1 / 31

“Regional” adjustments of SAA parameterization

“Regional” adjustments of SAA parameterization. Mark Dowell & Timothy Moore EC-JRC NURC & UNH. Definitions. “Regional”: Optical classification: Classes: Provinces:. Elements for Discussion (1/2). Do we need “regional” SAA parameterization to make further progress ?

darin
Download Presentation

“Regional” adjustments of SAA parameterization

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. “Regional” adjustments of SAA parameterization Mark Dowell & Timothy Moore EC-JRC NURC & UNH

  2. Definitions • “Regional”: • Optical classification: • Classes: • Provinces:

  3. Elements for Discussion (1/2) • Do we need “regional” SAA parameterization to make further progress ? • Do we need regional or optical class based parameterizations ? – how do we make this decision ? • Should effort be made not only for the IOP subcomponent models but also for the bulk IOP relationships (e.g. VSF variability)? • Should such methods be applicable independent of the inversion method (i.e. different inversion methods can be used for different optical water types)? • Will class-based approaches improve “ambiguity” issues associated with SAAs?

  4. Elements for Discussion (2/2) • Are existing global datasets “clean” enough to parameterize sub-component models at the scale of a class? • What would be acceptable as far as increased processing time, for a class based approach to be included in operational processing ? • Do we need additional statistics for assessing class based approaches ? • If we base our classifications on in-situ AOPs how do we incorporate the uncertainties associated with satellite derived nLw when mapping province distributions?

  5. Alternatives for SAA parameterizations(after Platt & Sathyendranath 2001) • Constant • i.e. constant aph* globally • Piecewise • i.e. different aph* for different geographic regional or “hard” optical classes • Continuous • i.e. aph* as function of unknown – aph*-f(CHL) • Piecewise continuous • Through mebership function and aph*-f(CHL)

  6. IOP sub-component model parameterization • aph*(l) • Sgd • Slope of bbp • In a spectral un-mixing algorithm many opportunities at intermediate steps. • Also Kd & Zeu models etc.

  7. Benefits • Flexible framework for updating algorithms based on availability of new “regional” datasets. • Feedback mechanism in prioritizing fieldwork for validation and parameterizations. • Avenue into spatial uncertainty analysis • Also of benefit for parameterizing new products from IOP e.g. POC, DOC

  8. Rationale • There is necessity to describe a considerable amount of variability in Inherent Optical Property (IOP) subcomponent models. • This is particularly true, if inversion algorithms are to be applicable at global scale yet remain quantitatively accurate in both the open ocean and coastal/shelf seas. • This is unlikely to be achieved in the foreseeable future, with a single representation of IOP subcomponents. • BEAM – Case2R • The proposed approach is an algorithm framework more than a specific algorithm.

  9. The approach undertaken adopts fuzzy logic to define and identify, in radiance space, distinct bio-optical provinces that implicitly reduce the variance in the IOP subcomponent models. Fuzzy Hard Fuzzy graded membership Traditional minimum-distance criteria Forest Forest Reflectance Band 2 Reflectance Band 2 Wetland Wetland Water Water Water = 0.05 Wetland = 0.65 Forest = 0.30 Unknown measurement vector Mean class vector 0 10 20 30 0 10 20 30 Reflectance Band 1 Reflectance Band 1

  10. c Satellite Measurements c Rrs() Individual class derived products Calculate membership c c Merged Product In-situ Database Sgd, aph*,……. Rrs() c c 8 classes Cluster analysis Station data sorted by class c c Class based relationships Class Mi, Si

  11. Advantages of fuzzy logic defined provinces • They allow for dynamics both seasonal and inter-annual in the optical properties of a given region. • They address the issue of transitions at the boundaries of provinces (through the fuzzy membership function of each class) thus resulting finally in the seamless reconstruction of a single geophysical product.

  12. 8 objectively identified classes in radiance space

  13. May 2004 SeaWiFS Composite Class 1 Class 2 Class 3 Class 4 Class 5 Class 6 Class 7 Class 8

  14. High resolution provinces for European Seas Med May 2004 Class 1 Class 2 Class 3 Class 4 Class 5 Class 6 Class 7 Class 8

  15. Channel 1-5 Channel 2,3,5 MERIS MODIS/Aqua May 2004 SeaWiFS Class 1 Class 2 Class 3 Class 4 Class 5 Class 6 Class 7 Class 8

  16. Chlorophyll agd(443) Rrs(665)

  17. Inversion Methods • Non-linear optimization (Amoeba) - a la GSM • Spectral Unmixing - a la Lee et. al. QAA • ……Neural Network , PCI Note: there is no need for the same inversion method to be applied to each class

  18. Class-based NLO (e.g. GSM) • Sgd varies based on class • [0.0180,0.0164,0.0139,0.0147,0.0153,0.0128,0.0138,0.0121] • aph*() varies dependent on class • Y (i.e. slope of bbp) using Carder’s relationship One could imagine applying a tuning algorithm (e.g. simulated annealing) to each class to determine optimimal class based model coefficients.

  19. Amoeba - NLO

  20. Class-based version of QAA • Sgd variable based on class • at(443) versus rrs(443)/rrs(555) class based • at(555) versus at(443) class based • aph(443) versus Chl class based aph*(443)

  21. a555 class parameterization N=41 N=59 N=110 N=169 N=71 Class-specific a555-rho parameterization: QAA

  22. RMS Error – NOMAD in situ data Relative Error (%) N=391

  23. Chlorophyll a – Aug. 31, 2006

  24. Chlorophyll a – Aug. 31, 2006

  25. Hard classes Aug. 31, 2006 MERIS MODIS

  26. MERIS fuzzy classes – March 23, 2008

  27. Spectral Unmixing Amoeba - NLO

  28. Conclusions • Fuzzy logic based dynamic provinces provide a powerful tool for describing the optical variability of the world oceans. • Statistically rigorous means of parameterizing bio-optical models. • Allow to produce algorithms capable of describing the strong non-linearity of optical variability across many decades of variability • Various inversion methods have been/ are being tested - not all classes need to adopt the same inversion scheme. • Uncertainty estimate with the approach can benefit from the availability of the membership functions. • Ongoing work to identify bio-optical “end-member” locations for use in identifying cal/val sites, as well as identifying “under-sampled” optical water types.

More Related