SSMI Sea Ice Concentrations in the Marginal Ice

  • Slides: 26
Download presentation
SSM/I Sea Ice Concentrations in the Marginal Ice Zone A Comparison of Four Algorithms

SSM/I Sea Ice Concentrations in the Marginal Ice Zone A Comparison of Four Algorithms with AVHRR Imagery submitted to IEEE Trans. Geosci. and Rem. Sensing 4 June 2004 Walt Meier NSIDC/CIRES Research Scientist

Motivation • Most previous algorithm comparisons have involved isolated case studies (a few days)

Motivation • Most previous algorithm comparisons have involved isolated case studies (a few days) • Comparisons have involved one or two algorithms • Comparisons often encompass primarily regions of compact ice where errors are expected to be smallest

This Study • Large scale independent comparison of SSM/I ice concentration algorithms – –

This Study • Large scale independent comparison of SSM/I ice concentration algorithms – – Four algorithms Several days Winter and summer Three regions • Focus on marginal/seasonal ice zone – – – Region of operational interest High small-scale variability both in space and time Region of large seasonal and interannual variability Algorithms have most difficulty in such regions Models of air-sea exchange most sensitive in such areas

Map of Study Regions Barents E. Greenland Baffin

Map of Study Regions Barents E. Greenland Baffin

AVHRR Imagery • Images with considerable cloud free areas collected over one year period

AVHRR Imagery • Images with considerable cloud free areas collected over one year period – June 2001 – August 2001 – November 2001 – March 2002 • Images collected from Eastern Arctic – Barents Sea – Baffin Bay – East Greenland sea • 2. 5 km resolution on NSIDC polar stereographic grid • >750 total scenes collected; 48 used in study

AVHRR Concentration: Summer • Mixing Method • June 2001 – August 2001 • Assume

AVHRR Concentration: Summer • Mixing Method • June 2001 – August 2001 • Assume Channel 2 (0. 72 -1. 10 m) albedo reflects amount of ice present in a pixel • Tiepoints defined for 100% ice and 100% water • Ice concentration derived from linear interpolation between tiepoints • Tiepoints determined locally in each image – Account for changes in sun and satellite angle and local ice changes • Similar methodology used in several past comparisons, e. g. : Comiso and Steffen, 2001, Zibordi et al. , 1995, Emery et al. , 1991, Steffen and Schweiger, 1990

AVHRR Concentration: Winter • Threshold Method • December 2001 – March 2002 • Assume

AVHRR Concentration: Winter • Threshold Method • December 2001 – March 2002 • Assume surface temperature is below freezing, thus ice is continually forming • Channel 4 (10. 3 -11. 3 m) brightness temperature indicates if ice is present in pixel or not • Ice/water threshold temperature (~271 K) defined – If Tb > threshold, Concentration = 100% – If Tb < threshold, Concentration =0% • Threshold chosen locally within each individual AVHRR image • Similar methodology used in several past comparisons, e. g. : Comiso and Steffen, 2001, Zibordi et al. , 1995, Emery et al. , 1991, Steffen and Schweiger, 1990

SSM/I Concentration Fields • 25 km fields on NSIDC polar stereographic grid – Algorithms

SSM/I Concentration Fields • 25 km fields on NSIDC polar stereographic grid – Algorithms run on 24 -hour composite brightness temperature fields acquired from NOAA at the National Ice Center • Subsampled to same region as AVHRR images • Rebinned (no interpretation) to same 2. 5 km resolution as AVHRR for pixel-to-pixel comparison • Weather filters used to eliminate false ice signals over open water (same filters used for all algorithms)

SSM/I Algorithms • Bootstrap (BT): 19 V, 19 H, 37 V – e. g.

SSM/I Algorithms • Bootstrap (BT): 19 V, 19 H, 37 V – e. g. , Comiso et al. , 1997 • Cal/Val (CV): 19 V, 37 V (37 V, H near ice edge) – e. g. , Ramseier et al. , 1988 • NASA Team (NT): 19 V, 19 H, 37 V – e. g. , Cavalieri et al. , 1984 • NASA Team 2 (N 2): 19 V, 19 H, 37 V, 85 H – Markus and Cavalieri, 2000

NASA Team 2 • Newest algorithm • Uses 85 GHz channels in addition to

NASA Team 2 • Newest algorithm • Uses 85 GHz channels in addition to standard 19 and 37 GHz channels – 85 GHz susceptible to atmosphere – N 2 uses inverse radiative transfer model to find ‘bestfit’ of 11 standard atmospheres – Atmosphere subtracted out from Tb signal – 85 GHz more sensitive to surface inhomogeneities potentially more accurate if no atmospheric problems • Standard algorithm for AMSR-E in the Arctic

SSM/I – AVHRR Difference Total 13897 pixels BT CV N 2 NT Mean -5.

SSM/I – AVHRR Difference Total 13897 pixels BT CV N 2 NT Mean -5. 3 1. 8 -1. 2 -9. 0 St. Dev. 12. 9 13. 7 14. 6 BT CV N 2 NT Mean -6. 1 -4. 3 -2. 6 -10. 5 St. Dev. 14. 6 16. 9 15. 7 15. 9 BT CV N 2 NT Mean -5. 0 0. 7 -0. 6 -8. 4 St. Dev. 12. 2 12. 3 12. 7 13. 9 Summer 4125 pixels Winter 9772 pixels Values in yellow are the lowest difference or are within 95% confidence level of lowest difference.

% Difference SSM/I-AVHRR Mean Differences Summer Differences for each case (numbered on x-axis) for

% Difference SSM/I-AVHRR Mean Differences Summer Differences for each case (numbered on x-axis) for each season. Error bars indicate 95% confidence levels. % Difference Winter

% Difference SSM/I-AVHRR St. Dev. Differences Summer Differences for each case (numbered on x-axis)

% Difference SSM/I-AVHRR St. Dev. Differences Summer Differences for each case (numbered on x-axis) for each season. Error bars indicate 95% confidence levels. % Difference Winter

Case Study Barents Sea 17 June 2001

Case Study Barents Sea 17 June 2001

% 0 20 40 60 80 100

% 0 20 40 60 80 100

BT 72% CV 81% N 2 74% NT 68% AV 79% % 0 20

BT 72% CV 81% N 2 74% NT 68% AV 79% % 0 20 40 60 80 100

% 0 20 40 60 80 100

% 0 20 40 60 80 100

AV BT CV 99. 6% 99. 2% 100. 0% % 0 20 40 N

AV BT CV 99. 6% 99. 2% 100. 0% % 0 20 40 N 2 60 97. 1% 80 100 NT 89. 8%

Case Study E. Greenland Sea 27 February 2002

Case Study E. Greenland Sea 27 February 2002

% 0 20 40 60 80 100

% 0 20 40 60 80 100

AV BT CV N 2 NT 96% 86% 93% 94% 83% % 0 20

AV BT CV N 2 NT 96% 86% 93% 94% 83% % 0 20 40 60 80 100

Clouds • Previous comparison limited to clear sky regions • Clouds prevalent – Over

Clouds • Previous comparison limited to clear sky regions • Clouds prevalent – Over 8 months of images in the three regions (~750 total) – <60 had enough clear sky to make comparisons • Algorithms likely do not perform as well under thick clouds, particularly N 2 • To investigate potential effects of clouds, a regional case study was conducted – Meier, W. N. , T. Maksym, and M. L. Van Woert, Evaluation of Arctic operational passive microwave products: A case study in the Barents Sea during October 2001, Proc. 16 th Int’l Symposium on Ice, Dunedin, NZ, 2 -6 Dec 2002, pp. 213 -222. – Barents Sea, October 2001 – USCGC Healy cruise – SSM/I concentrations compared with Radarsat imagery – N 2 did not show any noticeable degradation

1 Oct. BS SSM/I Contour Intervals • 5% • 15% • 50% • 90%

1 Oct. BS SSM/I Contour Intervals • 5% • 15% • 50% • 90% CV N N 2 Underestimates ice edge NT OLS © CSA 2001

11 Oct BS CV SSM/I Contour Intervals • 5% • 15% • 50% •

11 Oct BS CV SSM/I Contour Intervals • 5% • 15% • 50% • 90% Misses ice N 2 NT OLS Captures lower concentration © CSA 2001

Conclusions • Performance of algorithms varies depending on season, ice conditions, etc. – Overall,

Conclusions • Performance of algorithms varies depending on season, ice conditions, etc. – Overall, NASA Team 2 and Bootstrap have lowest differences from AVHRR • N 2 tends to have lowest bias • Bootstrap tends to have lowest difference SD – Cal/Val tends to overestimate concentration due to saturation to 100% concentration, especially in summer – NT is inferior algorithm in most situations • Algorithms yield similar difference SD values, due at least in part to low resolution of sensor no matter what algorithm is used, resolution limits the effectiveness of SSM/I

Acknowledgements • Canadian Space Agency for Radarsat imagery • DMSP and NOAA for OLS

Acknowledgements • Canadian Space Agency for Radarsat imagery • DMSP and NOAA for OLS and SSM/I data • Søren Anderson, Danish Meteorology Institute, for AVHRR data • Midshipman Nathan Bastar for initial analysis