1. Jin Wu 1, Neill Prohaska 1, Kenia T. Wiedemann 1,2, Loren P. Albert 1, Alfredo R. Huete 3*, and Scott R. Saleska 1* Observing canopy demography of a tropical moist forest with spectral unmixing of hyperspectral images: a model of HyspIRI science application Email Contact: jinwu@email.arizona.edu or saleska@email.arizona.edujinwu@email.arizona.edusaleska@email.arizona.edu (1: Department of Ecology and Evolutionary Biology, University of Arizona; 2: School of Engineering and Applied Sciences, Harvard University; 3: Plant Functional Biology & Climate Change Cluster, University of Technology Sydney) *: Corresponding Authors

2. Outline of the Talk 1: Motivation—canopy demography matters ecological processes of forest systems 2: Methodology—remote sensing perspective of demography 4: Canopy Demography Modeling—case study from a tower mounted hyperspectral imaging system 5: Inference to HyspIRI Study 3: Study Site and Materials 6: Summary and Future Work

3. Motivation O2CO2, H2O Both environments and phenology control photosynthesis seasonality; Phenology includes the seasonality of both Leaf Area Index (LAI) and canopy average leaf physiology Soil Moisture Nutrients Environments Leaf Area Index (LAI) Canopy Average Leaf Physiology* Phenology *Canopy leaf average physiology is an Euler Product of canopy demographic composition and physiology of each demographic component

4. Motivation (cont..) O2CO2, H2O Soil Moisture Nutrients Environments Leaf Area Index (LAI) Canopy average leaf physiology* Phenology *Canopy leaf average physiology is an Euler Product of canopy demographic composition and the physiology of each demographic component In a word, LAI seasonality and canopy demography seasonality are two important perspectives of plant phenology Both environments and phenology control photosynthesis seasonality; Phenology includes the seasonality of both Leaf Area Index (LAI) and canopy average leaf physiology

5. Motivation (cont..) Deciduous Forests & Evergreen Tropical Forests Deciduous Broadleaf Forest, Takayama, Japan. (Muraoka et al. 2010) 100 150200250 300 Day of Year 100 150200250 300 Day of Year Vcmax 25 (umol/m2/s) 0 20 40 60 80 LAI (m2/m2) 0 2 4 6 (a) (b) Canopy Level Leaf Level One Age Cohort

6. Motivation (cont..) Deciduous Forests & Evergreen Tropical Forests Deciduous Broadleaf Forest, Takayama, Japan. (Muraoka et al. 2010) 100 150200250 300 Day of Year 100 150200250 300 Day of Year Vcmax 25 (umol/m2/s) 0 20 40 60 80 LAI (m2/m2) 0 2 4 6 (a) (b) Canopy Level Albert & Wu. 1-m Branch Sample from Uxi Lista, 08/24/2012. Albert et al. In Preparation Leaf Age Tropical Moist Evergreen Forest, Tapajos, Para, Brazil (c) (d) Leaf Level Multiple Age Cohort One Age Cohort

7. Motivation (cont..) Deciduous Forests & Evergreen Tropical Forests Albert & Wu. 1-m Branch Sample from Uxi Lista, 08/24/2012. Albert et al. In Preparation Leaf Age Tropical Moist Evergreen Forest, Tapajos, Para, Brazil (c) (d) Leaf Level Multiple Age Cohort Tropical Moist Evergreen Forest, Tapajos, Para, Brazil LAI Photosyntehetic Capacity (umol/m2/s) Month of Year Observation Month of Year Canopy Level Levine, Restrepo-Coupe, and Saleska in prep

8. Motivation (cont..) Deciduous Forests & Evergreen Tropical Forests Albert & Wu. 1-m Branch Sample from Uxi Lista, 08/24/2012. Albert et al. In Preparation Leaf Age Tropical Moist Evergreen Forest, Tapajos, Para, Brazil (c) (d) Leaf Level Multiple Age Cohort Tropical Moist Evergreen Forest, Tapajos, Para, Brazil LAI Photosyntehetic Capacity (umol/m2/s) Month of Year Observation Month of Year Canopy Level Levine, Restrepo-Coupe, and Saleska in prep LAI seasonality and demography seasonality are both important to tropical photosynthesis processes

9. Model perspective of evergreen tropical photosynthesis seasonality ED: Ecosystem Demographic Model Levine, Restrepo-Coupe, and Saleska in prep IBIS: the Integrated Biosphere Simulator CLM: the NCAR Community Land Model JULES: the joint UK Land Environment Simulator Motivation (cont..) Dry Season GPP (gC m -2 d -1 ) Tapajos National Forest, KM67 site, Para, Brazil Observation Month of Year

10. Model perspective of evergreen tropical photosynthesis seasonality Motivation (cont..) Dry Season GPP (gC m -2 d -1 ) Observation Month of Year ED JULES CLMIBIS Soil moisture “Stress” ( 1=no stress) LAI Photosyntehetic Capacity (umol/m2/s) Month of Year Observation Levine, Restrepo-Coupe, and Saleska in prep Models driven Observation driven

11. Model perspective of evergreen tropical photosynthesis seasonality Motivation (cont..) Dry Season GPP (gC m -2 d -1 ) Observation Month of Year ED JULES CLMIBIS Soil moisture “Stress” ( 1=no stress) LAI Photosyntehetic Capacity (umol/m2/s) Month of Year Observation ED JULES CLMIBIS LAI (m2/m2) Summary of Motivation (1)Both LAI seasonality and demography seasonality are two important aspects of plant phenology (2) LAI seasonality and demography seasonality are both important for tropical forest photosynthesis processes; (3) There is still of great need to incorporate both LAI seasonality and demography seasonality to better model ecosystem level photosynthesis of tropical evergreen forest systems. Levine, Restrepo-Coupe, and Saleska in prep

12. Methodology: Leaf demography Field methods (accurate, time/cost consuming, spatial limit) (1) leaf level—tagging leaves and revisiting; (2) canopy level—regular harvesting branch sample and conducting demographic survey Remote sensing methods Seasonal RGB Images of a deciduous forest (Bartlett Forest, USA) in 2008 Richardson. 2010. Dublin Land Product Validation Subgroup. How about evergreen tropical forests?

13. Study Site: TNF-KM67, Para, Brazil Study Site: Tapajos National Primary Forest, KM67 site (TNF-KM67), Para, Brazil 57.8 m eddy sample 64 m cameras Eddy Covariance Hyperspectral Vegetation Imaging System (HVIS) Data available from 01/2002 to 01/2006, and 08/2008-12/2012 Equipped by Surface Optics Corporation (SOC) 710 camera Tropical evergreen moist forest with ~5 month dry season

14. Materials: Hyperspectral Vegetation Imaging System B. RGB Image A. SOC Camera C. SOC Composited Image Key parameters: * Spectral resolution: 4nm, 128 bands from 385nm to 1050 nm * Field of view: 36×27 degree * Temporal resolution: image/10 minutes, from 6 am to 6 pm * Data available: Jul-Dec, 2012 * Image dimension: 520*696 pixels Other parameters: * The camera system was programed to adjust their view angles from nadir view to very oblique view (75 o ) every 10 minutes, and repeated every hour

15. * 6 focal trees+14 understory trees were selected; * samples were collected in the early, middle, and late of dry season (Mid-Aug to Early-Dec) Materials: 2012 Fall Phenology Campaign

16. Materials: Leaf Demography Survey Bud scarsYoung LeavesOld Leaves Collected one meter branches multiple times throughout the dry season. Branches from sun and shade microenvironments within the trees. Counted young, mature and old leaves on sampled branches.

17. Canopy Demography Modeling Step1 : Acquire canopy hyperspectral reflectance from SOC camera images DOY 315 265 240 221 Tree #9, fully illuminated leaves, average reflectance for 10*10 pixels; the images selected were under sunny sky condition; all images selected for this analysis between 11 am-12pm local time

18. Canopy Demography Modeling Step2 : Spectral Unmixing Method Vegetation Percentage: f V Soil Percentage: f S One Image Pixel R composited = f V *R V + f S *R S Composited Reflectance Pure Vegetation Reflectance Pure Soil Reflectance Previous Studies (e.g. Asner and Lobell, 2000, RSE)

19. Canopy Demography Modeling Step2 : Spectral Unmixing Method Mature Leaves Percentage: f M Old Leaves Percentage: f O One Image Pixel R camera = f M *R M + f O *R O Camera observed Reflectance Pure Mature Reflectance Pure Old Reflectance This Study

20. 221 Old Recently Mature (a) (b) (a) Time series of 1-m branch composition; (b) Time series of SOC spectra 240 265 315 Canopy Demography Modeling Step3 : Identify pure old leaf reflectance and mature leaf reflectance

21. 221 Old Recently Mature (a) (b) Doy 221, pure old leaf 240 265 315 Canopy Demography Modeling Step3 : Identify pure old leaf reflectance and mature leaf reflectance (a) Time series of 1-m branch composition; (b) Time series of SOC spectra

22. 221 Old Recently Mature (a) (b) Doy 221, pure old leaf 240 265 315 Canopy Demography Modeling Step3 : Identify pure old leaf reflectance and mature leaf reflectance (a) Time series of 1-m branch composition; (b) Time series of SOC spectra Doy 315, pure mature leaf

23. 221 Old Recently Mature (a) (b) 240 265 315 Canopy Demography Modeling Step3 : Identify pure old leaf reflectance and mature leaf reflectance Mature Leaves Percentage: f M Old Leaves Percentage: f O One Image Pixel R camera = f M *R M + f O *R O Camera observed Reflectance Pure Mature Reflectance Pure Old Reflectance *All the channels from 450 nm-900 nm were used for spectral unmixing * The spectra unmixing result is dominated by the variation in NIR reflectance

24. Spectral Prediction Observation Canopy Demography Modeling Step5 : prediction of the seasonality of canopy demography R2=0.8656 P-value=0.0102 Mature Leaves Percentage: f M Old Leaves Percentage: f O One Image Pixel R camera = f M *R M + f O *R O Camera Observed Reflectance Pure Mature Reflectance Pure Old Reflectance Model Inversion

25. Canopy Demography Modeling Step5 : prediction of the seasonality of canopy demography Species NameFamily Tachigali eriopetala Leguminosae -Caesalp Erisma uncinatum Warm. Vochysiaceae Manilkara huberi Sapotaceae Ocotea sp.Lauraceae Four dominant tree species in TNF-KM67 forest site R2=0.7168 P-value<0.001 RMSE=15.1%

26. Inference to HyspIRI Study Earth Observing-1 Hyperion Key Parameters: (1) Launch date: 11/21/2000 (2) Spectral resolution: 224 bands from 400 nm to 2500 nm (3) Temporal resolution: 16-day frequencies (4) Spatial resolution: 30 m pixels, and 8-10 km wide imagery HyspIRI Mission Study

27. Inference to HyspIRI Study 188 221 269 315 Satellite EO-1 Hyperion (Landscape) EO-1 Hyperion in TNF-KM67 site in 2001, 2002 (1)25*25 pixels (~1 km*1km) (2)The data were radiometrically- calibrated, atmospherically corrected, and converted from radiance to apparent surface reflectance (3) Solar Zenith Angle: 39 o in early dry season (July) to 34 o in late dry season (December)

28. Inference to HyspIRI Study 315 265 240 221 188 221 269 315 Camera ( One Tree Crown) Hyperspectra comparison between near- surface remote sensing and satellite 315 265 240 221 Camera (Four Tree Crowns) Satellite EO-1 Hyperion (Landscape)

29. Inference to HyspIRI Study 188 221 269 315 Hyperspectra comparison between near- surface remote sensing and satellite Satellite EO-1 Hyperion (Landscape) *View Angle: ~70 degree (Camera); ~nadir (EO-1 Hyperion) *Spectra Contribution: (1) Limited number of tree crowns, and illuminated leaves (Camera); (2) More species diversity, sunny and shade leaves, branches, soils (EO-1 Hyperion) 315 265 240 221 Camera (Four Tree Crowns)

30. Inference to HyspIRI Study 188 221 269 315 Hyperspectra comparison between near- surface remote sensing and satellite Satellite EO-1 Hyperion (Landscape) *View Angle: ~70 degree (Camera); ~nadir (EO-1 Hyperion) *Spectra Contribution: (1) Limited number of tree crowns, and illuminated leaves (Camera); (2) More species diversity, sunny and shade leaves, branches, soils (EO-1 Hyperion) 315 265 240 221 Camera (Four Tree Crowns) mature leaf old leaf Landscape old leaf ? Landscape mature leaf

31. Inference to HyspIRI Study Applying spectra unmixing approach to landscape, or even Amazonian basin More near-surface remote sensing studies (tower based or airborne based) More field campaigns months 10 0 -10 -20 -50-60-70-80 0 2 4 6 8 10 12 Manaus K34 site Tapajos K67 site (University of Arizona) (INPA and University of Arizona) French Guiana (Oxford University) PASI, Peru Demography Campaign Hyperspectral Camera Hyperspectral Camera (Planned) More species in one site

32. Summary 1. Canopy demography seasonality is an equally important component of plant phenology as LAI seasonality 2. Canopy demography seasonality might be extremely important for understanding evergreen tropical forest photosynthesis processes 3. Spectral unmxing approach was attempted for the first time here for extracting canopy demographic composition based on canopy spectral reflectance acquired by a tower mounted imaging system. This may be the first time a leaf demography unmixing concept has been applied to hyperspectral data.

33. Summary 4. Canopy spectral predicted demography seasonality showed a good agreement with ground surveyed demography (R 2 =0.87, P-Value=0.01), suggesting a feasible way for monitoring canopy demography by using advanced space techniques in future. However, the robustness/applicability of this study still demands more quantitative analysis and similar studies. 5. Time-series of canopy spectra (via tower mounted camera system) showed similar trend as that from Hyperion. This suggests the former can be a model system for the study of Hyperion and HyspIRI, while the later can expand the spatial scale of the former measurements.

34. Future Work 1. Improve Methodology Mature Leaves Percentage: f M Old Leaves Percentage: f O One Image Pixel R camera = f M *R M + f O *R O Camera observed Reflectance Pure Mature Reflectance Pure Old Reflectance (b) 221 240 265 315 (1) Give more weight on the spectral reflectance at visible light region (2) Develop a 2or 3 band “canopy demography index”

35. Future Work 2. Hypothesis Test 221 240 265 315 (a) Recently Mature Old (b) (H1) (H2) The change of spectra in dry season is driven by the change of (H1) canopy demography and/or (H2) leaf amount, and/or other process (e.g. epiphyll)

36. Future Work 2. Hypothesis Test 221 240 265 315 (a) Recently Mature Old (b) (H1) (H2) It will be great to incorporate some processes based model (e.g. Prospect+SAIL) to quantitatively assess these two hypotheses

37. Acknowledgments Sensor Calibration and Discussion William van Leeuwen (UofA) Nikolaus Anderson (UofA) Stuart Biggar (UofA) David J. Moore (UofA) Joost van Haren (UofA) Field Work & Help Cecilia Chavana-Bryant (Oxford U) Mick Eltringham (UK) Kleber Silva Campos (UFOPA) Marielle Smith (UofA) Ty Taylor (UofA) Scott Stark (MSU) Brad Christofferson (Uof Edinburgh) Rodrigo da Silva (UFOPA) Raimundo Cosme Oliveiera (UFOPA) Steve Bissell (US) Funding Amazon-PIRE NASA Terrestrial Ecology Program Photo Credit: Neill Prohaska

38. Tree #/Name Nome CientíficoFamília Tr# 9 Erisma uncinatum Warm. Vochysiaceae Tr# 11 Ocotea sp.Lauraceae Tr# 118 Tachigali eriopetala (Ducke) L.G.Silva & H.C.Lima Leguminosae -Caesalp. Tr# 500 Manilkara huberi (Ducke) A.Chev. Sapotaceae Tr# 504 Coussarea paniculata (Vahl) Standl. Rubiaceae Sap Flow Tree Chamaecrista scleroxylon (Ducke) H.S.Irwin & Barneby Leguminosae -Caesalp. * 6 focal trees+14 understory trees were selected; * samples were collected in the early, middle, and late of dry season (Mid-Aug to Early-Dec) Materials: 2012 Fall Phenology Campaign

39. Leaf Spectra & Canopy Spectra 221 240 265 315 Young Young/Mature Mature Old Old & Epiphyll Leaf level hyperspectra (via a handheld ASD) Canopy level hyperspectra (via tower mounted camera) become older

40. R2=0.8656 P-value=0.0102 Spectral Prediction Observation Canopy Demography Modeling Model Prediction

41. Methodology: Reflectance Calculation #Tr9 #Tr11 #Tr118 #Tr13 #Tr14 Premise from equation (1) to equation (2) is DN linearly scales with radiation for each wavelength ( was tested in the lab!!) : leaf reflectance at wavelength : reference reflectance at wavelength : leaf surface radiation at wavelength : camera based digital number at wavelength Calibrated by an known surface reflectance Spectralon Under clear sky condition Diurnal pattern for characterizing BRDF of reference plates

42. Methodology (Example of ) SOC Camera Reference Plates Equipment controlling Camera View Angle Characterizing diurnal BRDF of Reference Plates at UA campus, clear sky condition

43. NDVI-Leaf Fresh Mass Tr9 Tr11Tr13Tr118 Tr14 Each point represents one time sample from 1-m sun branch; all the branches selected here for each tree have very similar branch diameter across the time.

44. Fresh mass (g) @ 1m branch Tr9 Fresh mass (g) @ 1m branch Tr11 NDVI-Leaf Fresh Mass

45. Spectra-Age Model Temporal changes in spectral reflectance (Roberts et al., 1998 ) Example species: Aldina (species name), Fabaceae (family name)

46. Spectra-Age Model Temporal changes in spectral reflectance (Roberts et al., 1998 ) Example species: Aldina (species name), Fabaceae (family name)

47. Spectra-Age Model Temporal changes in spectral reflectance (Roberts et al., 1998 ) Spectra-Age Model by Doughty and Goulden (2008)