1. Optimal Retrieval of Vegetation Canopy Characteristics Using Operational MODIS and MISR Surface Albedo Products LMD, Paris, October 5, 2007 JRC – Ispra Bernard Pinty, T. Lavergne, T. Kaminski, O. Aussedat, N. Gobron and M. Taberner with contributions from R. Giering, M. M. Verstraete, M. Vossbeck and J-L. Widlowski EC-JRC Institute for Environment and Sustainability, Ispra, Italy FastOpt, Hamburg, Germany

2. Complex land-surface RT effects on short term climate: the snow case with ECMWF/NCEP Ref: Viterbo and Betts, 1999, JGR Ref: http://eobglossary.gsfc.nasa.gov/ “…weather forecasts significantly underestimated air temperatures over boreal, sometimes by as much as 10-15 C…”

3. Ref: Viterbo and Betts, 1999, JGR Ref: http://eobglossary.gsfc.nasa.gov/ “…—the BOREAS team found that the models were overestimating albedo (the amount of light reflected by the surface). …” Complex land-surface RT effects on short term climate: the snow case with ECMWF/NCEP

4. How does radiation redistribute energy between the atmosphere and the biosphere? Absorbed Fluxes in Vegetation Absorbed Fluxes in Soil Scattered Fluxes by the surface

5. What do we measure at global scale that we should model as well? Albedo of the surface in the VIS and NIR (MODIS and MISR) Absorbed Flux by green Vegetation in the VIS (FAPAR)

6. Correct partitioning between the flux that is absorbed : 1- in the vegetation layer 2- in the background Pinty etal., (2006): Journal of Geophysical Research, doi:10.1029/2005JD005952 How do we model the absorbed fluxes in vegetation and soil ? Assessment of the fraction of solar radiant flux that is scattered (albedo) by, transmitted through and absorbed in the vegetation layer VIS+NIRVIS

7. Update/improve the current Land Surface schemes describing the radiation transfer processes in vegetation canopies see 2-stream model by Pinty et al. JGR (2006). What are the needs?

8. Requirements from a 2-stream model 3 (effective) state variables: 1. Optical depth: LAI 2. single scattering albedo : Leaf reflectance+ Leaf transmittance 3. asymmetry of the phase function Leaf reflectance/transmittance 2 boundary conditions: 1. Top: Direct and Diffuse atmospheric fluxes (known) 2. Bottom : Flux from background Albedo (unknown) Pinty etal., (2006): Journal of Geophysical Research, doi:10.1029/2005JD005952

9. The concept of effective LAI Effects induced by internal variability of LAI

10. The concept of effective LAI Structure factor

11. Update/improve the current Land Surface schemes describing the radiation transfer processes in vegetation canopies see 2-stream model by Pinty et al. JGR (2006). Prepare for the ingestion/assimilation of RS flux products into Land Surface schemes Retrieve 2-stream model parameters from RS flux products What are the needs?

12. The same RT models must be used in inverse mode and in forward mode to ensure consistency between RS products and large scale models’s simulations

13. Use the same model in INVERSE and forward SIMULATION mode INVERSE MODE with 3D model E. Bourderes SIMULATION MODE with 1D model

14. Retrievals of model Parameters for Land surface schemes Towards an integrated system for the optimal use of remote sensing flux products The inverse problem can be formulated in order to find solutions optimizing all the available information i.e., inferring statistically the state of the system

15. RS flux products from various sources + uncertainties Prior knowledge on model parameters + uncertainties Ancillary information, e.g. occurrence of snow and water Radiation transfer model Time-independent operating mode Posterior model parameters + uncertainties Diagnostic fluxes + uncertainties JRC-Two-stream Inversion Package: JRC-TIP Re-analysis generating consistent dataset

16. INPUTS : prior knowledge Updated/benchmarked 2-stream model from Pinty et al. JGR (2006) noted A priori knowldege/guess on model parameters noted uncertainty on the RS products is specified in the measurement set covariance matrix uncertainty associated the model parameter is specified via a covariance matrix RS Flux products, e.g., Albedo Vis/NIR and/or FAPAR noted

17. The core of the JRC-TIP Model parameters 2-stream model measurements Parameter knowledge Uncertainty measurements Uncertainty parameters Computer optimized Adjoint and Hessian model of cost function from automatic differentiation technique Assume Gaussian theory Posterior uncertainties on retrieved parameters are estimated from the curvature of Pinty etal., (2007): Journal of Geophysical Research, in press

18. OUTPUTS: posterior knowledge Assessement of all fluxes predicted by the 2-stream model and their associated uncertainty: PDFs of all 2-stream model parameters: a posteriori uncertainty covariance matrix Pinty etal., (2007): Journal of Geophysical Research, in press

19. Application results against model-based standard scenarios

20. a priori knowledge on model parameters Pinty etal., (2007): Journal of Geophysical Research, in press

21. Application results against model- based standard scenarios Application with measurements set d including various combinations of visible and near-infrared broadband radiant fluxes, i. e., Albedo (R), Transmission (T) and Absorption (A) PDFs for 7 model parameters to be estimated: LAI, Leaf and background properties in the two spectral domains

22. Single scattering albedo Asymmetry phase function Background albedo NEAR-INFRARED Single scattering albedo Asymmetry phase function Background albedo VISIBLE Lavergne etal., (2007): EUR 22467 EN measurement sets d a posteriori 1D Gaussian statistics @ 1σ

23. Single scattering albedo Asymmetry phase function Background albedo NEAR-INFRARED Single scattering albedo Asymmetry phase function Background albedo VISIBLE a posteriori 1D Gaussian statistics @ 1σ Lavergne etal., (2007): EUR 22467 EN measurement sets d

24. a posteriori 2D Gaussian statistics @ 1.5σ Lavergne etal., (2007): EUR 22467 EN

25. Results for scenarios built from Monte Carlo sampling of the a priori knowledge on model parameters together Pinty etal., (2007): Journal of Geophysical Research, in press

26. Retrieval of LAI (model-based cases)

27. Retrievals in Visible domain Single scattering albedo Background albedo RRA LAI<4 LAI>4

28. a posteriori covariance matrices Using the absorbed and scattered flux only Leaf Area Index Single scattering albedo Asymmetry phase function Background albedo VISIBLE Near-InfraRed

29. Application results over selected EOS validation sites Application with measurements set d limited to the visible and near-infrared broadband surface albedos, i.e., 2 measurements and 7 model parameters to be retrieved

30. prior knowledge on model parameters Pinty etal., (2007): Journal of Geophysical Research, in press

31. a priori covariance matrix

32. Dahra (Senegal) 15º 22’ N 15º 26’ W Semi-arid grass savannah Short and homogeneous over 1-2 km AGRO (US) 40º 00’ N 88º 17’ W Broadleaf crops and corn/soybean Mixed vegetation with different land cover type Konza (US) 39º 04’ N 96º 33’ W Grassland, shrubland and cropland Mixed vegetation with different land cover type Mongu (Zambia) 15º 26’ S 23º 15’ E Mixed shrubland and woodland Intermediate height but low density vegetation Example of application results Site identification and characteristics Pinty etal., (2007): Journal of Geophysical Research, in press

33. Application over AGRO: Measurements Pinty etal., (2007): Journal of Geophysical Research, in press

34. Application over AGRO: model parameters 2-stream model parameters Time-independent inversion Pinty etal., (2007): Journal of Geophysical Research, in press

35. Application over AGRO: Radiant fluxes 2-stream model radiant fluxes Time-independent inversion

36. in case snow occurs prior knowledge on model parameters Pinty etal., (2007): Journal of Geophysical Research, in press a priori ‘green’ leaves

37. Application over BOREAS NSA-OBS

38. Application over BOREAS: Measurements MODIS Snow indicator (MOD10A2) Specified uncertainty on BHRs is 5% relative

39. Application over BOREAS: Measurements MODIS Snow indicator (MOD10A2) a priori ‘green’ leaves Specified uncertainty on BHRs is 5% relative

40. Application over NSAOBS: model parameters a priori ‘green’ leaves

41. Application over NSAOBS: model parameters MODIS /Terra FAPAR MISR /Terra FAPAR a priori ‘green’ leaves

42. Application over NSAOBS: model parameters

43. MODIS /Terra FAPAR MISR /Terra FAPAR

44. Application over NSAOBS: model parameters a priori ‘green’ leaves

45. Application over NSAOBS: model parameters

46. Application over NSAOBS: radiant fluxes

47. a priori ‘green’ leaves

48. Application over NSAOBS: radiant fluxes JRC-FAPAR SeaWiFS MODIS FAPAR MISR FAPAR a priori ‘green’ leaves MERIS FAPAR

49. Application over NSAOBS: radiant fluxes a priori ‘green’ leaves JRC-FAPAR SeaWiFS MODIS FAPAR MISR FAPAR MERIS FAPAR our inversion based on TERRA albedos

50. Application over NSAOBS: radiant fluxes a priori ‘green’ leaves MERIS FAPAR current inversion based on TERRA albedos

51. Application over NSAOBS: radiant fluxes Assimilation of the MERIS FAPAR product Original MERIS FAPAR products Re-analysed MERIS FAPAR with MODIS/MISR surface albedos

52. Agriculture Semi-desert

53. Deciduous broadleaf beech forest Shrubland-woodland

54. Evergreen needleleaf pine forest Deciduous needleleaf larch forest

55. The GREEN solution The polychrome solution

56. Concluding remarks 1. Computer efficient inversion package has been designed and tested : estimate of uncertainty on all retrievals 2. This integrated package can be used for various purposes : retrieval of parameters from RS products, validation of RS products, assimilation of RS products into Land surface schemes. 3. Capability to generate global surface model parameters ensuring full consistency with measured (uncorrelated) fluxes from various sources: spectral albedos from MODIS- MISR (and any other sources). 4. Estimating radiant fluxes and surface parameters in the presence of snow.

58. a posteriori covariance matrices Using the absorbed flux only

59. a posteriori covariance matrices Using the absorbed and scattered flux only

60. a posteriori covariance matrices Using all radiant fluxes

61. Decompose the complex problem into simpler problems to solve

64. Features of 3-D versus 1-D systems 3-D heterogeneous vegetation systems : Are more transparent (solar domain) Absorb much less (PAR domain) Reflect much less (NIR domain) than predicted by 1-D theory using true state variables

65. Pinty etal., (2006): Journal of Geophysical Research, doi:10.1029/2005JD005952

68. Comparing/constraining or assimilating the radiation fluxes retrieved from RS against those generated by GCMs is not valid when using the true state variables in the GCMs simulations Pinty etal., (2004): Journal of Geophysical Research, doi:10.1029/2004JD005214

69. What did we improve? Domain of validity and accuracy of current implementations in GCMS: 1.can be extended to regimes where multiple scattering dominate e.g., snow-covered background. 2.Leaf reflectance can now be different from Leaf transmittance –needle shoots in coniferous canopies - 3.Corrections of current schemes in single scattering mode 4.Possibility to account for 3-D vegetation effects Possibility to ingest quantitative RS products: 1. Fluxes: Albedo, absorption 2. State variable: ‘effective’ LAI. 3.Adding one more layer to the atmospheric RT scheme