Status and description of Level 2 products and algorithm (salinity)

Transcription

1 SMOS 7th Workshop, ESRIN, Frascati, Oct Status and description of Level 2 products and algorithm (salinity) J. Font, J. Boutin, N. Reul, P. Waldteufel, C. Gabarró, S. Zine, J. Tenerelli, M. Talone, F. Petitcolin, J.L. Vergely, S. Delwart

2 SMOS 2-D Field of view (1 FOV / 10) (F. Petitcolin, ACRI-st) SEPSv3 simulations S90 S80 S70 S60 S50 S40 S30 S20 S10 S1 216s 192s 168s 144s 120s 96s 72s 48s 24s T0 Satellite pass: 06 & 18 h local time (Equator)

3 Measurements variability within FOV Multi-angular and spatially variable nature of the measurements: incidence angle (dashed lines) ranges from 0 to 65º spatial resolution (dash-dotted lines) from 32 to 100 km radiometric sensitivity (dashdotted) from 1.8 K at boresight to 5 K. data interpolated into an ISEA grid (approx. 15 km spacing) Figure generated by the SMOS End-to-end Performance Simulator (UPC, 2004)

4 Sea surface emission at L-band At 0 incidence and flat sea Sensitivity of T b to S over a smooth surface is 0.2 to 0.8 K/psu depending on ocean temperature, incidence angle and polarization SSS retrieval more difficult at high latitudes

5 Emitting surface geometry Influence of incidence angle and surface roughness on emissivity at 0 wind at 40 psu (Camps et al., 2001)! effect of 5 psu ~ 10 m/s Observation geometry to be carefully controled

6 SMOS SSS retrieval approach To model the polarised L-band emission of the ocean surface using a guessed SSS value Auxiliary data needed to describe the environmental conditions Compare modelled Tb with SMOS measurements for all available angles Reach the best fit through iterative modifications of guessed values

7 SSS retrieval algorithm The overdetermination of T b at a fixed ocean location can be used to reduce the measurement noise and to adjust several geophysical variables in the iterative minimization process The forward model to be used considers separately the emission of a flat sea (well determined at L band) plus the effect of the roughened surface, parameterized using external information T b,p = T b,p flat (!, SST, SSS) + "T b,p rough (!, ", SST, wind waves, swell, other wave characteristics, foam coverage, foam emissivity, rain) The measured polarized T b will be due to this emission of the sea surface, plus emission from other sources (directly or reflected on the surface), plus other effects on the radiation when travelling from the surface to the SMOS antenna (atmospheric absorption, Faraday rotation )

8 SSS Level 2 processing scheme As being implemented in the Processor Per ISEA grid point on each half-orbit: Preprocessed (interpolated) auxiliary data (SST, wind, ) from ECMWF For all T b measurements available (up to 78) Decision on processing/flagging according to several tests Forward model for T b at each measurement angle!: Emissivity of flat sea at! with auxiliary SST and guessed SSS Add roughness effects (3 model options + foam) Add external noise (reflected signals: atmosphere, galactic, ) Add atmospheric attenuation and emission to antenna Transport modeled T b to antenna level (including Faraday from TEC) Compare modeled with measured T b Iterative convergence Output values (adjusted): SSS, SST, WS, TEC (depending on options)

9 Pre-processing Set up configuration and read input files (L1c, Look-up Tables) Geometry: Compute angles using EE CFI v3.6 for Satellite to target direction Target to Sun direction Target to Moon direction Target to sky, specular to viewing direction Specific angle for galactic noise 2 model Forward Model support: Atmospheric effects Measurement Discrimination: Set up flags at measurement level (Sun glint, outliers, ice, moon, ) => Compute first guess of measurement using forward model (configuration 1), ECMWF data and SSS climatology Set up flags at grid point level (retrieval conditions, valid measurements)

10 Processing and post-processing Loop: 3 +1 Configurations Loop: Selected grid points Compute priors, uncertainties on priors and first guess for iterative scheme Retrieval Compute uncertainties on retrieved parameters Compute brightness temperature at 42.5 deg. incidence Set quality flags and descriptors End loop End loop Write User Data Product and Data Analysis Product Write product reports

11 Grid points selection

12 Measurements selection

13 The sea surface L-band emission Quite well modelled for a flat sea (dielectric constant + geometric optics) Roughness effects, the main issue Consider emission from other sources that reach the radiometer Atmospheric effects

14 Modular Forward Model Atmosphere Ocean! Sea surface emissivity models Dielectric constant of sea water (Klein and Swift, 1977) 3 roughness models Model 1: Dinnat et al., 2002 (2-scale, Durden and Vesecky spectrum#2) Model 2: Johnson and Zhang, 1999 (SSA, Kudryavtsev spectrum) Model 3: Gabarró et al., 2004 (empirical) Foam (Reul and Chapron, 2003)! Other contributions Atmosphere Tropospheric (from Liebe, 1993) Faraday rotation (Waldteufel et al, 2004) Sky radiation (reflected / scattered), sun glint (Reul et al., 2007)

15 Pseudo-dielectric constant Modified cardioid model (Waldteufel et al., 2004): Difficult to retrieve real and imaginary part of dielectric constant Change of variables $' = A_card (1 + cos(u_card) ) cos (U_card) + B_card $" = A_card (1 + cos(u_card) ) sin (U_card) Assuming a simplified emissivity model (flat sea) and standard corrections, A_card can be retrieved with good accuracy (and independent from roughness corrections) A_card values very different between ice and sea water Retrieved A_card can be used for sea ice detection, better than other implemented methods

16 Full forward model flat sea sw1 cardioid All four Stokes parameters computed Roughness 2-scale sw2 Roughness SSA Roughness empirical Dual pol. => Tx / Ty Dual pol. => pseudo Stokes 1 IPSL/ LOCEAN Galactic Noise 0 sw3 Galactic Noise 1 Galactic Noise 2 Full pol. => Tx / Ty / St3 / St4 Full pol. => Tx / Ty Full pol. => pseudo Stokes 1 IFREMER sw4 foam ICM-CSIC Sum of contributions From target to antenna Atmosphere

17 Surface roughness effects Three different roughness model options have been implemented in the algorithm: -Two theoretical models, including the statistical description of the sea surface and the asymptotic solution for electromagnetic scattering (twoscale approach, and small slope approximation), - A semi-empirical formulation derived from the WISE campaign data set. It fits measured polarized Tb to different combinations of available roughness descriptors An additional specific model has been developed to account for the effect of foam on the sea surface emissivity (thickness and coverage) During SMOS commissioning phase (and beyond, if necessary) the different options, and suboptions, will be tested until identifying the best one. This may not be the same for different ocean regions or seasonal conditions

18 Surface roughness effects The figure shows the sensitivity of L-band sea surface emissivity to wind speed as function of incidence angle. Over measured Th and Tv data we plot predictions from SSA (blue) and two-scale (green) models at SST=15 C and SSS=35 psu. The red curve shows a best-fit through the observations. It appears that all available data is not enough to discriminate the best adapted correction between these three models of the roughness impact.

19 Roughness models Horizontal polarisation Vertical polarisation Tb rough (K) Model 1 Model 2 Model 3 #= Tb rough (K) Model 1 Model 2 Model 3 #= Wind speed (m/s) Wind speed (m/s) WS > 3 m/s: models 1 & 3 ~ linear in wind speed model 2 strongly non linear foam impact significant for WS > 10 m/s

20 Parameters in the roughness models R o u g h n e s s model Dual pol Dual pol + Stokes 1 Full pol Full pol + Stokes 1 1. Two-scale SSS SST TEC WSx WSy SSS SST WSx WSy SSS SST TEC WSx WSy SSS SST WSx WSy 2. SSA SSS SST TEC U* % $ SSS SST U* % $ SSS SST TEC U* % $ SSS SST U* % $ 3. Empirical (2 or 4 roughness prameters) SSS TEC WS Hs % U* MSQS SSS WS Hs % U* MSQS SSS TEC WS Hs % U* MSQS SSS WS Hs % U* MSQS

21 Galactic noise correction Three models are implemented in the L2 OS processor: GN0 model: constant incidence galactic noise of 3.7 K multiplied by specular reflection coefficients GN1 model: incidence galactic noise obtained from Reich & Reich map convoluted by the mean WEF and multiplied by specular reflection coefficients GN2 model: reflected galactic noise obtained from Reich & Reich map convoluted by bistatic coefficients according to the measurement geometry and the wind speed

22 Galactic noise correction Galactic Noise model 1 Galactic Noise model 2 with WS = 3 and 10 m/s, at 0º

23 Galactic noise correction Ascending orbits Descending orbits

24 Radiative transfer function: Tb m = Tb s e - %atm + Tb up + & Tb down e - %atm Tb m measured Tb Tb s Tb emitted from surface % atm equivalent optical thickness of atmosphere Tb up emitted upwards by atmosphere and attenuated along upward path & surface reflection coefficient Tb down emitted downwards by atm. and attenuated along downward path 4 atmospheric components to be considered in computing % atm, Tb up and Tb down : dry atmosphere water vapour clouds rain available models implemented Atmospheric effects

25 Transformation from surface to antenna From T H, T V to T X, T Y including atmospheric effects and Faraday rotation Faraday rotation! '= f (total electron content TEC,...) Stokes 1 I = T H +T V = T X +T Y

26 Iterative retrieval algorithm Cost Function to be minimized: * 2 = N! i= 1 meas ' Tbi ( % & ) i Tb mod i $ " # 2 + K! k = 1 ' Pk ( % & P ) k k 0 $ " # 2 Tb meas : Tb measured Tb mod in the antenna reference frame : Tb from a direct forward model N: number of Tb observations P: geophysical parameters driving Tb variations (depend on forward model) ( i : errors on Tb meas Gaussian assumption ( k : prescribed errors on auxiliary parameters Four retrievals: 3 salinities (3 roughness models) + Pseudo-dielec. const. Retrieved parameters: SSS/A_card, SST, roughness parameters, TEC in dual pol mode Minimization: Levenberg-Marquardt algorithm + convergence criteria

27 The SMOS Salinity Prototype Processor Tb measurements (level 1C, on fixed grid) Auxiliary data: SST, wind, Preprocessing (sort & flag): coast, ice, heavy rain, sun, moon, sky radiation, outliers Inversion: Iterative method Salinity, SST, wind, (level 2) Forward models: flat sea emissivity, roughness (3 models), foam, sky radiation, atmosphere, geometry

28 SMOS SSS L2 processing scheme Input data files preprocessor ECMWF files LUT and coef. files L1C product SMOS-ECMWF files local files Configuration files SMOS L2 SSS Prototype processor Product files User Data Product Data Analysis Report Breakpoint Reports

29 The SMOS SSS L2 User Data Product SSS computed at each ISEA grid point for a semi-orbit (ascending or descending) Product structure (per grid point): Identification and pixel characteristics Three SSS values (with uncertainties) Pseudo-dielectric constant retrieved WS and SST used in retrieval Polarised Tb at 42.5º at surface and antenna Flags and confidence and science descriptors

30 Overview of L2 OS User Data Product Loop grid points: ISEA grid point ID, latitude and longitude Equivalent footprint diameter Mean acquisition time SSS and ( SSS (x3) A card and ( Acard (new) Wind Speed and SST from ECMWF and uncertainties Modelled surface and antenna Tb at 42.5 deg, and uncertainties Control Flags (41) Descriptors (25, 1 byte each) Retrieval fit quality index (x3+1) High value acceptability probability (x3+1) SSS Quality index (x3) Number of iterations (x3+1) Number of valid measurements for retrievals Number of valid measurements with flag raised: BORDER_FOV, EAF, AF, Sun, Moon, etc. Science Flags (22), Science Descriptor (1)

31

32

33

34

35 Overview of L2 OS Data Analysis Product Loop grid points: ISEA grid point ID, latitude and longitude Out of Look-Up Table (x3 +1) Distance to satellite track Discarded measurements: outliers, large footprint, RFI(L1), Sunglint (L1) Atmosphere: thickness and equivalent temperature at nadir Priors and uncertainties on priors (x7x(3+1)) Results and uncertainties (x7x(3+1)) Loop Measurements: Differences in brightness temperatures (x3+1) Galactic noise contributions (H and V polarizations) Total: 112 grid point parameters, 5 descriptors, 29 flags 5 measurement descriptors, 19 flags

36 Algorithm Validation: simulations 15 km Homogeneous area: & 10 in latitude & full swath width (1200 km) & 7300 grid points & ascending orbit Different configurations selected area Sample of the hexagonal grid Ocean surface with P true = {SSS, SST, wind,...} " Simulated brightness temperature measurements : Tb meas = Tb mod (P true ) +! Tb " Simulated geophysical parameters measurements : P meas = P true +! P

37 Algorithm Validation: tests We study the effects of: Noise on the geophysical parameters P meas = P true +! P Bias on the geophysical parameters P meas = (P true + B P ) + ' P Sea surface conditions 2 polarisation modes: Dual polarisation: T x, T y " TEC retrieved First Stokes parameter: I = T x + T y ~ not sensitive to Faraday rotation

38 Algorithm Validation Plan Priority Scenarios 1.1 Noise study 1.2 Bias study 1.3 Model cross checking 2.1 Valid measurement selection 2.2 Sea surface type 2.3 Galactic noise correction 2.4 Galactic noise mask 3.1 Coast effect 3.2 Ice effect 3.3 Cardioid Model Objectives Sensitivity to uncertainty on priors Sentitivity to biases on priors Sensitivity to model error and bias Tune thresholds for measurement selection Test assumption of homogeneity, sensitivity Validation of galactic noise corrections Tuning of galactic noise mask Tune threshold: how close to the coast can SSS be retrieved? Sensitivity to Ice, tune threshold for detection Sensitivity study

39 SSS retrieval behavior for the 3 models N = 7300 SSS = 35 psu SST = 15 C WS = 7 m/s TEC = 10 TECu ( SSS = 100 psu ( SST = 1 C ( WS = 1.5 m/s ( TEC = 5 TECu Good behaviour of the retrieval - distribution close to Gaussian - theoretical error given by the retrieval = rms error of SSS retrieved SSS true Model 1 Model 2 Model 3 theoretical error given by the retrieval rms error of SSS retrieved SSS true

40 SSS retrieval behavior for the 3 models N = 7300 SSS = 35 psu SST = 15 C WS = 7 m/s TEC = 10 TECu ( SSS = 100 psu ( SST = 1 C ( WS = 1.5 m/s ( TEC = 5 TECu Model 1 Model 2 Model 3 Good behaviour of the retrieval - distribution close to Gaussian - theoretical error given by the retrieval = rms error of SSS retrieved SSS true Higher errors at the edge of the swath " decreasing number of measurements per gridpoint " increasing radiometric noise Lower error for model 2 " lower dependence on WS at moderate WS

41 Errors on SSS retrieved from dual pol and Stokes 1 measurements Dual pol Stokes 1 ( WS = 3 m/s ( SST = 0 C, 1 C, 2 C! WS = 1.5 m/s ( WS = 0 m/s Model 1 SSS = 35 psu SST = 15 C WS = 7 m/s ( WS = 3 m/s ( SST = 0 C, 1 C, 2 C! WS = 1.5 m/s ( WS = 0 m/s High sensitivity of SSS error to errors on roughness parameters SSS error weakly sensitive to SST error Better ability of dual pol to correct for large WS errors

42 Error on SSS retrieved at high & low SST Model 1 SSS = 35 psu SST = 15 C WS = 7 m/s Model 1 SSS = 38 psu SST = 25 C WS = 7 m/s Model 1 SSS = 33 psu SST = 5 C WS = 7 m/s ( WS = 0 m/s ( WS = 3 m/s ( SST = 0 C ( SST = 1 C ( SST = 2 C SSS error increases with decreasing SST Small sensitivity of SSS errors to SST errors lower than 1 C

43 Mean SSS bias (psu) Bias on SSS retrieved with biases on the geophysical parameters Dual pol Model 1 ( SSS = 100 psu SSS = 35 psu ( SST = 1 C SST = 15 C ( WS = 1.5 m/s WS = 7 m/s ( TEC = 5 TECu TEC = 10 TECu B SSS = -0.4 and -0.7 psu for B WS = -1 and -2 m/s Mean SSS bias (psu) Distance to track (km) Distance to track (km) Stokes 1 TEC biases cause asymmetry in SSS biases in dual pol Better ability of dual pol to correct for WS biases close to the center of the swath

44 Algorithm improvements Implementation of neural network approach Auxiliary data other than ECMWF 3 h forecasts Coastal area measurements processing Empirical roughness model optimisation Roughness effects at low winds Other improvements in theoretical roughness models Use more information on sea state/fronts in forward models Polarised sky radiation and galactic noise issues Cost function and convergence strategy Computation of theoretical uncertainties Computation of Tb at surface level Use of in situ salinity data (cal/val feedback) Possible use of Aquarius and ocean models (L4 feedback)