Figure 1: GPS receiver network used for processing at KNMI. KNMI is generating two types of ZTD’s: an observation available every hour (sites denoted by open circles) and every 15 minutes (sites denoted by asterisk).
Figure 1: GPS receiver network used for processing at KNMI. KNMI is generating two types of ZTD’s: an observation available every hour (sites denoted by open circles) and every 15 minutes (sites denoted by asterisk).
Ground based Global Navigation Satellite Systems (GNSS), such as the well known Global Positioning System (GPS, United States) and, in the future, Galileo (Europe), can partially fill this lack of information. Because GPS is currently the only operational GNSS we will use the term GPS in the text below. GPS signals, transmitted by a GPS satellite and received on earth are bended by the atmospheric refractivity resulting in a signal delay. When the satellite and receiver position are known an estimate of the total refractivity along the signal path can in principle be made. Refractivity depends on temperature, atmospheric pressure and humidity. When temperature and pressure are known, information on the humidity can be derived. In practice, also other unknowns such as receiver clock offsets have to be estimated.

The benefit of GPS humidity observations lies in the fact that the observations are possible in any weather situation and the observation frequency can be as high as 5 to 15 minutes. The deficit is that the observations are (currently) restricted to land and that an integrated quantity is observed and not a profile.

Method of observing upper air humidity using GPS
To accurately estimate the total delay due to the atmosphere, the positions of the GPS receivers have to be known with an accuracy of a few millimetres. This requires a fixed stable network with high standard GPS receivers. Note that low-cost GPS receivers have a position accuracy of a few meters. In the Netherlands a network of 35 receivers is installed as a result of close collaboration, called NETPOS, between the Kadaster (the National Surveying Department of the Netherlands) and the Ministry of Transport, Public Works and Water Management. On three remote automatic weather stations operated by KNMI a GPS receiver is installed. In return KNMI has access to raw GPS time delay observations of the network. This data is available in batches of one-second values every 5 minutes with a latency (timeliness) of less then 2 minutes.

Surveying with GPS with an uncertainty of a few centimetres is achieved using NETPOS. All raw one-second GPS data received by the network are processed to calculate a first order atmospheric correction. For atmospheric applications a higher accuracy in positioning is needed. This higher accuracy requires estimation of several unknowns such as satellite position and receiver clock errors. By collecting and processing all raw GPS time delay observations from the network over a period of several hours this accuracy can be met. Furthermore, a network of GPS receivers with baselines longer than 1000km, will improve the absolute accuracy of the atmospheric estimate. In Figure 1 the network of the GPS receivers used for meteorological applications is shown.

The open circles denote the sub-network used to calculate atmospheric delays every hour and the asterisks show the sub-network for real-time applications. The first product is available after 45 minutes of the last observations and is aimed to be used in numerical weather prediction models, while the real-time delay is less accurate but still suitable for nowcasting purposes.

The atmospheric delay is called the Zenith Total Delay (ZTD) and is determined for each GPS receiver. ZTD can be expressed as a sum of a dry and wet part, the so-called Zenith Hydrostatic Delay (ZHD) and the Zenith Wet Delay (ZWD). The ZHD can be approximated using the surface atmospheric pressure1), and thus the ZWD can be computed from the total delay by subtraction, that is

ZWD = ZTD – ZHD.

The ZWD is associated with the vertically integrated column of water vapour (IWV) over the GPS receiver

IWV = 1/k ZWD.

The factor k depends on the weighted mean temperature of the atmosphere, which can be approximated by a function of the surface air temperature2). So IWV can be obtained from a ZTD observation.

Applications of GPS humidity observations
Figure 2: Residual signal of the GPS receiver for three different intervals of one hour. The middle panel shows the residual from two satellites; for the other panels only one residual was observed above 50 degrees elevation.
Figure 2: Residual signal of the GPS receiver for three different intervals of one hour. The middle panel shows the residual from two satellites; for the other panels only one residual was observed above 50 degrees elevation.

Figure 3: Scatter plot of Convective Available Potential Energy (CAPE) as observed by radiosonde measurements and spectral power density (P) for two periods (two weeks in November 2000 and May 2003) and three GPS sites and four radiosonde launch sites. Symbols are connected by a line when more than one GPS satellite was visible.
Figure 3: Scatter plot of Convective Available Potential Energy (CAPE) as observed by radiosonde measurements and spectral power density (P) for two periods (two weeks in November 2000 and May 2003) and three GPS sites and four radiosonde launch sites. Symbols are connected by a line when more than one GPS satellite was visible.

Atmospheric instability and GPS
In fact, ZTD is an average over time and space of the delay of the GPS signal. The time dependent residual signal, which is the difference between estimated ZTD mapped back on the line of sight between the receiver and satellite and the observed observation, contains information on the state of the atmosphere. Figure 2 shows the residual signal with an observed elevation larger than 50 degrees mapped to the zenith. The major systematic errors (such as e.g. multipath signal reception) are removed from the signal, so the remaining signal will contain only noise and, when present, an atmospheric signature. In case of a convective atmosphere with significant updraft of humid air, the GPS signal will be influenced by fluctuations of the water vapour. Spectral analysis of a time series of one hour of GPS residual observations is performed to calculate the power spectral density (P) of the GPS signal. This parameter is compared to a measure of buoyancy called Convective Available Potential Energy (CAPE), which can be determined using radiosonde observations. In Figure 3 a scatter plot of P versus CAPE is shown for collocations of three GPS sites and four radiosonde launch sites. Despite the fact that the two measurements spaces are distinct (i.e. CAPE is based on a profile at a fixed time; P is based on a single value for the vertical averaged over a period of an hour) the correlation is remarkable (around 0.6). The continuous availability of GPS estimates may help the forecaster to detect atmospheric (in)stability3).

Figure 4: Locations in Europe with available GPS ZTD estimates. Operational sites are green; potential sites are black.
Figure 4: Locations in Europe with available GPS ZTD estimates. Operational sites are green; potential sites are black.

EUMETNET E-GVAP
KNMI participates in the EUMETNET GPS water vapour programme (E-GVAP).This programme aims at providing the EUMETNET partners with European GPS delay data for use in operational meteorology. This is done in close collaboration with the geodetic community in Europe. EGVAP started April 2005 and is planned for four years.
A snapshot of the locations at which GPS ZTD estimates are measured and derived is shown in Figure 4. Implementation of ground based GPS data in operational NWP and nowcasting has requirements regarding quality, homogeneity, stability, actions to take in case of problems, extent of observation network, etc. A key goal of E-GVAP is to gradually improve the ground based GPS (near-)real time delay data to meet these requirements.

E-GVAP-programme is based on a combination of centralised tasks, which will be carried out by the E-GVAP team, and distributed activities, which will be handled by the national met-service members in collaboration with their geodetic colleagues.

The centralised tasks include database setup and maintenance, processing of GPS data from special selected sites and quality monitoring and feedback. The distributed functions include, amongst other things, enlargement of the GPS network in areas with poor coverage in liaison with the geodetic community.

Figure 5: Example of the possible use of real time two dimensional GPS water vapour fields; a south westerly flow transported instable air which generated lightning events (black dots) in areas with a strong water vapour gradient, contoured from low values (blue) to high values (yellow).
Figure 5: Example of the possible use of real time two dimensional GPS water vapour fields; a south westerly flow transported instable air which generated lightning events (black dots) in areas with a strong water vapour gradient, contoured from low values (blue) to high values (yellow).

Two dimensional water vapour fields
From real time GPS IWV two dimensional water vapour fields can be generated. These fields can be used for nowcasting of convection and thunderclouds. In the example shown in Figure 5 strong lighting events occur at the edges of water vapour ridges. The possible use of these water vapour fields for nowcasting will be a topic of research for the next few years.

Expected Impact of GPS Tropospheric Slant Delays
An alternative to vertically aggregated propagation delay observations is to use the actual delay along the slanted path of signal propagation from satellite to receiver as observation. The intrinsic geometrical information in these so-called ‘slant delays’ offer the possibility of improved sampling of the local refractivity atmosphere in the vicinity of the receiver. A disadvantage of these observations is that they are likely to contain more noise. To find out if in spite of this detriment these observations contain useful information, an impact study was conducted with simulated slant delays.

The study comprised a System Simulation Experiment (SSE) carried out with the HIRLAM Forecasting System (HFS). The SSE is based on the Assimilation Ensemble Method (AEM), a probabilistic method originally used for the estimation of background errors4,5) that has been extended and applied to simulate observation impact6).
Key to the AEM is the creation of independent sets of random perturbations for a finite number of members in one or more ensembles to perturb the forecasting system, in this case through a fixed set of observations from a specific set of observation types. The rationale behind the AEM is that if the perturbations in observations are representative of observation errors, then the differences in contemporaneous analyses (forecasts) between the members of an ensemble are a surrogate for analysis (forecast) errors.

In this experiment three ensembles were generated, each defined by the set of observation types used in the ensemble:

  • Control - The reference, incorporating all observations used in operations: SYNOP, SHIP, DRIBU, TEMP, PILOT, WINDPROF, AIREP and SATOB,
  • Denial - As Control but without TEMP, PILOT and WINDPROF observations,
  • SDSIM - As Control but with simulated slant delay observations added.

The ensembles above consist of 4 members where each member represents an independent 14-day run of the forecasting system starting on May 2nd 2003 at 00H. The model runs are performed on a 10km resolution grid with the Netherlands at the centre. The Denial ensemble is carried out to determine the sensitivity of the HFS for the input of moisture observations. It acts as a means to calibrate the SSE. For the SDSIM ensemble simulated slant delay observations were generated before the experiment. They were computed from model fields of a nature run consisting of ECMWF analysis boundaries for the HFS and were created with the HFS forward operator for GPS slant delays. In each assimilation cycle of the SDSIM ensemble on the order of 100 simulated slant delay observations are available from a GPS network mainly situated in the Netherlands. The slant delays were created using the actual GPS satellite constellation geometry at observation time.

For an ensemble with N members the RMS of N-1 independent sets of analysis difference fields is called the analysis spread and represents a measure of the uncertainty in the analysis. The difference in the analysis spread between two ensembles that differ only in the set of observation types used in the analysis gives insight in the impact of a particular observation type.
Figure 6: Differences in the spread between the members of the SDSIM and Control ensemble for surface parameters PMSL, T2m and Td2m.
Figure 6: Differences in the spread between the members of the SDSIM and Control ensemble for surface parameters PMSL, T2m and Td2m.
Figure 7: Differences in the spread between the members of the SDSIM and Control ensemble for dew point temperature at standard model levels 850, 500 and 250 hPa, indicated by dTd850, dTd500 and dTd250 respectively.
Figure 7: Differences in the spread between the members of the SDSIM and Control ensemble for dew point temperature at standard model levels 850, 500 and 250 hPa, indicated by dTd850, dTd500 and dTd250 respectively.

The results for analysis spread difference in the calibration ensemble generally confirmed the beneficial impact of radiosondes (not shown). To illustrate the impact of slant delays the analysis spread difference between the SDSIM and the Control ensemble for surface parameters and dew point temperature at the 850, 500 and 250 hPa level is presented in the maps in Figures 6 and 7 respectively. The maps in these figures show a reduction in the spread in red which signifies positive impact. The yellow dots in and around the Netherlands in the panels represent the receiver stations in the GPS network.

For model surface fields of pressure reduced to mean sea level (PMSL), air temperature (T2m) and dew point temperature (Td2m) at 2 meter above the surface (at screen height) the impact is fairly neutral (Figure 6). The structure of the spread difference looks patchy for temperature and dew point temperature and amplitudes are small.

At 850 hPa positive impact is present around most stations of the GPS network and along the coast of Brittany (c.f. Figure 7). These structures reflect the dominant flow regime at the time. At 500 hPa the impact is generally positive mainly over central Europe but remarkably not at the hart of the receiver network. There is no significant impact over the Atlantic Ocean. At 250 hPa the impact is neutral over Europe. Impact structures, positive or negative, are related to frontal systems over the Atlantic and are more large-scale than those at lower levels.

In general it was found that slant delay observations could not improve the accuracy of the analysis for PMSL and 2-meter air temperature beyond what can be achieved with radiosondes. For these parameters slant delay impact is mainly neutral. The spread differences for dew point temperature indicate that slant delays can achieve small local improvements in the accuracy of the upper air analysis.

Outlook
In the next few years GPS ZTD observations will become available on an operational basis at European scale, also at KNMI. Applications of GPS with respect to nowcasting will be further investigated, with emphasis on convective systems. As GPS ZTD becomes routinely available the observations will be assimilated in HIRLAM. In the case of GPS slant observations more research is required, especially in the pre-processing stage of creating the observations. Implementation of realistic spatial and temporal observation error characteristics in the assimilation system of the HFS is needed, followed by impact studies with real observations on a very high resolution grid (<5km) that have to be carried out for an extended period of time to study the effects of seasonal variations in atmospheric humidity.

References

  1. Saastamoinen, J., 1972. Atmospheric Correction for the Troposphere and Stratosphere in Radio Ranging of Satellites. Geophysical Monograph Series, 15, 247-251.
  2. Bevis, M., S. Businger, T.A. Herring, C. Rocken, R.A. Anthes and R.H. Ware, 1992. GPS meteorology: Sensing of atmospheric water vapour using the global positioning system. J. Geophys. Res., 97, 15.787–15.801.
  3. Haan, S. de, 2006. Measuring atmospheric stability with GPS. J. Appl. Meteor., 45, 3, 467-475.
  4. Fisher M., 2003. Background error covariance modeling. Recent Developments in data assimilation for Atmosphere and Ocean. ECMWF Seminar Proceedings, Reading, UK, 8-12 September 2003, 45-64.
  5. N. Zagar, E. Andersson, M. Fisher, 2005. Balanced tropical data assimilation based on a study of equatorial waves in ECMWF shortrange forecast errors. Quart. J. Royal Meteor. Soc., 131, 987-1011.
  6. D.G.H. Tan and E. Andersson, 2005. Simulation of the yield and accuracy of wind profile measurements from the Atmospheric Dynamics Mission (ADM-Aeolus). Quart. J. Royal Meteor. Soc., 131, 1737 - 1757.