Accurate Estimation of Atmospheric Water Vapor Using ... - IEEE Xplore

3764

IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. 53, NO. 7, JULY 2015

Accurate Estimation of Atmospheric Water Vapor Using GNSS Observations and Surface Meteorological Data F. Alshawaf, T. Fuhrmann, A. Knöpfler, X. Luo, M. Mayer, S. Hinz, and B. Heck

Abstract—Remote sensing data have been increasingly used to measure the content of water vapor in the atmosphere and to characterize its temporal and spatial variations. In this paper, we use observations from Global Navigation Satellite System(s) (GNSS) to estimate time series of precipitable water vapor (PWV) by applying the technique of precise point positioning. For an accurate quantification of the absolute PWV, it is necessary to combine the GNSS observations with meteorological data measured directly or inferred at the GNSS site. In addition, measurements of the surface temperature are used to calculate the empirical constant required to convert the GNSS-based delay into water vapor. Our results show strong agreement between the total precipitable water estimated based on GNSS observations and that measured by the sensor MEdium Resolution Imaging Spectrometer with a mean RMS value of 0.98 mm. In a similar way, we compared the GNSS-based total PWV estimates with those produced by the Weather Research and Forecasting (WRF) Modeling System. We found that the WRF model simulations agree well with the GNSS estimates with a mean RMS value of 0.97 mm. Index Terms—Atmospheric sounding, Global Navigation Satellite System(s) (GNSS), MEdium Resolution Imaging Spectrometer (MERIS), precipitable water vapor (PWV), Weather Research and Forecasting (WRF).

I. I NTRODUCTION

T

HE performance of using Global Navigation Satellite System(s) (GNSS) for highly precise positioning or measuring Earth’s surface displacement is limited by the successful correction of the time delay produced by the atmospheric water vapor. Although the contribution of water vapor does not exceed one-tenth of the total neutrospheric delay, this error source is difficult to correct due to its high variations in time and space. In this paper, we want to exploit the sensitivity of the GNSS

Manuscript received May 9, 2014; revised September 29, 2014; accepted December 3, 2014. F. Alshawaf and S. Hinz are with the Institute of Photogrammetry and Remote Sensing, Karlsruhe Institute of Technology, 76131 Karlsruhe, Germany (e-mail: [email protected]; [email protected]). T. Fuhrmann, A. Knöpfler, M. Mayer, and B. Heck are with the Geodetic Institute, Karlsruhe Institute of Technology, 76131 Karlsruhe, Germany (e-mail: [email protected]; [email protected]; michael.mayer@ kit.edu; [email protected]). X. Luo was with the Geodetic Institute, Karlsruhe Institute of Technology, 76131 Karlsruhe, Germany. He is now with the GNSS Product Management Group, 9435 Heerbrugg, Switzerland (e-mail: [email protected]). Color versions of one or more of the figures in this paper are available online at http://ieeexplore.ieee.org. Digital Object Identifier 10.1109/TGRS.2014.2382713

signals for measuring the integrated content of water vapor in the atmosphere. We will describe a new approach for precise water vapor mapping by combining GNSS and meteorological data, and we will evaluate the performance. There are a variety of methods that have been used to measure the integrated water vapor content in the atmosphere, either from space or ground. Meteorological instruments such as radiosondes or water vapor radiometers provide water vapor measurements that are limited in the temporal or spatial resolution or in both. Near-infrared sensors such as the MEdium Resolution Imaging Spectrometer (MERIS) only produce continuous water vapor maps under cloud-free weather conditions. Therefore, observations from microwave remote sensing systems were exploited for water vapor mapping. Since the 1990s, GNSS observations have been considered as an efficient tool for atmospheric sounding [1]. Since then, numerous studies have been carried out for estimating water vapor using GNSS data, see, for example, [2], [3]. On the other hand, interferometric synthetic aperture radar (InSAR) data have recently been investigated to derive atmospheric water vapor maps of high spatial resolution [4]. In addition to observational studies, progress has also been made in numerical weather prediction models such as the Weather Research and Forecasting (WRF) modeling system for simulating atmospheric parameters. However, these models are still limited by the settings of the boundary layer and the model initial conditions [5]. In this paper, we present a new method to precisely estimate the precipitable water vapor (PWV) using a combination of GNSS data and surface meteorological observations of air pressure, temperature, and relative humidity. The main advantage of GNSS is that they offer rich information about the temporal variability of water vapor content due to the high sampling rate of the GNSS measurements. Using the technique of precise point positioning (PPP), GNSS provides accurate estimates of the absolute PWV at GNSS sites. We improve the accuracy of the PWV estimates to a submillimeter level by combining site-specific GNSS data with meteorological measurements, which are directly measured or interpolated at the GNSS site. The GNSS-based PWV estimates can be used in tomographic approaches to monitor the Earth’s atmosphere for purposes of weather forecasting and climate research. With the increasing number of satellite constellations (e.g., GPS, Galileo, and GLONASS), the spatial coverage of the atmosphere with slant PWV estimates is highly improved by combining observations from multiple systems [6]. This information about water vapor,

0196-2892 © 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

ALSHAWAF et al.: ESTIMATION OF WATER VAPOR USING GNSS OBSERVATIONS AND METEOROLOGICAL DATA

3765

sites. Most of the German sites are operated by the Satellite Positioning Service (SAPOS) of the German State Surveys of Baden–Württemberg and Rheinland–Pfalz. Aside from the German sites, we also used data from French data providers (e.g., RENAG, RGP). The GNSS measurements are available since 2002 at a sampling rate of 30 s [7]. In this study, only GPS data were considered, whereas GLONASS data will be included in the future. The meteorological measurements were collected at four sites located close to the GNSS sites (see Fig. 1). Moreover, meteorological data are received from the model WRF, including maps of the PWV. Additionally, we used PWV maps from MERIS for evaluating our results. MERIS is a passive imaging spectrometer located on board the Envisat platform. It measures the solar radiation reflected from the Earth’s surface or clouds. The ratio of the radiance values measured at the channels 14 and 15, located at 885 and 900 nm, respectively, are used to determine the vertical PWV content in the neutrosphere [8]. MERIS provides maps of the PWV at a spatial resolution of 260 m × 290 m (full-resolution mode). Under cloud cover, the PWV measured by MERIS represents only the water vapor existing between the sensor and the clouds top; therefore, MERIS does not provide useful PWV information under cloudy weather conditions. Fig. 1. Study region in the Upper Rhine Graben in Germany. The red squares indicate the locations of the GNSS sites, whereas the green circles indicate the locations of the meteorological sites. The black box illustrates the grid over which the WRF model data are simulated.

either integrated or tomographic, can be used to adjust the initial and boundary conditions in atmospheric prediction models. Since these models often overestimate the PWV, particularly under high atmospheric advection, the GNSS-based PWV estimates can be useful for suppressing the bias in the model. The work presented in this paper is carried out within the frame of a project initiated for constructing PWV maps by combining GNSS and InSAR data. GNSS and InSAR have complementary properties; GNSS provides estimates of the absolute PWV at sparse sites, whereas InSAR produces maps of partial PWV at a high spatial resolution. Hence, by properly combining of both data sets, we should be able to accurately derive maps of the absolute PWV at a high spatial resolution. This paper is organized as follows. Section II presents the study region and the datasets used in the study. In Section III, the strategy to estimate the PWV from GNSS data is presented. The conversion factor required for transforming the wet delay into PWV using measurements of surface temperature is also described in this section. Validations of the PWV estimates against MERIS and WRF data are presented in Section IV. The approaches used for spatially interpolating temperature and pressure measurements are presented in Section V. The conclusion is presented in Section VI. II. S TUDY A REA A ND DATASETS This study is carried out using data collected in the Upper Rhine Graben area in the southwest of Germany and the northeast of France, as shown in Fig. 1. This region is well covered by a relatively dense network of permanent GNSS

III. E STIMATING PWV Here, we introduce the methods for obtaining the path delay and PWV from GNSS data. The Saastamoinen a priori empirical model is described in Section III-A. In Section III-B, the method for estimating the site-specific neutrospheric parameters during the GNSS data processing and the inclusion of meteorological measurements to derive the PWV are described. In Section III-C, the formula for converting the wet path delay into PWV is discussed. A. Path Delay From Empirical Models Based on the empirical neutrospheric a priori models, such as the model of Saastamoinen [9], it is possible to calculate the neutrospheric delay at any point on the ground based on the air pressure, temperature, and relative humidity. The models for the Zenith Dry Delay (ZDD) and the Zenith Wet Delay (ZWD) are as follows: = 0.002277D(P − 0.155471e) Δρsaas d 1255 saas + 0.205471 e Δρw = 0.002277D T

(1) (2)

where Δρsaas denotes the ZDD, caused by dry gases, and d denotes the ZWD, due to water vapor. P and e are the Δρsaas w air pressure and partial pressure of water vapor in hectopascals, respectively. T is the air temperature in kelvins. Factor D depends on the altitude and the geographical latitude of the point at which the delay is computed. If reliable meteorological observations are available at the GNSS site, the ZDD can accurately be calculated from the model in (1). In contrast, the ZWD model in (2) is often inaccurate [10]. According to [11],

3766


the partial pressure e in (1) and (2) can be calculated from the following empirical formula: RH exp(−37.2465 + 0.2131665T − 0.000256908T 2 ) e= 100 (3) with T in kelvins. The relative humidity (RH) in % is either measured or calculated from RH = 100 − 5(T − Td )

(4)

Td is the measured dew point temperature. Note that e can also be calculated directly given the dew point temperature, as described in [12]. B. Path Delay From GNSS Data The total neutrospheric delay can be subdivided into azimuthally isotropic and azimuthally anisotropic components. The anisotropy in the neutrosphere originates from first, the natural tilting of the neutrospheric layer, which means that the normal direction of the neutrosphere (i.e., the direction with the minimum neutrospheric delay) can slightly mismatch the topocentric zenith. Second, anisotropic effects can be caused by local turbulent or dynamic phenomena in the atmosphere, which only affect the signals travelling through. Here, we describe how these components are determined during the GNSS data processing, and how the PWV is extracted from the total neutrospheric delay. The GNSS data were processed using the Bernese GNSS Software Version 5.0 (BSW5) [13]. We applied the technique of PPP, where a linear combination of the GPS phase observations in the L1 (1.57542 GHz) and L2 (1.2276 GHz) frequency bands is computed. By applying the ionospheric-free linear combination, the first-order ionospheric delay is eliminated from the observations, and we can estimate the absolute (total) neutrospheric delay. Precise satellite orbits and clock information, both provided by the Center of Orbit Determination Europe (CODE) (see [14, Tab. 3]), are used to obtain the best quality for the estimated parameters. Details of the data processing are presented in [14]. The steps for estimating the wet delay by GNSS data processing are illustrated in Fig. 2 and are explained in the following. During the GNSS data processing in the BSW5, the neutrospheric path delay is estimated from the GNSS observations by least squares adjustment. The zenith site-specific neutrospheric parameter, i.e., Δρz , is obtained using the following model of the azimuthally isotropic slant path delay Δρgnss s,iso = Δρz, mod · M Fd + Δρz · M Fw

(5)

where Δρgnss s,iso is the azimuthally isotropic slant path delay. The settings of the software are selected such that Δρz, mod gives the delay in the Saastamoinen a priori model, and M Fd is the dry Niell mapping function [15]. The a priori model requires input values of P , T, and e, which are, in the BSW5, derived from the standard atmosphere [16]. M Fw is the wet Niell mapping function. The azimuthally isotropic zenith total delay, i.e., Δρgnss z,iso , corresponding to (5) is Δρgnss z,iso = Δρz, mod + Δρz .

(6)

Fig. 2.

Procedure for quantifying the ZWD and PWV from GNSS data.

While the first term in (6) contains the dry delay, the second delay term is caused mainly by water vapor, but it may also contain a fraction of the dry delay. In order to accurately derive the wet delay from the total neutrospheric delay, precise information about the ZDD is required. The azimuthally isotropic ZWD, Δρgnss zw,iso , is then retrieved as follows: gnss met Δρgnss zw,iso = Δρz,iso − Δρz, mod

(7)

where Δρmet z, mod is the ZDD at the GNSS site calculated from model (1) using surface meteorological observations, which have to be available at the GNSS site. If the site is not equipped with suitable sensors, measurements from the adjacent meteorological sites have to be interpolated or extrapolated at the GNSS site, as described in Section V. The total slant wet delay along the satellite line-of-sight, i.e., Δρgnss sw , is calculated using gnss n Δρgnss sw = Δρzw,iso M Fw + Δ

∂M Fw cos A ∂z ∂M Fw + Δe sin A + Δr ∂z

(8)

where Δn and Δe are the estimated northing and easting neutrospheric gradients, respectively. A is the azimuth angle, and z is the zenith angle of the satellite. The phase residuals after least squares adjustment may contain effects from multipath and antenna phase center variation, as well as anisotropic wet path delay. This delay can be extracted from the phase residuals, for example, by residuals stacking; this part is denoted as Δr in (8). An approach to reduce the systematic site-specific error sources, such as multipath effects, from the residuals and to reconstruct the anisotropic water vapor signal is presented in [14]. In (8), the first term represents the isotropic wet delay, whereas the remaining terms account for the anisotropic delay. The slant wet delay can be mapped to the zenith direcgnss tion to derive the entire ZWD, i.e., Δρgnss zw = Δρsw /M Fw , which can be converted into PWV using (12). We discussed the principle to derive the ZTD by considering isotropic and anisotropic components; however, the results presented in this paper consider only the isotropic wet delay [first term in (8)], whereas we use (8) in current studies to include the anisotropic component.


Fig. 4.

3767

PWV estimates at ten GNSS sites plotted against the site altitude.

Also, the estimates of site-specific parameters represent a temporal average because the processing approach uses GNSS observations within a predefined time window (1 hour for this work) to achieve high accuracy. C. Conversion of Wet Delay Into Water Vapor The ZWD is considered as a measure of the total PWV in the atmosphere. The two quantities are related by the empirical constants Π resp. Q [17], i.e., PWV = Π · ZWD or

ZWD = Q · PWV

(9)

where Q = Π−1 is calculated based on the weighted mean temperature of the atmosphere, Tm , such that Q = 0.10200 +

Fig. 3. Time series of PWV estimated from GPS observations at ten GNSS sites located, as shown in Fig. 1 on (a) June 27, 2005 and (b) January 3, 2005.

1708.08 Tm

(10)

where In Fig. 3, we show time series of the PWV estimated from GPS measurements at ten sites on June 27, 2005 [summer (a)] and January 3, 2005 [winter (b)]. In summer, due to local advection effect, the PWV content is high and more variable, whereas in winter, the Upper Rhine Graben region has dry weather conditions resulting in lower and smoother PWV content. The content of PWV depends on the altitude of the GNSS site. In general, the PWV content decreases as the site altitude increases (see Fig. 4). However, due to local advection effects, the PWV content may increase even if the site is located at a higher altitude. It should be noticed that a GNSS receiver records the signals from all visible satellites with elevation angles above the cutoff elevation (7◦ in this paper). The GNSS-based PWV represents a measure of the average effect of a conical section of the neutrosphere above the GNSS receiver. Since most of the water vapor is located near the Earth’s surface, we can approximate the radius of the cone. Owing to the schematic diagram in Fig. 5, if the minimum elevation angle is 7◦ and assuming that water vapor is concentrated in the lower 1 km of the atmosphere, the corresponding cone radius is approximately 8 km.

Tm ≈ 70.2 + 0.72Ts .

(11)

Ts is the surface temperature in K [1]. As a rule of thumb, Π = 0.15 (Q = 6.67) is commonly used in atmospheric research [14], but the actual value varies with the surface temperature. If the amount of water vapor in the atmosphere is large, inaccurate values of Π can result in a significant PWV bias, i.e., if the ZWD is 200 mm, then the corresponding PWV is 30 mm for Π = 0.15 (rule of thumb). However, if Π is calculated using the measured surface temperature to be 0.16, then the corresponding PWV is 32 mm. The larger the ZWD, the more critical is the value of Π. Therefore, rather than using the value of Π = 0.15, we computed the value of Q using measurements of the surface temperature. In order to test the sensitivity of Q to the surface temperature, we used (9) to compute its value over 96 days in the years 2002 and 2004, as shown in Fig. 6. The figure plots Q against the atmospheric mean temperature Tm that is calculated using observations of the surface temperature from three meteorological sites. The value of Q varies with the temperature measured at each day, and we observed that Q is in the range of 6.039 to 6.633 using the observations of year 2002,

3768


Fig. 5. GNSS antenna receives a satellite signal at a minimum elevation angle (θmin ) that defines a cone-like neutrospheric section above the antenna. For θmin = 7◦ , rc ≈ 8 km. If PWV is available on a grid, the values within the circle should be averaged to emulate GNSS-based PWV.

Fig. 7. MERIS PWV against GPS PWV at ten sites on five days of Envisat overpass time. The MERIS observations are averaged within circles of 8 km radius centered on the GNSS site. The slope of the black line is 1.

the value of Π = 0.15 is not suitable for the selected test area, rather its value should be calculated using surface temperature measurements. IV. VALIDATION OF E STIMATED WATER VAPOR

Fig. 6. Empirical constant Q calculated based on the surface temperature measured at three meteorological sites and their average. The measurements were taken at 10:00 A . M . over 96 days (8 days/month) in the years 2002 and 2004.

whereas it varies between 6.047 and 6.570 in year 2004. The average curve is linearly decreasing with slopes of −0.0236/K and −0.0232/K for the years 2002 and 2004, respectively. There is a significant decrease in the value of Q from winter (low temperature) to summer (high temperature). Therefore, using

Here, we present a comparison of the PWV estimated from GPS data and PWV measured by MERIS, as well as that simulated by the model WRF. Since the site-specific PWV estimates correspond to spatially averaged values, we average the PWV from MERIS within circles of 8 km radius centered at the location of the GNSS site. Fig. 7 shows the PWV mean values from MERIS against those estimated from GPS observations at ten sites on five days, at MERIS acquisition time (9:51 UTC). The results show a strong correlation (93% in average) between MERIS and GPS. When calculating the difference between the two data sets for the five days, the mean value does not exceed 0.4 mm and a maximum standard deviation of 1 mm. The average RMS value is 0.98 mm. Indeed, we should not expect a perfect correlation between MERIS and GPS PWV values since we approximate the conical effect of GPS with a circle, which radius depends on our assumptions about water vapor residence. MERIS PWV maps are limited in temporal resolution, with a map of PWV being available only every a few days at a fixed time. Therefore, we used PWV maps simulated by the model WRF to further evaluate the estimates of PWV from GPS observations. WRF generates PWV maps at a spatial resolution


3769

Fig. 8. Time series of the PWV at four GPS sites and the corresponding PWV from WRF on January 3, 2005 and June 27, 2005. The PWV simulated by WRF are averaged within blocks of 15 km × 15 km to emulate GPS. The mean RMS value of the differences is 0.97 mm. TABLE I M EAN , S TANDARD D EVIATION , AND RMS VALUES FOR THE D IFFERENCES B ETWEEN GPS AND WRF DATA

of 3 km × 3 km. We compared time series of PWV estimated from GPS observations at four sites with those obtained from the model, as depicted in Fig. 8. In a similar way to MERIS, the WRF data are averaged within a block of 15 km × 15 km to emulate the conical effect of GPS. We compare the model data on the days January 3, 2005 and June 27, 2005 with the GPS-based PWV estimates. The first example (January 3, 2005) shows strong agreements between the GPS and WRF data, whereas the second example (June 27, 2005) shows a bias in the data. The mean, standard deviation, and RMS values of the differences of the two datasets are given in Table I. It is observed from Fig. 6 that the GPS estimates of PWV show more variability over time compared with the simulations of the WRF model. It is here important to note that the model output depends on various factors such as the initial conditions and the setting for the boundary conditions; therefore, the WRF model might show scenarios where the simulated PWV values may have a poor correlation with the GPS estimates of PWV or may be biased.

the surrounding area. The meteorological site might be 100 km distant, if the topography is smooth, since the temperature is more variable in the vertical than the horizontal direction. At local scales, the variations in the air pressure and temperature in horizontal directions are small compared with variations in the vertical direction. Therefore, if meteorological measurements from one adjacent meteorological site are used, we calculate the pressure and temperature at the GNSS site by considering the altitude difference without accounting for the horizontal variations. The air pressure at the GNSS site is computed using the hydrostatic equation g/RL L(hgnss − hmet ) gnss met =P (12) 1− P T met where P gnss is the air pressure computed at the GNSS site altitude hgnss , P met is the air pressure measured at the altitude hmet of the meteorological site, and the measured air temperature is denoted by T met . Factor g defines the mean Earth’s gravity acceleration (9.80665 m/s2 ). R denotes the universal gas constant (8.31447 J/mol K), and L is the temperature lapse rate (0.0065 ◦ C/m). Since the temperature varies linearly with the site altitude, it is calculated at the GNSS site (T gnss ) using the following formula: T gnss = T met − L(hgnss − hmet ).

(13)

If multiple meteorological sites are located close to the GNSS site, we can account for vertical and horizontal variations in air pressure and temperature. We model the pressure as a sum of a component proportional to the altitude of the meteorological site and a correction term that accounts for the horizontal variations, i.e.,

V. I NTERPOLATION OF A IR P RESSURE A ND T EMPERATURE

P = P0 (h) + ΔP (x, y)

When meteorological measurements are not available at the GNSS site, we need to find at least one meteorological site in

with P0 (h) = ah + b, and ΔP (x, y) accounts for the horizontal variations of P . Using measurements from different

(14)

3770

Fig. 9.


Interpolated maps of pressure and temperature and the aggregated maps from the model WRF on 06-27-2005. The resolution of the grids is 9 km × 9 km.

meteorological sites, we first estimate the regression parameters a and b and then calculate the residuals. The linear regression model is used to compute the elevation-dependent pressure at different altitudes. To account for spatial variations, the residuals are used to calculate a correction value at any point by applying, for example, inverse distance weighting. In a similar way, the temperature can be computed at any point using T = T0 (h) + ΔT (x, y)

(15)

with T0 (h) and ΔT (x, y) modeled in a similar way as for the pressure. In order to evaluate the models used for interpolating the air pressure and temperature, we compare the interpolation results to WRF maps. As input, we used meteorological data downsampled by a factor of ten to produce continuous grids of 9 km × 9 km cells. The original WRF data are aggregated to the resolution of 9 km × 9 km to compare the results. Fig. 9 shows the maps of pressure and temperature produced by interpolation and the aggregated maps from WRF as well as the difference maps. There is strong agreement between the maps for both pressure and temperature with only small differences. The mean values of the difference maps are close to zero and the standard deviations are 1.9 hPa for the pressure maps and 0.18 ◦ C for the temperature maps. Applying the models in (14) and (15) using measurements from several meteorological sites, we calculate the pressure and temperature at the GNSS site by accounting for both the horizontal and vertical variations.

estimates from GPS data with the corresponding measurements from MERIS and the model WRF show strong agreement with RMS values of less than 1 mm. In order to improve the quality of the GNSS estimates of water vapor, we first performed three-dimensional interpolations of pressure and temperature at the GNSS site. This was achieved by splitting the pressure or temperature into two components: the first and largest accounts for vertical variations, and it is calculated at the GNSS site using a linear model. The second accounts for horizontal variations, and it is calculated using, for example, the method of inverse distance weighting. Second, we calculated the conversion factor Π using surface temperature measurements rather than relying on a constant value. ACKNOWLEDGMENT The authors would like to thank the GNSS data providers: RENAG, RGP, Teria, and Orpheon (France), Satellite Positioning Service (SAPOS)–Baden–Württemberg and Rheinland– Pfalz (Germany), European Permanent Network, and CODE. The authors thank ESA for the MERIS data. The authors also thank the Institute of Meteorology and Climate Research, Atmospheric Environmental Research, Karlsruhe Institute of Technology for preparing the WRF simulations and the Landesanstalt für Umwelt, Messungen und Naturschutz Baden– Württemberg for providing the meteorological observations. R EFERENCES

VI. C ONCLUSION In this paper, we have presented a method to determine the total PWV in the Earth’s atmosphere using GNSS data and surface meteorological observations. For an accurate determination of PWV, we found that it is important to use meteorological observations either measured at the GNSS site or interpolated from adjacent meteorological sites. Comparisons of the PWV

[1] M. Bevis et al., “GPS meteorology: Remote sensing of atmospheric water vapor using the global positioning system,” J. Geophys. Res., vol. 97, no. D14, pp. 15787–15801, Oct. 1992. [2] M. Bender et al., “Validation of GPS slant delays using water vapor radiometers and weather models,” Meteorologische Zeitschrift., vol. 17, no. 6, pp. 807–812, Dec. 2008. [3] A. Karabati´c, R. Weber, and T. Haiden, “Near real-time estimation of tropospheric water vapor content from ground based GNSS data and its potential contribution to weather nowcasting in Austria,” Adv. Space Res. vol. 47, no. 10, pp. 1691–1703, May 2011.


[4] F. Alshawaf, “Constructing Water Vapor Maps by Fusing InSAR, GNSS, and WRF Data,” Ph.D. dissertation, Karlsruhe Instit. Technol., Karlsruhe, Germany, 2013. [5] I. Jankov, W. A. Gallus Jr., M. Segal, and S. E. Koch, “Influence of Initial Conditions on the WRF-ARW model QPF response to physical parameterization changes,” Weather Forecast., vol. 22, no. 3, pp. 501–519, Jun. 2007. [6] M. Bender et al., “GNSS water vapor tomography—Expected improvements by combining GPS, GLONASS and Galileo observations,” Adv. Space Res., vol. 47, no. 5, pp. 886–89, Mar. 2011. [7] M. Mayer et al., “GURN (GNSS Upper Rhine Graben Network): Research goals and first results of a transnational geo-scientific network,” In Geodesy for Planet Earth, IAG-Symposia, vol. 136, S. Kenyon, M. C. Pacino, and U. Marti, Eds. Berlin, Germany: Springer-Verlag, 2012, pp. 673–681. [8] R. Bennartz and J. Fischer, “Retrieval of columnar water vapor over land from back-scattered solar radiation using the Medium Resolution Imaging Spectrometer (MERIS),” Remote Sens. Environ., vol. 78, no. 3, pp. 271–280, Dec. 2001. [9] J. Saastamoinen, “Contributions to the theory of atmospheric refraction,” Bull. Géodésique, vol. 107, no. 1, pp. 13–34, Mar. 1973. [10] P. Misra and P. Enge, Global Positioning System: Signals, Measurements and Performance, 2nd ed. Lincoln, MA, USA: Ganga-Jamuna Press, 2001. [11] M. G. Lawrence, “The relationship between relative humidity and the dew point temperature in moist air: A simple conversion and applications,” Bull. Amer. Meteorol. Soc., vol. 86, no. 2, pp. 225–233, Feb. 2005. [12] D. Balton, “The computation of equivalent potential temperature,” Mon. Weather Rev., vol. 108, no. 7, pp. 1046–1053, Jul. 1980. [13] R. Dach, U. Hugentobler, P. Fridez, and M. Meindl, “Bernese GPS Software Version 5.0,” Astron. Inst., University of Bern, Bern, Switzerland, 2007. [14] T. Fuhrmann, X. Luo, A. Knöpfler, and M. Mayer, “Generating statistically robust multipath stacking maps using congruent cells,” GPS Sol., vol. 19, no. 1, pp. 83–92, 2015. [15] A. Niell, “Global mapping functions for the atmosphere delay at radio wavelengths,” J. Geophys. Res.: Solid Earth, vol. 101, no. B1, pp. 3227–3246, Feb. 1996. [16] K. S. W. Champion, A. E. Cole, and A. J. Kantor, “Standard and reference atmospheres,” in Handbook of Geophysics and the Space Environment, 4th ed. Washington, DC, USA: U.S. Air Force, 1985, pp. 14–1. [17] M. Bevis et al., “GPS Meteorology: Mapping zenith wet delays onto precipitable water,” J. Appl. Meteorol., vol. 33, no. 3, pp. 379–386, Mar. 1994.

F. Alshawaf received the Ph.D. degree in remote sensing from Karlsruhe Institute of Technology (KIT), Karlsruhe, Germany, in 2013. Since 2010, she has been a Research Assistant with the Institute of Photogrammetry and Remote Sensing and the Geodetic Institute at KIT. Her research interests include water vapor mapping using interferometric synthetic aperture radar and GNSS, and improving the quality of these maps by statistical data fusion with numerical weather models.

T. Fuhrmann received the Diploma degree in geodesy and geoinformatics from the Karlsruhe Institute of Technology (KIT), Karlsruhe, Germany, in 2010. He studied the GNSS-based estimation of water vapor using precise point positioning. He is currently working on a project to detect crustal deformations based on leveling, interferometric synthetic aperture radar, and GNSS data at the Geodetic Institute, KIT, Germany.

3771

A. Knöpfler received the Diploma degree in geodesy and geoinformatics from the Karlsruhe Institute of Technology, Karlsruhe, Germany, in 2005. His research interest is GNSS in science and education with a special focus on geodynamics and sitespecific effects.

X. Luo received the Ph.D. degree in geodesy from the Karlsruhe Institute of Technology, Karlsruhe, Germany, in 2012. Since September 2013, he has been a GNSS Application Engineer of the GNSS Product Management Group, Leica Geosystems. His research interests include analysis of stochastic models, and atmospheric and site-specific effects in GNSS data, with a special focus on statistical testing and timeseries modeling.

M. Mayer received the Ph.D. degree in geodesy from the Karlsruhe Institute of Technology (KIT), Karlsruhe, Germany, in 2005. He studied the modeling of the GPS deformation network of the Antarctic Peninsula. He is currently the Head of the GNSS Working Group, Geodetic Institute, KIT. His research interests include analysis of atmospheric and site-specific GNSS effects with a special focus on continuously operating reference sites.

S. Hinz received the Ph.D. degree and Habilitation in computer vision and remote sensing from the Technical University of Munich, Munich, Germany, in 2003 and 2008, respectively. In 2008, he became a Full Professor and Director of the Institute of Photogrammetry and Remote Sensing with the Karlsruhe Institute of Technology (KIT). He is currently the Dean of the KIT-faculty of Civil Engineering, Geo- and Environmental Sciences. His research interests include theory and methods of computer vision and remote sensing, with particular focus on semantic image understanding, image sequence analysis, hyperspectral and radar remote sensing.

B. Heck received the Ph.D. degree and Habilitation in geodesy from the Karlsruhe Institute of Technology (KIT), Karlsruhe, Germany, in 1979 and 1984, respectively. Since 1991, he has been with the Geodetic Institute, KIT, where he fills the position of Professor and Chair of Physical and Satellite Geodesy. His research interests include geometrical geodesy, deformation analysis, gravity-field analysis, and GNSS positioning. Prof. Heck has been the Director of the Black Forest Observatory since 1999 and is an active member of the International Association of Geodesy.