SYNERGIC USE OF EO NWP AND GROUND BASED

SYNERGIC USE OF EO, NWP AND GROUND BASED MEASUREMENTS FOR THE MITIGATION OF VAPOUR ARTEFACTS IN SAR INTERFEROMETRY N. Pierdicca 1, F. Rocca 2, P. Basili 3, S. Bonafoni 3, G. Carlesimo 4, D. Cimini 5, P. Ciotti 4, R. Ferretti 4, F. S. Marzano 1, V. Mattioli 3, M. Montopoli 4, R. Notarpietro 6, D. Perissin 7, E. Pichelli 4, B. Rommen 8, S. Reising 9, G. Venuti 2 1 Sapienza Univ. of Rome, 2 Politechnic of Milan, 3 University of Perugia, 4 Univ. of L'Aquila, 5 IMAA-CNR, 6 Politechnic of Turin, 7 Chinese Univ. of Hong Kong, 8 ESA/ESTEC 9 Colorado State University

Introduction § § SAR interferometric maps displays not only very small terrain motions, but also the so called Atmospheric Phase Screen, i. e. , the excess path delay mainly associated to the variable columnar amount of water vapor at the time of the passage, that delays the e. m. wave. The atmospheric contribution to the interferograms is a major artifact to be corrected to get reliable motion estimates. By PS multipass techniques and for long term displacements APS’s can be retrieved and could provide a valuable information on the atmospheric conditions at very good spatial resolution Note that interferometric measurements are doubly differential as they are time differences referred to a single point in each image, that is constrained to be stationary in time. Outline § § I will briefly review the METAWAVE (Mitigation of Electromagnetic Transmission errors induced by Atmospheric Water Vapor Effects) ESA sponsored study. We focus on the exercise to assimilate APS derived information on water vapor within Numerical Weather Prediction models 2

METAWAVE project § Funded by ESA/ESTEC § Objectives ü Use any additional information to correct, at list partially (mitigate), the atmospheric WV artefacts in In. SAR ü Assess usefulness of In. SAR for atmospheric applications and particularly weather prediction § Requirements for In. SAR are very demanding: ü resolution order of 100 m ü thematic accuracy order of mm DZWD ( 0. 16 mm DIWV) ü Timeliness 3

Project activities § During the METAWAVE project several techniques have been exploited to map path delay due to water vapor: ü ü ü Numerical Weather Prediction (NWP) models Earth Observation products Ground based microwave radiometers and GPS receivers Data interpolation and downscaling processing techniques Tomographic techniques § Two experiments were set up ü Rome area: regional scale applications and ground based microwave radiometers (including Colorado State University CMR for 3 -D tomography) ü Como area: local scale applications by exploiting a dense network of GPS receivers 4

Regional scale: Rome experiment setup § Radiosonde launch Colosseo ü 8 launched from Sapienza, 6 successful ü 4 Daytime, 2 nighttime ü 75 from nearby operational station § Radiometers and LIDARS ü One 2 -channel + one 4 -channel ü LIDAR nearby ü Two other 4 -channel to form a triangle 5 km 10 km 5

Local scale: Como experiment setup Regional operational GPS network Intermediate GPS network Local network 6

From PS to APS § On land, numerous scatterers exist that maintain their scattering characteristics very stable in time (the Persistent Scatterers: PS). § Once these points are detected, their apparent motion can be recorded from the phase of the radar returns with millimeter precision. § The atmosphere, particularly due to the high water vapour spatial and temporal variability, introduces an unknown delay in the signal propagation (the Atmospheric Phase Screen: APS). § Under certain hypothesis and using a huge number of interferograms the apparent motion of the PS’s due to the APS‘s can be singled out wrt to real displacements and APS’s can be estimated 7

Samples of APS’s of PS’s Winter Summer 8

The APS and the atmosphere § Atmospheric path delay L can be converted into (columnar) Integrated Water Vapor (IWV) and passed to the meteorologists. § In. SAR could become a tool for high resolution water vapour retrieval and provide routinely water vapour maps to be assimilated into high resolution Numerical Weather Prediction (NWP) models. § A major difficulty is associated to the differential nature of the APS's provide an insuperable high resolution mapping of the atmospheric path delay differences (i. e. , in time and space) over stable PS’s, but they do not furnish absolute values. § This difficulty can be overcome by relating on external information providing suitable climatological values in order to provide the reference atmospheric signal associated to the master SAR image of the interferometric stack which cannot be known using SAR data only. 9

From APS to water vapour § DIn. SAR interferometric phase contains DISPLacement phase F and excess path ATMOspheric delay L differentiated wrt time i and j and referred to point x 0 § For a steady (or known motion) surface, DFDISPL =0 (or known) and an image sequence provides the atmospheric delay (APS) in each point x wrt a unique master j=M, with arbitrary unknown consti=(l/4 p)Di. MF (x 0): § L has a dry and a wet component, the latter proportional to the Integrated Water Vapour (IWV) One could derive the “absolute” atmospheric delay at time i, and than the wet contribution proportional to IWV, assuming the master contribution is known by EXTernal sources (e. g. , NWP, EO products): § § The associated error variance is a combination of APS error and EXTernal source error, which can be significant: s 2 L= s 2 APS+ s 2 EXT 10

From APS to water vapour § Alternatively, by averaging many APS’s and corresponding EXTernal information : § Again using an EXTernal source one can estimate the master contribution and than the actual “absolute” atmospheric delay from APS by: § Which is more reliably since the associated error variance depends on error variance of the mean L field provided by the EXTernal source, which is much smaller than that of an individual LEXT field: s 2 L= s 2 APS+ s 2 Mean. APS+s 2 Mean. EXT § § There still an ambiguity due to const which can be removed by relying on EXTernal sources provided by NWP or independent EO products (e. g. MERIS). Absolute path delay, or its wet component proportional to IWV, can be assimilated into NWP. 11

The water vapour maps from APS § The local circulation in the urban area of Rome was studied using the high-resolution Mesoscale Model (MM 5) and Weather Research and Forecasting (WRF) model. § These are well established fully compressible nonhydrostatic NWP models, reaching resolution of 1 km or higher. § A long sequence of ASAR images has been processed to derive APS maps, subsequently converted to absolute IWV maps by using as climatological background and bias (const) correction the MERIS IWV products. § Basically In. SAR provides small scale components of the vapour field, whereas the large scale (including mean) of traditional meteo data is retained 12

The water vapour maps from APS § § APS turned into absolute IWV (left panel), known except a constant, using MM 5 as external background and then embedded into MM 5 map (right panel) Note higher map resolution inside the SAR frame 13

Assimilation experiment Questions: ü Are the NWP outputs sensitive to the assimilation of APS small scale features ? ü Is the effect positive ? § § MM 5: 4 domains 2 way nested, 1 km resolution for inner domain, 33 s levels Initialization: warm start to assimilate APS derived IPWV on 3/10/2008 at 9: 30 UTC, being the ECMWF initial conditions only available at synoptic hours MERIS used as completely independent background to produce IWV from APS Case study during the Metawave experiment in Rome to be compared to ground truth (October 3, 2008) MM 5 nested domains 14

Assimilation of In. SAR derived IWV § § § xb is the atmospheric “background” status B and O are the covariance matrix of background field and observation errors H(x) is the observation operator and Qv is the mixing ratio [g/kg] 15

MM 5 assimilation results MM 5 without APS § MM 5 (1 km) hourly rain rate field at 17: 00 UTC compared to interpolated raingauge (right panel) at 20: 00 UTC § SAR assimilation (top-right) doesn’t correct MM 5 time anticipation (3 h) of maximum rain. § It reduces overestimation (>18 mm/h vs observed 16 mm/h) produced by MM 5 without SAR (top-left) MM 5 + APS Interpolated rain gauges 16

WRF assimilation results WRF without APS WRF + APS § WRF time anticipation only 1 hour in both cases § SAR assimilation (top-right) again better reproduces the observed rainrate values Interpolated raingauges 17

Conclusions § The METAWAVE project has been quickly reviewed § A strategy to assimilate In. SAR APS into NWP has been illustrated § It was shown that, using APS, NWP forecasts can be improved, albeit slightly. More can be done: ü Running models at higher resolution to retain the high frequency structures of In. SAR data. ü Assimilating directly the path delay L instead of IWV ü Making the warm start more reliable by using local observations ü Avoiding the 3 DVAR geostrophyc adjustment of the meteorological fields to the large scale § Future availability of low revisit time SAR’s (e. g. , Sentinel 1) could ring the change for the use of In. SAR for meteorological purposes 18

19

Triple collocation § It is useful to validate measuring systems when none of them are immune from errors (apart biases). § Suppose three measurement systems X, Y, and Z measuring a true variable t (IWV in our case). § dx, dy , dz are random zero-mean observation errors § Errors have variances ex 2, ey 2, ez 2 § sy, sz are scaling factors § It is assumed that systems X and Y can resolve smaller scales than system Z by introducing r 2 as the variance common to these smaller scales. § Variable t with variance s 2 refers to the large scale features of the observed field and has § By collecting observations collocated in time and space, from a statistical analysis it is possible to estimate error variances and scaling factors 20

Triple collocation results § We applied the triple collocation techniques to data collected during METAWAVE in Rome § We identified GPS (X) and MERIS (Y) as the systems providing IWV at the smaller scale, whereas MM 5 (Z) is supposed to be less resolved § Assuming different hypothesis for r 2 we found the following: § MERIS exhibits smaller random error (≈0. 8 mm IWV or 5 mm L) wrt GPS (≈1. 1 mm IWV, 7 mm L) and calibration factor similar to GPS (0. 99) § MM 5 shows larger error (≈ 1. 5 mm) and underestimation of IWV (0. 88) 21

Project rationales § § As for no sudden, long term, ground motion, multi-pass technique can mitigate path delay artifacts and provide APS to meteorologists, to be exploited for other applications. For sudden, short term, ground motion and/or traditional (few passes) In. SAR, two frameworks are identified: ü Regional scale applications • Relaxed spatial resolution (goal 1 km), integration of many data sources by interpolation/downscaling, and NWP model assimilation ü Local scale applications • Small coverage, support of ground based systems (e. g. , GPS, radiometers) Long term Short term Regional scale Local scale Averaging possible. Are APS useful for WV mapping ? Need for high-resolution nearby PS estimates Averaging not possible. Need for non-SAR APS estimates ! 22

SAR and Permanent Scatterers (PS) Arno R 1 R 2 Pisa Livorno Migliarino Serchio Dr VS VS PS 23

DF l/4 p [mm] PS displacement time series Non linear motion or noise or atmospheric delay residuals (APS) Linear motion 1 rad≈5 mm 24

Atmosphere properties from APS An example of using APS for atmospheric studies, § Semivariogram of the APS’s derived from a huge number of interferograms, providing an insight into the spatial distribution of the water vapour at small scale (Ferretti et al. ). 25