Ocean Remote Sensing with Reflectometry: Current Status and Future Directions Keynote James L. Garrison Purdue University, West Lafayette, USA 4th International Conference on GPS Radio Occultation (ICGPSRO2018) Taipei, Taiwan April 18-20, 2018 1
Outline Ocean reflectometry: Basic Principles Spaceborne reflectometry on CYGNSS Current Status: Ocean winds retrievals Future Directions: Sequential Estimation Improved GMF Development Data Assimilation Non-GNSS Signals of Opportunity Conclusions 2
GNSS-R Basic Principles Receiver glistening zone glistening zone 3
Rough Surface Scattering Phenomenon observed in visible light at sunset: (water is calm inside the red ellipse) (From Chapron and Ruffini, 2003 GNSS-R workshop, Barcelona. Photo taken at Le Conquet, Brittany) 4
Rough Surface Scattering Barrick [Proc. IEEE, V 56, No 10, 1968] showed cross-section proportional to slope PDF Physical interpretation: Probability of surface slope Giving specular reflection In direction of receiver Green = normal to surface Red = bisector (q) vector 5
Rough Surface Scattering Delay-Doppler Map (DDM) Integral over surface Masking of Surface by delay-doppler mapping Masking of Surface by Receiver antenna Path loss Surface-receiver Cross-section Proportional to Probability of slope scattering in correct direction Path loss Trans-surface 6
Waveform Features Peak Power Trailing-Edge Slope (TES) Leading-Edge Slope (LES) Peak Delay 7
Early Airborne Wind Retrievals [Garrison, et al TGARS 2002]: Measured Waveform Best Fit of Model Waveform [Garrison, et al. GRSL 2011]
GNSS-R Model Functions [Katzberg, et al, GRL 2006]
NASA Earth Ventures Mission 8 satellite GNSS-R constellation Improved forecast of tropical cyclone intensification: 1. Better penetration of rain (L-band vs K-band) 2. Higher revisit rate (2.8 H med. vs 11-35 H) CYGNSS 10
CYGNSS Simulated Coverage of Hurricane Frances: ASCAT vs. CYGNSS C. S. Ruf, et al., Bull. Am. Met. Soc., V. 97, N. 3, pp. 385 395, 2016. DOI: 10.1175/BAMS-D-14-00218.1 11
Going to Space Maximum Delay for 25 km resolution 12
GNSS-R from Space 25 km resolution requirement: 3 X 5 pixels used in L2 retrieval Downloaded to the ground: 17 x 11 pixels around specular point 13
GNSS-R from Space Typical space-borne GNSS-R Observable Decrease in sensitivity with wind speed N. Rodriguez-Alvarez and J. L. Garrison, TGARS 2016. DOI: 10.1109/TGRS.2015.2475317 14
CYGNSS Observables Ruf, et al, IEEE JSTARS 2018 DOI:10.1109/JSTARS.2018.2825948 15
CYGNSS Retrievals (U<20 m/s) Groundtruth: ECMWF 30.9 M matchups 1.96 m/s RMS (incl. error in ECMWF & interpolation) Block IIR IIR-M (IIF Excluded) Ruf, et al, IEEE JSTARS 2018 DOI:10.1109/JSTARS.2018.2825948 16
CYGNSS Retrievals (U>20 m/s) Matchup: 674 pairs RMS = 6.45 m/s (ensemble statistics) Ruf, et al, IEEE JSTARS 2018 DOI:10.1109/JSTARS.2018.2825948 17
Calibration very important: Transmitter EIRP Antennas CYGNSS Winds: Discussion Block II-F (variable power) currently excluded Sensitive to both winds and waves - model depends on fetch/wave age, consider: Coupled wind/wave retrievals [Clarizia and Ruf, JAOT 2017, DOI: 10.1175/JTECH-D-16-0196.1] Roughness as the retrieval variable [Garrison, et al. GRSL 2011] Sensitivity decreases with winds (a lot for U > 20 m/s) 18
Outline Future Directions: Sequential Estimation Improved GMF Development Data Assimilation Non-GNSS Signals of Opportunity Conclusions 19
Sequential Estimation 20
Sequential Estimation Observation vector (y): DDM samples @ 1Hz (36x20) State vector (x k ) = 10 km gridded ocean mss (related to surface wind field using Katzberg model) Observation equation (h(x)): Discretized Zavorotny-Voronovich Integral 2 h( i, fd, tk, x) Bp( tk ) p p i, fd, tk ( i, fd, tk, x) j i i p Constant Properties Of Pixel p Ambiguity Function Averaged Over Pixel p Linearized Observation equation (Jacobian) (H(x)) Slope PDF At Pixel p 21
EKF Results: 2/12/2018 22
Data Assimilation: The Way Forward Experience from GNSS Radio-occultation (GNSS-RO): Water vapor -> Refractivity -> bending angle Direct inversion (bending angle -> refractivity) assumes uniform properties over area covered by the integral Alternative: assimilate bending angle directly into weather models. 23
Data Assimilation: The Way Forward Reflectometry (GNSS-R): Winds -> MSS -> DDM Similar problem can we assimilate DDM s directly into forecast models? NASA Grant: NNX15AU18G Assimilation of GNSS-R Delay-Doppler Maps into Hurricane Models 24
Validation of Forward Model Tests on CYGNSS data from hurricane Irma on Sep. 4, 2017 case1 case2 33 rd AMS Hurricane Conference Ponte Vedra, FL, Apr 17, 2018 25
Validation of Forward Model Tests on CYGNSS data from hurricane Irma on Sep. 4, 2017 Case1: near hurricane, U ~ 30 m/s. [33 rd AMS Hurricane Conference Ponte Vedra, FL, Apr 17, 2018] 26
Validation of Forward Model Tests on CYGNSS data from hurricane Irma on Sep. 4, 2017 Case2: medium wind speed, U ~ 20 m/s. [33 rd AMS Hurricane Conference Ponte Vedra, FL, Apr 17, 2018] 27
Validation of Forward Model Tests on CYGNSS data from hurricane Irma on Sep. 4, 2017 Case 3: low wind speed, U ~ 5 m/s. Can adjust noise floor in forward model for better match-up Currently testing with more accurate CYGNSS antenna patterns [33 rd AMS Hurricane Conference Ponte Vedra, FL, Apr 17, 2018] 28
Tests on CYGNSS data near Cyclone Enawo, Mar 6, 2017 Case4: very low wind speed, U < 2 m/s. Validation of Forward Model [33 rd AMS Hurricane Conference Ponte Vedra, FL, Apr 17, 2018] 29
Why GNSS? Continuous global coverage L-band good penetration of atmosphere, vegetation, rain, etc Pseudorandom noise (PRN) code designed for ranging My career began with GNSS!
What about other signals? Approximately 400 communication satellites in GEO High-powered (~30 db above GNSS) signals Allocations in most bands used for remote sensing: L,S, C, Ku/Ka Designed for data transmission Not ranging! Assumption: Compression & Encryption are very efficient at filling available spectrum Data is nearly random Direct signal can be used as reference
(Self-) Ambiguity Function (SAF) Measurement PSU PC DirecTV Antennas USRPs AMPs
2014 Hurricane season: 2-Jul-2014 to 17-Sep-2014 5 Hurricanes: Arthur, Bertha, Cristobal, Dolly, and Edouard S-band Satellite Digital Audio Radio Service (SDARS) signals Signals of Opportunity (SoOp) Winds Goals: Develop empirical PDF model for S-band (2.3 GHz) Assess high wind sensitivity of S-band reflectometry Processor RF front end, A/D Sampler 33
SoOp Hurricane Winds Wind Speed Retrieval Cox & Munk Model MSS u (U 10 ) = 0.45(0 + 0.00316 U 10 ) MSS c (U 10 ) = 0.45(0.003 + 0.00192 U 10 ) Katzberg s model (L-band empirical correction to C&M) U 10 = U 10true 0<U 10true <3.49 U 10 = 6 ln U 10true 4 3.49<U 10true <46 U 10 = 0.411 U 10true 46<U 10true 34
SoOp Hurricane Winds [Zhang, et al., GRL in preparation] 35
SoOp Altimetry Sample coverage in 10 days from traditional altimetry orbit Ho, et al., PIERS 2017 Sample coverage in 3 days with 6 receivers in LEO observing transmitting constellation: DBS 9
Ku/K-BAND Antennas Experiment Description Beacon 27 m Receiver at JPL Shed Platform Harvest Ho, et al., PIERS 2017 9
SoOp Altimetry (Ku-Band, LHCP) Theoretical Experimental SNR(dB) 39.5 40.5 σ (cm) 1.53 2.78 Theoretical Experimental SNR (db) 36.1 36.5 σ (cm) 1.50 2.58 Ho, et al., PIERS 2017 13
SoOp Altimetry (Ku-Band, LHCP) σ = 5.63 cm Ho, et al., PIERS 2017 16
Summary Spaceborne GNSS-R wind retrievals successfully demonstrated under some conditions. Sensitivity decreases for high winds (esp. U > 20 m/s) Calibration remains a big concern Perhaps wind speed is not the best variable: Wind/Wave coupling Fetch/Wave age dependence Direct DDM Assimilation GNSS-R just the beginning - many other signals are out there! Diversity of frequenices High EIRP (+30 db Vs. GNSS) 40
Acknowledgements NSPO generously provided for Prof. Garrison s travel expenses to attend ICGPSRO-2018 41