InSAR measurements of volcanic deformation at Etna forward modelling of atmospheric errors for interferogram correction Rachel Holley, Geoff Wadge, Min Zhu Environmental Systems Science Centre, University of Reading
Atmospheric errors in InSAR Atmospheric errors are a well established problem in interferometry, particularly in areas of high relief. ERS tandem interferogram of Etna from 5-6 September 1995 shows atmospheric fringes surrounding summit Etna good location for studying atmospheric errors due to high relief and coastal climate
Possible Mitigation Methods Stacking Time series methods PS methods Calibration with GPS water vapour estimates Removal of atmospheric errors Calibration with spectral water vapour estimates (e.g. MERIS) Statistical Calibratory Forward atmospheric modelling
Advantages of atmospheric modelling approach Requires no field operations or ground-based equipment Available quickly after image acquisition. Requires no back-catalogue of SAR data Does not require daytime or cloud-free conditions Available anywhere in the world MERIS image showing cloud obscuring Etna
Unified Model Horizontal resolution Initialisation Surface characteristics Water vapour representation Convection scheme Time step Vertical levels NH3D 1.7km Radiosonde, or single UM grid point No ground exchange Passive tracer None 4sec 30 Unified Model (UM) 1km Global model, nested domains Met Office Surface Exchange Scheme (MOSES II) Mixed phase scheme with prognostic rain Mass flux scheme in outer domains 30 sec 38
UM Nested Domain scheme Three model domains nested within the UM global model. Outer domains provide initial state and boundary conditions for the higher resolution domains nested within Retains spatial continuity of larger fields when downscaling Global 1 2 3 Grid length 60km 12km 4km 1km Timestep 20 min 5min 1.33 min 30 sec Boundary condition update 1 hour 1 hour 15 minutes
Initialisation NH3D used local radiosonde initialisation data, or closest UM global grid point (left), very coarse. Nested domains preserve 3D spatial continuity of the global model, incorporates spatial data from more grid points. Etna 60km Boundary conditions in the nested domains updated regularly throughout the model run. European Centre for Mesoscale Weather Forecasting (ECMWF) global model data assimilates SSMI water vapour data during the model run. NH3D domain with global UM grid points (black squares) at 60km grid spacing, (Webley et al 2004)
Atmospheric Errors Model cross section shows importance of hydrostatic effects due to the intersection of topography with vertical water vapour gradient. Dynamic effects such as turbulence in the lee of the volcano add further atmospheric errors, dependant on synoptic conditions. Cross section of wind (m/s, stream lines) and mixing ratio (g/kg, shaded) at 0900 UTC 24 November 2004
IWV field, 24 th November 2004 A B Wind vectors (m/s) and mixing ratio (g/kg, shaded) at 1500m asl, 0900 UTC 24 November 2004, showing line of cross-section A-B.
Integrated Water Vapour from MERIS Validation against meteorological observations only possible for surface parameters. Satellite measurements of the 2D water vapour field available. MERIS instrument aboard Envisat synchronous with the ASAR Full resolution integrated water vapour (IWV) product has ~300m resolution at nadir. Theoretical accuracy of 1.6mm. Data valid only for cloud-free pixels, unmasked cloud gives anomalously low water vapour values. MERIS pseudocolour (above) and corresponding IWV image showing anomalously low values due to cloud
MERIS with cloud Direct correction of interferograms possible for arid areas (eg. Li et al 2005), but less than 10% of potential images are completely cloud free, due to Etna s high relief and coastal climate. However, partially cloudy scenes can be masked and used to validate the atmospheric model in clear areas 75% Clear dates Proportion of potential interferogram dates which would have been cloud free, derived using MODIS data from 2002-2005. Dark area over summit shows the area has higher probability of cloud compared to the rest of Sicily. 0%
ESA cloud mask (red) is provided: Adequate for thick cloud Poor at edges of cloud Does not detect high cirrus MERIS cloud mask The Low Pressure data flag (green) can be used to mask high cloud. Better at detecting cirrus, but not missed cumulus. Flag wrongly raised over volcano summit due to DEM error However currently a manual mask is used.
25 June 2005 IWV fields Modelled IWV (left) compared to MERIS data (below) for the 25th June 2005, an entirely cloud-free scene. 3.3cm IWV 0cm Good match between model and MERIS IWV fields: Bias= 0.8 mm RMS= 1.9 mm Correlation coefficient= 0.94
Dynamic atmospheric model method for InSAR water vapour correction The 3D atmospheric water vapour field is calculated for each SAR scene using the UM, and integrated along the radar line of sight for each pixel. Difference between the two fields converted to InSAR signal delay δ a,b using empirical scaling factor Q=6.3, based on the effect of water vapour concentration on the atmospheric refractive index. Signal delay field then removed from the interferogram δ a, b = IWV Q = a, b λφ a, b 4π
Etna ASAR Data October 2004 October 2006 Ascending/Descending 38 o N ascending 37 o N 14 o E 15 o E descending 3000 m asl 0
Interferograms 24th Nov 04 29th Dec 04 17th Nov 04 2nd March 05 35 days 105 days Descending Ascending 22th Dec 04 2nd March 05 17th Nov 04 24th Aug 05 70 days 280 days Ascending Ascending
Interferogram and model comparisons 45 Descending unwrapped interferogram (left) and model (right) for a 35 day Nov-Dec 04 interferogram, Baseline 171m. Model also masked for comparison. mm -45 30 mm -30 Ascending interferogram (left) with the model (right) for a 70 day temporal separation, Dec 04 - March 05, 110m baseline. Masked as above.
Land surface representation Refinements to method Surface exchange scheme uses initial soil parameters from regional climatology model, affecting surface energy partitioning. Poor quality soil moisture data has been shown to introduce a cold bias in night-time temperatures, affecting the water vapour model. Initialisation data European Centre for Mesoscale Weather Forecasting (ECMWF) produce a global model which assimilates SSMI water vapour data. Diagnostic runs show initialising with the ECMWF data reduces the systematic dry bias in the UM model.
Refinements to method Temporal Sampling Nov best fit June best fit Radar acquisition Temporal errors in the advection of synoptic water vapour gradients, (eg. fronts) can be present in model. Can produce a large apparent error in the model when considering only the time step closest to image acquisition. Where partial MERIS data is available, this can be used to compensate for the temporal error. Correlation, RMS and bias between MERIS and modelled PWV, red lines represent 24 th November 04 case, green lines are 25 th June 05.
Deformation at Etna Deformation associated with the 2001-02 and 2002-03 eruptions is well documented. The latest eruption took place from September 2004 to March 2005 Early evidence suggests the event represented passive emptying during an overall phase of recharge. Patanè et al 2005
Long timespan interferogram 17 th Nov 04 24 th Aug 05, baseline 71m Around four fringes of apparent subsidence centred on summit. Typical small wavelength deformation and hydrological features also visible. Initial IWV fields from ECMWF global data suggest signal could be deformation. Comparison with other interferograms suggests deformation may be present in the Dec-March interferogram. +2π LOS (rad) -2π
Conclusions and further work Comparison of UM and NH3D models UM has improved representation of physics, higher spatial resolution, better initialisation data and retains spatial continuity. Comparison with available cloud-free MERIS data shows good agreement Recognised deficiencies in the modelling Surface representation needs improvement Initialisation data should assimilate available water vapour data Temporal mismatch between model simulation and radar acquisition Aim to produce atmospheric corrections for all individual interferograms, and compare results with time series and GPS representation of water vapour field.
Nov-Aug, Aug is wetter all over. Nov summit 2 plain 15 Aug summit 4 plain 30 So difference of 2 at summit and 15 on plain, differential of 13mm PWV, ~8cm apparent defm, ~ 3 fringes Apparent uplift?
Interferograms na.a041117.a050302.4lks.filt etna.a041117.a050824.4lk s.r-00117.red.filt etna.a041222.a050302.4 lks.orb.int.filt etna.a050126.a05040 6.4lks.int_r- 0048_az000132.red.filt etna.a050302.a0508.red.filt etna.d041124.d0412 29.4lks.orb.red.filt etna.d041124.d0502 02.r843.int.filt etna.d050413.d 050518.4lks.int. red.filt etna.d050622.d050 727.4lks.orb_r002_ ax-00006.int.red.filt