Processing of ground based GNSS data to produce near real time (NRT) tropospheric zenith path delays (ZTD) Jan Douša (jan.dousa@pecny.cz) Geodetic Observatory Pecný, Research Institute of Geodesy, Topography and Cartography, The Czech Republic E GVAP Workshop November 6, 2008
Outline introduction to GNSS the concept of GNSS contribution to meteorology different GNSS processing approaches (PPP x Network) general aspects of the network processing (in brief) the requirements and features of near real time (NRT) solution some results and comparisons historical view 2
GNSS Global Navigation Satellite Systems GPS NAVSTAR NAVigation System using Timing And Ranging The United States military service (1972, fully operational since 1994) GLONASS GLO GLObalnaja NAvigacionnaja Sputnikovaja Sistema Russian (the Soviet Union ) military service (1978, scheduled for restoration by 2010) GALILEO European Space Agency (ESA) European commercial service (1999, scheduled to be fully operational by 2013) DORIS (France), COMPASS or Beidou (China), QZSS (Japan), IRNSS (India) 3
Satellite tracks projected onto the surface GPS 31 (32) satellites / 6 orbital planes / 11h 58min GLONASS 21 (24) satellites / 3 orbital planes / 11h 15min Galileo 27 (30) satellites / 3 planes 4
basic GNSS observables GPS oscillator with fundamental frequency 10.23 MHz multiplied by 154x > 1575.42 MHz (L1) 120x > 1227.60 MHz (L2) code pseudorange the measure of the transit time from satellite to receiver using autocorrelation of received and replicated signal (the time is coded in signal) observables: C1 = L1 C/A, P1 = L1 P(Y), P2 = L2 P(Y) and many others in future 1m absolute positioning for civil usage phase pseudorange the measure of the phase difference btw. received and replicated carrier frequency observables: L1, L2 and others in future subcentimeter level relative positioning doppler data the measure of doppler shift due to a mutual motion of satellite and receiver 5
Error sources for GNSS Satellites: ephemeris, clocks, differencial code biases (AS antispoofing, S/A selective availability before 2000) Receivers: clocks, phase center offsets and variations, differencial code biases Environment: troposphere, ionosphere, multipath, Earth s kinematics Processing: cycle slips in phases, model errors Elimination by observable differences by introducing precise models and products 6
Parameters in GNSS mathematical model thus we have to handle somehow these parameters in GNSS processing: satellite and receiver position satellite and receiver clock corrections Earth orientation parameters and geocenter coordinates satellite and receiver code differential bias satellite and receiver phase center offsets and patterns troposphere effect ionosphere effect ambiguities 7
Observable differences to eliminate some of the errors in mathematical GPS model, we often create and use differences from the original observables: single difference (SD) difference between two stations difference between two stations (baseline generation), which eliminates the satellite clock errors observed at both stations double difference (DD) difference between two SDs (measurement to two satellites from the single baseline), which eliminates reciever clock errors tripple difference (TD) diferences between two DD in different epochs, which is useful to detect the phase skips (e.g. when signal from satellite was discontinued) the original observables we often call as zero difference (ZD) 8
GPS meteorology concept we know precise receiver and orbit positions, we eliminate ionosphere effect (receiver and satellite clock error), we introduce (PCVs, OCTIDE,...) (PCVs, OCTIDE,...) we estimate: zenith path tropospheric delay (receiver and satellite clocks) GPS x NWP 9
GPS observation equation Basic GPS carrier phase observable (scale to distance): Lrecsat = σrecsat + c*δsat + c*δrec + λ*nrecsat + ION + TRP + εrecsat σrecsat.. receiver satellite distance in vacuum (receiver and satellite coordinates) c.. speed of light δsat, δrec.. satellite and receiver clock errors λ.. wavelength of the carrier phase nrecsat.. unknown initial phase ambiguities ION.. ionospheric (slant) delay TRP satellite and receiver position need to be accurately known eliminated using double-differences (estimated in PPP!!!) need to be resolved integer or float first-order eliminated in the ionosphere-free combination.. tropospheric (slant) path delay TRP = mf (z) * ZHD + mf (z) * ZWD h w (z = zenith distance) ZTD = ZHD + ZWD ZTD [m] Zenith Total Delay (usually site & time dependent parameters) mfw / h mapping function (wet / hydrostatic) Lklij = Lkli Lklj = ( Lki Lli ) ( Lkj L lj ) double differences in network sol. 10
Least Squares Adjustment Observations: GPS distance measurements residuals (code and/or phase) unknown parameters stochastic information coordinates ambiguities ztd s after linearization user usually knows the models for the orbits, tides, etc. 11
Normal Equations (NEQs) minimizing the residuals: e P e min. normal equation parameters of interest (coordinates, troposphere,...) parameters to be eliminated (ambiguities) parameter estimation 12
Sequential Adjustment: Idea often applied in two ways: - time domain (sequential solutions) - space domain (network clusters) time Processing of sequential solutions : identical with processing all observations in a common adjustment, if there are no correlations of the original observations 13
Processing strategies & software Precise Point positioning (PPP) Advantages Small NEQ Keeping CPU with increasing number of sites / parameters (e.g. ZTD every 15 min, estimation of gradients) Disadvantages Network using double differences Correlations between parameters of all stations are taken into account Investigations of site dependent effect Independence of external products (except for small networks) Correlations btw stations are ignored Large NEQ Use of external products (orbits, clocks) Increasing CPU with incr. number of sites/parameters PPP approach: Epos - GFZ Gipsy - NGAA Network approach: Gipsy - ASI Bernese - BKG, GOP, KNMI, LTP, ROB, METO, SGN 14
PPP processing strategy (example GFZ) GFZ EPOS Software Part 1 - Network orbit improvement: Adjustment of precise orbits & clocks Global network : ~20 IGS+German sites Input orbits: GFZ 3h Ultra-rapid (pred.) CPU (Linux PC): ~6 to 8 minutes Part 2 - PPP Analysis: Estimation of trop. parameters Large set of parameters possible (high sampling rate, trop. gradients) NEW: slant delays estimation CPU (Linux PC): <5 min for 220 sites courtesy of Galina Dick (GFZ) 15
General network processing steps creating data batches (x hourly or sliding window) data quality check single point positioning for rough receiver clock synchronization network design by double differencing (clusters possible) data screening for phase cycle slips, ambiguities set up iterative site & satellite quality check and outliers rejection ionosphere product & ambiguity resolution reference frame realization & coordinate estimation ZTD product generation 16
Network processing strategy (example GOP) pre processing is based on two hours data batches 1 hour redundancy with the previous run easier ambiguity resolution, coordinates also for regularly late RINEX ( > 30min ) normal equations (NEQ) 1h for ZTD and 2h for coordinates processing in clusters of the network coordinates are combined from last 28 days using 2h NEQs with ambiguity fixed, free network solution, IGS05 reference frame ZTD product based on last 12h stacking of 1h NEQs ionosphere product for ambiguity resolution 17
GOP processing scheme 18
Ambiguity resolution in near real time initial phase ambiguities represent a huge number ( > 90%!) of necessarilly estimated parameters in mathematical GPS model in network solution, they can be resolved for integer numbers, which has strong impact for the coordinate estimation in short time data span ambituity resolution depends on time window and baseline lenght in GOP solution, for example, the ambiguities are resolved for 70% in total within two hour data batch applying two step approach (wide lane ambiguities at Melbourne Wubbenna phase+code linear combination resolved in 80 90% and narrow lane ambiguities at ionosphere free phase linear combination resolved with 70% success) resolved ambiguities are introduced as known at least for the official coordinate estimation (North/East/Up coordinate repeatability improved from 10/10/25mm to 6/6/16mm) a positive bias of aprox. 1mm observed in ZTD solutions btw ambituity free and fix solution! 19
NRT coordinate solutions The coordinates, which are fixed or tightly constrained in NRT ZTD solution should be as good as possible ( 3:1 for CRD:ZTD) example: GOP solution for the coordinates the coordinates are based on ambiguity fixed solution using last 28 days of two hourly NEQs, the solution is updated every hour. the coordinates are expressed in local datum close to the last ITRF realization by IGS (currently IGS05) by applying the Helmert transformation (fidutial stations are iteratively checked) 20
Troposphere model Bernese GPS software Slant tropospheric path delays = wet + dry (hydrostatic) are mapped into zenith using a mapping function (mf) SPD = mfh(z) ZHD + mfw(z) ZWD [z = zenith distance] where ZHD can be well a priori estimated if atmospheric pressure and station heiht+latitude are known (e.g. Saastamoinen, 1972) Because its variability, ZWD should be estimated for baselines > 20km Extended model could apply additionally the azimuthal dependency expressed as horizontal tropospheric gradients (GN north, GE easth): SPD = mfh(z) ZHD + mfw(z) ZWD + mfw/ z [ GN cos(a) + GE sin(a)] [A = azimuth] Constant or piece wise linear function is applied for ZTD Standard atmosphere (or in situ atm. pres. measurement) for a priori ZHD Dry and wet Niell mf ( Global or Vienna mf in future) 21
Troposphere model impact study example Some impacts in past using older models: no a priori model (zero value) and dry Niell mapping function used for the total zenith delay estimated (used until May 2005). 3. a priori ZHD based on standard atmosphere and wet Niell mapping function estimating ZTD (hopefully most of the ZWD). bias variable in time and space 2. Another site dependent bias was introduced in 2006 due to changing relative absolute Phase Center Variations and Offsets model used (upto 5mm) 22
Tropospheric product (GOP example) ZTDs for every hour (HH:00 + HH:59) a linear trend is considered between the values coordinates are heavily constrained to our estimated values realizing the IGb00 reference frame and written to the COST 716 format. ZTD product filtering: Sites with less than 4 hours of data in ZTD solution are excluded from the product Sites with less than 2 days of data in coordinate solution are excluded. ambiguity free (AF) and ambiguity fixed (AX) ZTD solutions are provided (officially AF), both using the same a priori coordinates values (ambiguity fixed). 23
Near real time aspects of ZTD estimation Requirements: hourly GNSS data (IGS, EPN, national,...) last hour precise orbits (IGS ultra rapids,...) for PPP: precise satellite clocks, DCB bias Features: processing started every hour usually ZTD at the edge of the processing window correlation with respect to previous estimates (physical, via processing, possible constraints depends on time resolution) Other important models: ocean and Earth tides (station coordinate, geocentr, satellite orbits) receiver and satellite phase center offsets and variations troposphere mapping function 2 nd, 3 rd order ionosphere many others especially in precise orbit determination 24
GNSS hourly data availability 25
Requirements on predicted orbits for ZTD 2008 2001 predominantly IGS ultra rapid orbits used errors in ZTD Synthetic error in orbit position 1m in along track 1m in cross track 1m in radial (mostly eliminated in DD) 26
ZTD results PPP vs Network ZIMM and GOPE one of the 12 supersites 27
Some ZTD/PWV comparison at GOP 2001 2003 comparison NRT x post processing StdDev : 4 7mm Bias : 1 3mm 28
ZTD comparison NRT GOP HIRLAM (NWP) weekly Sdev and Bias GPS ZTD from GOP near real time NWM Hirlam from DMI StdDev: 8 16mm (28mm) Bias: upto 16mm (25mm) (strong seasonal variation) 29
AC s NRT ZTD x post processing @ GOP ztd differences freqency & distribution functions (2004/2005) ACRI ASI BKG GFZ GOP IEEC LPT NKG NKGS B O R 1 G O P E H E R S P O T S W T Z R O N S A M A R 6 C A G L M A T E 30
Hour x day plots (ztd differences) NRT x post processing 31
Ground based GPS meteorology (Europe) COST-716 Action (1998-2003): "Exploitation of Ground-Based GPS for Operational Numerical Weather Prediction and Climate Applications 15 Institutions 7 ACs > 200 GPS sites TOUGH (2003-2006): Targeting Optimal Use of GPS Humidity Measurements in Meteorology 15 Institutions (Coordinator DMI) 12 ACs > 400 GPS sites E-GVAP (2006-2009): The EUMETNET GPS Water Vapor Programme 13 Institutions 10 ACs > 800 GPS sites 32