retrieval needs to be below 20 g m −2 , which is in the order of the retrieval uncertainty. To avoid scenes with large scale advection of moist or dry air, the difference between the maximum and minimum IWV z within one hour around the scan needs to be smaller than 2 kg m −2 , which is above the instrument sensitivity. These requirements need to be fulfilled for at least three consecutive scans. The first and last scan of each sequence is neglected. The choice of the thresholds showed to be a 20 good trade-off between excluding apparent cloudy situations, but still allowing a sufficient number of scans to generate a large data sample. Only the months between April and September between 2012-2018 are regarded, since the highest diurnal IWV variability is observed between spring and autumn at JOYCE (Löhnert et al., 2015) and the influence from the land surface is expected to be larger. During the observational period 316 days with in total 7261 single scans are selected with a mean IWV z of 18.00 ± 6.37 kg m −2 measured in a 1 h window around the scans.

25
Instead of using the total slant column IWV, the humidity profile is integrated up to a scaling height of 2.5 km (hereafter: IWV 2.5 ) for an analysis of the lower tropospheric water vapor patterns. This height, where the humidity profile drops below 1/e, was found as mean value for the zenith MWR measurements 1 h around each of the selected scans. A similar scaling height was also found using satellite data (Simon and Joshi, 1994). For all scans, the mean value per scan is subtracted to investigate  in number of cases during daytime is due to the formation of convective clouds, since overcast situations would influence the number of cases independent of the time of the day. Also the mean standard deviation for each scan increases from 1.1% to 1.94% during daytime indicating the influence of convective activity, which is shown by high TKE dissipation rates and a mean CBL height up to 1.28 km ( Fig. 2(b)). Note that these deviations are median values to detect the long-term pattern and that single scan deviations from the mean can get over 5%. Also the IWV standard deviation from the zenith MWR measurements 5 in Fig. 2(b) reveals a diurnal cycle during this measurement period of late spring until early autumn, which is in agreement with the seasonal statistics derived in Löhnert et al. (2015). While the IWV standard deviation follows the rate of the CBL height development in the morning hours, an abrupt decrease is only evident in the turbulence measurements in the afternoon transition period. This suggests that water vapor is mixed into the upper layers of the atmosphere during daytime and is still present in the residual layer throughout the night and would explain the consistent pattern after sunset in Fig. 2(a).

10
Separating all cases according to the low-level wind direction from the Doppler lidar, no directional dependence is found.
For assessing the impact of the large scale water vapor transport, the ERA5 reanalysis product is used.  Fig. 2(a)). The wind direction ranges from a southerly flow during night to an east to north direction during the day corresponding to fair weather situations and anticyclonic flow at this site. The wind direction turns clockwise with height for the ERA5 product and the Doppler lidar observations, but stays relatively constant within the CBL as there is no large difference between DWL s and the Doppler lidar wind direction at 1005 m (DWL b ) between 10-15 UTC.
The wind direction in the free troposphere at 700 hPa shows no significant diurnal cycle. The same applies to the IWVT, that corresponds to the westerly wind direction at 700 hPa, showing the west-wind-zone transport of humid air at the mid-latitudes, which might contribute to the IWV 2.5 deviation scan pattern, especially during night. But at midday and early afternoon the positive deviations increase and shift to the southeast. Despite the fact, that the ERA5 IWV shows a diurnal cycle, this shift can 5 not be seen in the IWVT, suggesting that also local influences contribute to the observed IWV signal. This is further analyzed in the next section.

Comparison of daytime MWR and MODIS derived IWV deviations
For a comparison with an independent IWV measurement, the MWR results are compared to the MODIS-NIR derived IWV around JOYCE. For a fair comparison of the column amount of water vapor from MODIS to the path-integrated water vapor 10 observations from the MWR scans, a virtual MWR scan is derived from the MODIS observations. Therefore, the total IWV is distributed to an absolute humidity profile assuming a linear decrease by 20% in the CBL and an exponential decrease above similar to Schween et al. (2011). The CBL height is determined from the boundary-layer classification (Manninen et al., 2018) available for each overpass. The CBL height is assumed to be constant in the area of interest, as well as the 1/e height for the exponential decrease, which is calculated from the MWR humidity profile. In this way a virtual scan corresponding to the 15 MWR scan configuration can be performed around JOYCE where the amount of water vapor is integrated for each beam. Only overpasses without missing data due to the MODIS quality checks are considered. A circular area with a radius of 4.3 km is chosen. This radius corresponds to the distance, where the beam at 30 • reaches the water vapor scaling height of 2.5 km, which was found on average in the zenith MWR humidity measurements. 20 As an additional comparison of MWR and MODIS, the IWV z measurements of the MWR (IWV z,MWR ) and the MODIS mean total column amount 1 km around JOYCE (IWV z,MODIS ) are compared ( Fig. 3(a)). The zenith IWV values are highly correlated (0.98) with a RMSE of 2.79 kg m −2 , which is about 1 kg m −2 higher than found in Steinke et al. (2015). This discrepancy is probably caused by a greater IWV variability shown in Fig. 2(b). For larger IWV values, the MODIS observations tend to an overestimation. For the 61 MODIS overpasses occurring between 9-13 UTC, the corresponding MWR scans 25 within 1 h around the overpass are selected (172 scans) and for both data sets the mean IWV deviation from the (virtual) scans are calculated (Fig. 3(b)). In general, the relative deviations from the MODIS virtual scans are higher by a factor of about 3.
With both observations, a noticeable negative deviation between 270 • -60 • is visible, but also the agreement in the location of the maximum positive deviations around 225 • is evident. This area shows a high fraction of crop and grassland and one of the pit mines, whereas less water vapor seems to be present in the vicinity of the urban area and forested hill. Regarding  to the topography and a higher roughness length at the forested hill can cause decreased water vapor fluxes into the atmosphere.

LES case study analysis
The influence of the land use type on the atmospheric water vapor pattern is further investigated in a case study (25 July 2012) by means of a large-eddy simulation using the ICON-LEM model. On this day, with a northwesterly wind direction, no clouds 10 are observed until 12:30 UTC and the timings of the MODIS overpasses are 10:00 UTC and 11:40 UTC. In this time interval, four MWR scans are performed and the CBL height determined by the Doppler lidar increases from 885 m to 1305 m. In the first ICON-LEM simulation (ICON1) using the simplified GLOBCOVER land use data ( Fig. 4(a)), the model boundary-layer height reaches these heights about one hour later than in the observations. In order to compare a similar state of the boundarylayer in terms of convection, the analysis time for the simulations is shifted by one hour. Again, like for the MODIS data,  In a second simulation (ICON2), the land use types are changed according to Fig. 4(b) (crop/grass to bare ground, bare 10 ground to water, urban to forest, forest to crop/grass and water to urban). In this way, a significant reconstruction in the spatial distribution of the land use types is achieved without changing the scale of heterogeneity. The comparison in the IWV deviation between ICON1 and ICON2 shows higher IWV values in the northwest for ICON1 and in the southwest for ICON2. To elaborate the details of this difference, the spatial fields of height and time averaged vertical velocity and mean IWV are analyzed (Fig. 6). The averaging domain is the same as shown in Fig. 4   regarding the vertical velocity, which is also evident in Fig. 6(a). In addition, the wind is lifted by the hill and a downdraft above the hill can be seen. This was already discussed in Marke et al. (2018) and might explain parts of the lower water vapor flux discussed in Sect. 3.1. In the ICON2 simulation, with a larger fraction of bare ground, the differences in surface properties and the size of the heterogeneous land use patches seem to be large enough to cause a secondary circulation (Garcia-Carreras et al., 2011;van Heerwaarden et al., 2014;Eder et al., 2015)  The change in the circulation pattern due to the different land use input can explain the differences in IWV deviation (Fig. 5), since more water vapor is transported in the enhanced updraft streak (Fig. 6(d)). The strong circulation effect might also reduce the influence of other humidity sources, like the introduced water bodies at the positions of the pit mines in ICON2. The formation of convective clouds in the simulations is affected as well. Whereas in ICON1 the cloud distribution is rather patchy, clouds form only along the high IWV region in ICON2. Also the mean cloud cover of 8.55% in ICON1 compared to 10.55% 5 in ICON2 is closer to the observed maximum cloud cover of 6% determined by a total sky imager at JOYCE. The maximum integrated cloud water content of these clouds is 36.96 g m −2 (ICON1) and 5.61 g m −2 (ICON2). The specific cloud water content in ICON1 is significantly higher and the clouds grow taller compared to the ICON2 simulation (Fig. 7), which is connected to a higher latent heat flux in ICON1 due to more vegetated areas. On the other side the boundary-layer grows deeper (by about 30 m) in ICON2 because of the increased sensible heat flux caused by a higher fraction of bare ground.

10
This shows the connections of the different Earth system compartments and stresses the importance of further monitoring and modeling the interactions between the land use, water vapor and the transition to clouds.