1. Martin Rutherford Technical Director Oceanography and Meteorology Royal Australian Navy 02 Jun 2009 Towards a Bias Corrected L4 Median

2. Scope n Motivation – Spoilt with choice! n Creating a simple median using ArcGIS n Analysing the Median n Bias corrections n What next!

3. Motivation n 1984 – 1992 -CSIRO -Weekly ‘best image’ based on cloud content -AVHRR only -Hard copy by priority post -Customers were happy (previously had nothing) n 1992 – 2003 -BoM regional SST image -Updated after each AVHRR pass -Weighted blend based on age, but no OI -Sourced using McIDAS, delivered via the web -Customers were happy (timely, fewer gaps)

4. Motivation n 2003 - 2007 -NAVO MCSST K10 -Global daily 10 km analysis -OI, no cloud holes -Processed as web map service and posted on internet -Customers were happy (more timely, higher resolution) n 2007 -Added OSTIA as web map service -Added K10 – OSTIA differences ~ confidence -Customers NOT happy (which one do they use?)

5. Lessons Learned n Customers want: -Web-based delivery -Frequent updates -High resolution -A measure of confidence/uncertainty -Animation n Customers do NOT want: -Choice

6. A (Not so) Simple Median n Use Python and GIS tools to: -Download all L4 products -Interpolate to common grid -0.1, 0.25, 0.5?? -Bilinear, cubic ?? -Calculate median and standard deviation -Display as Web Map Service -Deliver as GIS compatible layer

7. Best Spatial Resolution ?

12. Best Interpolation Method ?

13. Decision Point ! n Re-sample to 0.1 Degree n Use Cubic Interpolation

14. The Daily Product

15. Contributing to the Median

25. Calculating the Bias n Find daily (Interpolated - Median) n Bias = Median of daily (Interpolated – Median)

26. Bias: ABOM GAMSSA

27. Bias: EUR ODYSSEA

28. Bias: NAVO K10

29. Bias: NCDC AVHRR

30. Bias: NCDC AVHRR + AMSR

31. Bias: REMSS MW+IR

32. Bias: UKMO OSTIA

33. Applying the Bias n Apply bias correction to daily Interpolated n Find new daily median

34. Testing for Impact n Compare Standard Deviation n Compare contributions to median

35. Testing for Impact – Std. Dev.

36. ABOM GAMSSA ‘Bias’

37. GAMSSA Uncorrected Contribution

38. GAMSSA Corrected Contribution

39. EUR ODYSSEA ‘Bias’

40. ODYSSEA Uncorrected Contribution

41. ODYSSEA Corrected Contribution

42. NAVO K10 ‘Bias’

43. K10 Uncorrected Contribution

44. K10 Corrected Contribution

45. NCDC AVHRR ‘Bias’

46. AVHRR Uncorrected Contribution

47. AVHRR Corrected Contribution

48. NCDC AVHRR + AMSR ‘Bias’

49. AVHRR + AMSR Uncorrected Contribution

50. AVHRR + AMSR Corrected Contribution

51. UKMO OSTIA ‘Bias’

52. OSTIA Contribution

53. OSTIA Corrected Contribution

54. REMSS MW + IR ‘Bias’

55. REMSS Uncorrected Contribution

56. REMSS Corrected Contribution

57. Percentage of Observations in Median

58. What Next? n Seek GHRSST Science Team Advice!! n Re-do all calculations! n Publish uncorrected median and Std. Dev. n Compare with GMPE

59. Bigger Questions! n Should users be correcting GHRSST products? n Has this technique any merit?? n How do I provide the ‘best’ information for my users from existing GHRSST products: -Global -10km or better -0.5 K or better -Six hourly -Near real time -Including diurnal variability

60. Comments?