The Ice Contamination Ratio Method: Accurately Retrieving Ocean Winds Closer to the Sea Ice Edge While Eliminating Ice Winds

Transcription

1 The Ice Contamination Ratio Method: Accurately Retrieving Ocean Winds Closer to the Sea Ice Edge While Eliminating Ice Winds David Long Department of Electrical and Computer Engineering Brigham Young University 12 June 212 IOVWST Meeting

2 QuikSCAT Swath and Polar Ice Arctic 28 Antarctica 28

3 Comparison to NCEP Winds Evidence of Ice Winds from RMS wind speed difference between NCEP (or ECMWF) and QuikSCAT. 8 million WVCs / year

4 Problem and Motivation Satellite Scatterometers accurately measure ocean winds but can suffer from contamination from nearby sea ice Sea ice backscatter is much brighter than ocean backscatter Sea ice in the main lobe or nearby sidelobes biases the ocean backscatter measurements Can result in erroneously high wind speeds -- Ice Winds To avoid, processing within 5 km of ice edge discarded Throws a lot of data away Ice winds still occur due to fast moving ice Want to eliminate ice winds and estimate winds closer to sea ice edge Adapt proven land contamination ratio approach for sea ice

5 SeaWinds on QuikSCAT SeaWinds is commonly referred to by its platform QuikSCAT Dual polarization, scanning pencil beam spaceborne scatterometer Transmits and receives at 13.4 GHz VV at 54.1 incidence angle HH at 46 incidence angle Measures normalized backscatter crosssection (σ ) at (up to) four looks Vpol fore Vpol aft Hpol fore Hpol aft Inner/outer swaths High resolution images are created using the SIR algorithm

6 σ QuikScat Wind Overview Conventional and UHR ( K ) = σ 1+η true p 2.5 km UHR-AVE Algorithm 25 km Spatial Response Function Wind Retrieval using GMF Wind Vectors 25 or 2.5 km σ true = σ x, y footprint R R x, y footprint x, y dxdy dxdy

7 Ice Contaminated Winds measurement footprints Sea ice in/near mainlobe of response causes sigma- contamination Goal: detect and reject contaminated slices. Contaminated Wind Speeds sea ice ocean m/s 12/15/4

8 σ true = = where x, y footprint σ x, y footprint i Ice x, y x, y footprint = σ ICR + σ i σ R R R R ICR Assumption : σ x, y dxdy σ dxdy dxdy o x, y x, y ( 1 ICR) = dxdy Ice Contribution Ratio + Ice σ R o Ocean x, y x, y footprint x, y x, y footprint R R R dxdy dxdy dxdy dxdy σ i is constant over the ice within a σ o is constant over the ocean within a footprint. footprint.

9 ICR Processing for Identifying and Discarding Ice Contaminated Measurements Calculate ICR ICR<threshold AVE and Wind Retrieval Find local wind and ice brightness Threshold LUT LR Wind (25 km) HR Wind (2.5 km) σ Measurement (m/s)

10 ICR Estimation ICR E [ ICR] where P P P n n n footprint P n ( ice σ ) ( σ ice) Ice = [ ] I[ n] R[ n] R n R[ n ] P n footprint ( ice σ ) ( ice σ ) R[ n] footprint footprint footprint R[ n ] P R[ n ] ( ice) : prior probability of ice = = n Pn ( ice) Pn ( σ ice) ( ice) P ( σ ice) + P ( ocean) P ( σ ocean) : posterior probability of ice : observation probability n Random Sequence n n

11 Constructing ervational PMFs P ice P σ ocean ( ) σ and ( ) Day 6, 24 Day 12, 24 Day 18, 24 P ( σ ice) P( σ ice) σ (db) Day 24, 24 Day 3, 24 Day 36, 24 P ( σ ice) P ( σ ice) σ (db)

12 P ervation Probabilities ( ocean) σ ( P σ ice) P ( σ ice) Note: Example PMF from DOY , 24 UHR sigma- (avewr) data Actual PMF s generated in the four dimensional measurement space

13 Ice Prior Generation Wind Speed (m/s) P( ice) Prior Probability Ice prior is generated by averaging operational QuikSCAT ice masks within a local sliding time window. Window length of 23 days is used here.

14 Computed Posterior Probability ( ) P iceσ

15 ICR Algorithm (tests individual sigma- for sea ice contamination) Ice Masks Prior 24 Training Set Histograms AVE Posterior σ Measurement Calculate ICR ICR<threshold AVE and Wind Retrieval Find local wind and ice brightness Threshold LUT Monte Carlo Simulation HR Wind LR Wind

16 Monte Carlo Simulation Choose w ICR σ ice GMF w σ ocean Incidence/azimuth angles from cross-track location [ ICRσ + ( 1 ICR σ ]( ) σ Sim = ice ) ocean 1+ηK p Simulation parameters: ICR wind ice backscatter cross track location 15 simulated winds for each parameter set RMS error computed η ~ N(, 1) 15 Realizations σsim GMF 1 w 15 ŵ ( w w ˆ Sim,i ) Sim ε i= 1 RMS = 15 2

17 Thresholding Example Simulation Results Thresholding Example Cross track 7 Ice brightness.375 ε rel ε = Ice ε ε Ice Free Ice Free 1% Relative Error Threshold

18 Simulation Results Simulated ICR Thresholds (db)

19 ICR Algorithm RL Ice Masks Prior 24 Training Set Histograms AVE Posterior σ Measurements Calculate ICR ICR<threshold AVE and Wind Retrieval Find local wind and ice brightness Threshold LUT Monte Carlo Simulation UHR Wind L2B Wind

20 Case Study December 15, 24 Location: South of Africa Contaminated HR Wind Speed (m/s) ICR Processed HR Wind Speed (m/s)

21 Case Study Location: South of Africa L2B Wind Speed (m/s) ICR Processed LR Wind Speed (m/s)

22 October 8, 2 Case Study Location: East of the Drake Passage Wind Speed (m/s)

23 Validation Metrics 1) Stand-off Distance (SOD) SOD = mean( d, d2,..., d 1 N ) Wind Speed (m/s) 2) Relative Error Compares against National Center for Environmental Prediction (NCEP) ε rel ε = ICR ε ε Ice Free Ice Free

24 2/24 Metric Validation UHR 25 km L2B 1, wvcs for 25 km & L2B, 1 million wvcs for UHR

25 28 Wind Distributions 1) ICR winds, L2B agree (Ice-free) Antarctic Wind Probability Distributions 2) L2B winds, ICR agree (Ice-free) 3) ICR winds, L2B disagree (Ice-free) 4) L2B winds, ICR disagree ( Ice winds ) 5133 revs, 4 million wvcs

26 Conclusion New ICR processing retrieves wind an average of 22.5 km (Antarctic) 15.9 km (Arctic) from sea-ice edge Improvement from original 5 km ice-edge masked winds ICR processing improves wind estimate integrity in regions of possible sea ice Eliminates so-called Ice Winds ICR processing applied to Arctic and Antarctic regions Augmented L2B product (only near-ice regions modified) UHR near-ice product (new) Plan to apply to ASCAT and OSCAT in near future Journal paper currently in review: W.J. Hullinger and D.G. Long, Mitigation of Sea Ice Contamination in QuikSCAT Wind Retrieval, IEEE Transactions on Geoscience and Remote Sensing, in review, 212

27 Backup Slides

28 Overview Brief review of QuikSCAT wind retrieval Ice Contribution Ratio (ICR) Measurement Model ICR Calculation ICR Threshold Determination Case Studies Performance Analysis Conclusion

29 Metrics using Various Priors Stand-off Distance (m) Relative Error (%)

30 Effect of Ocean Prior ICR Processed Winds (m/s)