Improved estimation of the tropospheric delay component in GNSS and InSAR measurements in the Western Corinth Gulf (Greece), by the use of a highresolution meteorological model: The PaTrop Experiment N. ROUKOUNAKIS, P. ELIAS, P. BRIOLE, A. ARGIRIOU, I. KIOUTSIOUKIS, A. RETALIS, D. KATSANOS
Background During the past decade, extensive research has been carried out, with respect to the application of space geodesy techniques (INSAR and GNSS) for mapping regional surface deformations due to tectonic movements. High accuracy (to mm level) is required in most cases, as horizontal and vertical surface velocities are of the order of a few mm/yr. In the region of the Corinth Gulf, Greece, GNSS and InSAR techniques have been extensively deployed as part of an international research effort (Corinth Rift Laboratory) to monitor local tectonic deformities and study seismicity. Corinth Rift max extension rate 16mm/year. High seismicity, intense geophysical interest. Monitoring network includes inclinometers, strain meters, seismometers, tide-gates, GNSS stations, active satellite observations (SAR).
G2 Conference, 16-18 novembre 2015, Toulouse Roukounakis et al
Background Multi-disciplinary approach Meteorology GNSS: Large time-series (>10 years) available, from existing network of 12 permanent GNSS stations that has been gradually installed in the region + measurements from semi-permanent stations. Higher uncertainty for the vertical component InSAR: ERS-1 & 2 from 1992 to 2002 ENVISAT from 2002 to 2010 RADARSAT-2 from 2009 to now TerraSAR-X from 2010 to now } Higher uncertainty for the horizontal component
GNSS Troposheric Delay Figure 1: First row shows the horizontal (left) and vertical (right) positioning error using (blue) or not using (red) the tropospheric correction. The variation in range is shown in the second row at left (ESA, 2011).
GNSS Tropospheric Delay A variety of tropospheric models and mapping functions have been successfully deployed for the estimation of Zenith Total Delays (ZTD): Climate re-analysis models: VMF1, GMF-GPT, GPT2 (Boehm et al, Steigenberger et al, Lagler et al, Wang et al). GNSS data combined with other satellite observations (eg MERIS) and radiosonde measurements (Bonafoni et al, Shangguan et al, Vasquez et al). Figure 2: Global distribution of total delay RMS error from the GALILEO tropospheric model (ESA, 2012).
Physics of the atmospheric propagation delay Saastamoinen [1972] showed that the total atmospheric delay is partitioned in two quantities: Hydrostatic delay (dependant only on surface pressure), and wet delay (function of water vapour distribution) Elgered et al. [1991] adopted a model in which the zenith hydrostatic delay (ZHD), in millimeters, is given by: L 0 h =(2.2779±0.0024) P s /f(λ, Η) (3) where P s is the total pressure (mbar) at the Earth's surface, and f(λ, Η)=(1 0.00266 cos 2λ 0.00028Η) (4) Accounts for the variation in gravitational acceleration with latitude λ and the height H of the surface above the ellipsoid (in km). The zenith wet delay (ZWD) is given by: L 0 w =(0.382±0.004) K 2 mbar 1 ( P v / T 2 )dz (5)
The PaTrop Experiment A network of twenty permanent and temporary GNSS receivers (Topcon GB1000 fitted with Topcon PG-A1 antennas, at 30s acquisition) provides the data for the PaTrop experiment, covering an area of approximately 100x80 km in the region of the Western Corinth Gulf, Greece. Receivers are distributed in both flat and mountainous terrains (elevation range 0-1500m), in order to study different topographical and meteorological conditions and achieve the desired spatial resolution. The duration of the experiment is 12 months, with the aim to investigate seasonal effects. Processing of the data involves the calculation of detailed tropospheric delays (ZTDs, ZWDs and ZDDs) every 5 min, with the GIPSY-OASIS 6.12 software. Results will be compared with the tropospheric delays derived from the WRF weather re-analysis (Ps, T, Pv), at 5 km and 1 km horizontal resolution, and 34/45 pressure levels in the vertical tropospheric column, respectively.
The PaTrop Experiment Figure 3: Map showing the PaTrop GNSS network (in blue) and the CRL permanent stations (in yellow) providing data for the troposphere experiment.
WRF local reanalysis A regional weather forecasting model, such as WRF, provides high resolution scenarios, which can be used for impact assessment. High resolution allows for a more precise description of regional topographic forcings due to orography, land-sea contrasts and vegetation characteristics and therefore processes strongly forced by topography, such as orographic precipitation and relative humidity, can be represented much more accurately at increased resolution. Both are important meteorological parameters that have a strong effect, as it is water vapor content in the lower troposphere which is the main uncertainty in the estimation of GPS atmospheric delay. Figure 4: Map showing surface temperatures as calculated by the 1x1 km WRF re-analysis run in the Corinth Gulf, for October 2014
WRF 5x5 km re-analysis Figure 5: Map of the study area, showing the horizontal 5x5 km WRF grid, the location of the GNSS stations and the nearest grid points used for the calculation of ZHDs, ZTWs and ZTDs. A 1x1 km configuration will provide better horizontal approximation, increased vertical pressure level stratification as well as improved terrain detail, thus greater accuracy with regards to Ps, T and Pv estimated by the model. G2 Conference, 16-18 novembre 2015, Toulouse Roukounakis et al
Preliminary results GNSS derived delays 300 250 KRIN KOTE 200 150 100 50 300 250 200 GALA MESA Diagrams 1-2: GNSS derived tropospheric residuals, for 4 stations of the PaTrop network, October- November 2014. 150 100 50 30/09/2014 10/10/2014 20/10/2014 30/10/2014 09/11/2014 19/11/2014 29/11/2014 09/12/2014
WRF 5x5 km re-analysis: Preliminary results ZWD and ZTD values calculated, for October 2014. Every run lasts 48h, with initial conditions entered from the ERA-Interim global dataset. Values compared with 30 min residual averages from tropospheric processing of GNSS data after a priori tropospheric model has been applied (GPT2). Correlation between GNSS and WRF ZWDs is particularly strong, indicating that the biggest part of the residual values is the wet component which is not adequately modeled by the a priori troposphere. Hydrostatic component possibly still remains after data processing, as shown by higher values of GNSS residuals vs WRF ZWD. As the model moves away from initial conditions, there is a considerable offset: When model output corresponds to the first 24h of the 48h run, the resulting values are much more consistent with observations. ZTDs and ZWDs are lower in mountainous locations (MESA h=480m and KALA h=717m) than in stations close to sea level (MESO h=4m and KOTE h=82m), as expected, due to decreased surface pressures and temperatures, but show a higher degree of temporal variation.
WRF 5x5 km re-analysis: Preliminary results Diagrams 3-6: WRF vs GNSS Tropospheric delays, for 4 stations of the PaTrop network, 11 th October 2014.
InSAR tropospheric correction Techniques being used to mitigate atmospheric effects: Limitations: Using large number of SAR scenes to } require larger data sets (>10 20 scenes) generate an averaged image. that may not be available for a given Persistent scatterers technique (PSInSAR) geographic region and time of interest. estimates delay through a series of images based on the residuals after low-pass timedomain filtering. Correction techniques that can be applied on a scene-to-scene basis are highly desirable: Multi-spectral observations (e.g. MERIS and MODIS) Weather models Tropospheric correction from GPS stations Phase-based empirical methods (linear and power-law approximations) Spectrometers can only provide useful corrections under cloud-free and daylight conditions. Produce an estimate for the wet component only. Timing and location issues, making them unable to correctly resolve the turbulent variation of water vapor. Absent or sparsely distributed Only estimate a topography correlated component of delay. Cannot account for the turbulent and coherent short-scale component.
InSAR tropospheric correction Figure 6: Atmospheric path delay corrections for interferometric pair of 29 July 2000 and 18 August 2001 over California. (a) original interferogram (deformation field has been modeled with GPS observations and removed from the interferogram); (b) interferogram corrected using MODIS data. (from Li et al, 2005).
InSAR tropospheric correction For selected cases where seismic/aseismic deformation is observed and InSAR images are available, the tropospheric effect will be accurately estimated and minimised with a combination of GNSS/high resolution meteorological modeling. Thus, problems related with model validation, timing and location capabilities will be overcome with the use of a dense network of GNSS receivers. The successful modeling of the lower troposphere with smaller pixel size (a few km), will enable us to correct the tropospheric delays in ERS interferometry acquired before the deployment of the permanent GNSS network. Figure 7: Interferogram of the Movri earthquake (2008) showing significant tropospheric noise
Conclusions Future work The application of WRF is expected to provide improved tropospheric modeling, especially with respect to the wet component (ZWD). Corrected GNSS derived Precipitable Water Vapor (PWV) may be an additional source of data for the NWP in order to obtain more accurate weather forecasts. The accuracy provided by the NWP will be improved with the use of further downscaling (1x1 km) and assimilation of surface data from weather stations (NOA, UPAT, HNMS) and possibly radiosonde data. Further WRF 5x5 km and 1x1 km runs are needed to cover the period Oct 2014-Sep 2015. A second objective of the PaTrop experiment is to correct the tropospheric delays in InSAR acquisitions with a combination of GNSS/meteorology.