InSAR atmospheric effects over volcanoes - atmospheric modelling and persistent scatterer techniques Rachel Holley 1,2, Geoff Wadge 1, Min Zhu 1, Ian James 3, Peter Clark 4 Changgui Wang 4 1. Environmental Systems Science Centre, University of Reading 2. Fugro NPA Ltd 3. Meteorology Dept., University of Reading 4. JCMM/Met.Office
Overview Modelling atmospheric phase delays using the Unified Model Contributions to atmospheric refractivity Comparison between model and MERIS Correction of DifSAR interferograms Comparison between modelled fields and persistent scatterer atmospheric phase screen results Corrected DifSAR deformation results Persistent scatterer deformation results Conclusions
Contributions to phase ΔΦ int =ΔΦ def + ΔΦ orb + ΔΦ topo + ΔΦ noise + ΔΦ atm Atmospheric refractivity Errors caused by changes in atmospheric refractivity between SAR dates is a wellknown problem for InSAR, particularly in areas of high relief. Etna is a good location for studying this problem due to high relief and coastal climate Interferogram of Etna from 5-6th Sept 1995 shows atmospheric fringes surrounding summit
Possible Mitigation Methods Stacking Time series methods Persistent Scatterer methods Calibration with GPS water vapour estimates Removal of atmospheric noise Calibration with radiometric water vapour estimates (e.g. MERIS) Statistical Calibratory Forward atmospheric modelling
Unified Model set-up Nested domain scheme, outer domains provide initial state and boundary conditions for the higher resolution domains nested within Model physics includes land surface, planetary boundary layer, radiation, cloud microphysics Global Grid length 60 km 12 km 4 km 1 km 300 m Timestep 20 min 5 min 1.33 min 30 sec 10 sec Boundary update 1 hour 1 hour 15 min 15 min
Running the Unified Model Inputs Outputs Initialisation data: European Centre for Medium-Range Weather Forecasts (ECMWF) global atmospheric analysis 300 m domain orography from SRTM DEM 300 m domain landuse and vegetation, derived from the IGBP 1 km dataset Run time - 6 hour model integration finishes within 1 hour using > 32 processors Model outputs include 3D fields of: Temperature Pressure Water vapour Liquid cloudwater Solid cloudwater Wind speed and direction
Atmospheric refractivity P 7 ne e e N = k1 + 4.028 10 + k' 2 k3 + 1. 45W 2 + T f T T Hydrostatic Ionospheric Water vapour Liquid water Long wavelength Model Model Model k 1 =77.6 K kpa -1, P = total atmospheric pressure in hpa T = absolute temperature in K K' 2 = 23.3 K hpa -1, e = partial pressure of water vapour in hpa, n e is the electron density per cubic metre, f = radar frequency W = liquid water content in gm -3.
MERIS comparisons 11 partially-clear MERIS scenes compared with the equivalent modelled fields. Standard deviation of difference fields 0.5-1.7 mm PWV, around accuracy of MERIS data Slight wet bias in the modelled fields (+0.6 mm average), especially at low PWV values (high altitudes) Some cases where opposite bias present - different gradients with altitude shows potential for apparent deformation
Direct correction with MERIS Direct correction of interferograms possible using cloudfree MERIS However even partially cloud-free scenes rare for Etna Six interferogram pairs possible, but all lack any significant coverage over volcano
Interferogram correction SAR Data Unified Model ASAR1 ASAR2 PW1 PW2 Q factor Ps1 Ps2 Interferogram PWV difference dry difference Delay difference map Corrected interferogram
Planar gradients Long-wavelength atmospheric gradients can be removed from interferograms along with orbital gradients, resulting in planar mis-matches in corrected interferograms. Requires removal of a best-fit planar gradient calculated in areas where no deformation expected.
Interferogram correction 72 interferograms corrected using modelled atmospheric fields Reduces standard deviation of interferograms to ~12 mm (calculated in areas where no deformation expected) Reveals some unwrapping errors, could be improved by unwrapping to model rather than assuming minimum phase change
Interferogram correction - limitations Mis-representation of small scale processes (e.g. small convection cells) Location errors of model features can cause errors Some evidence of tendency to overcorrect vertical gradients
Model vs APS First challenge is separation of atmospheric signal from non-linear deformation using temporal filtering Seasonal atmospheric signals occur on same timescales as deformation, impossible to separate fully
Model vs APS Division between cases with strong seasonal difference (summerwinter) and strong correlation, and weak seasonal difference (summersummer) with poor correlation. Those showing good correlation show larger variation in model than APS, model appears to overestimate magnitude of topographic signal. Highly dependant on master date.
Comparison with other data
Modelling conclusions Use of the model allows consideration of more components of atmospheric phase delay than other methods (dry delay, liquid water, effect of temperature on Q scaling factor) Modelling improved by assimilation of PWV data into initial fields, and by use of high-resolution topography Sensitivity to initialisation data reduced relative to previous work, but still an issue; coarse resolution can cause timing/position errors in modelled fields. Tendency for the model to misrepresent PWV fields at higher altitudes, gradient differences can cause phase errors with altitude. Modelling not yet suitable for operational use, but possibilities for improvements identified
DifSAR Ascending (left, 3115 days) and descending (right, 1715 days) stacked corrected interferograms using data from 2004-2006. Deformation features visible (e.g. Pernicana Fault, Catania Anticline), however atmospheric influence still present.
Persistent Scatterers
Persistent Scatterers Deformation data at a variety of scales, time series data show correlation with eruptions
Future modelling work Global model initialisation data now available at higher resolutions, should reduce timing/position errors Investigate assimilation of other PWV data sources into model (MERIS/MODIS, GPS etc.) Use other data sources to investigate timing errors in model, and attempt a network correction method Assimilate higher-resolution land use and soil moisture data into model, for example ESA s SMOS