Advancing  Techniques  for  Informing    Terrestrial  Ecosystem  Models  with    Leaf  and  Imaging  Spectroscopy  to    

Improve  the  Representa?on  and  Predic?on  of  Vegeta?on  Dynamics  and  Carbon  Cycling  

 

Ankur  R.  Desai,  Sean  DuBois,  Shawn  P.  Serbin,    Toni  T.  Viskari,  Michael  C.  Dietze,  and    

Philip  A.  Townsend    

#EuroSpec2013  Final  Conference,  Nov  6-­‐8  2013,  Trento,  Italy    

Terrestrial  carbon  cycle  feedback  is  a  leading  order  uncertainty  for  climate  simula5on  

IPCC  AR5  WG1  CH6  (draB)  

This  is  true  at  the  site  level,  too!  Willow  Creek  EC  Tower  Site,  Wisconsin    

www.pecanproject.org  

NEE  

GPP  

Reco  

Leaf  and  canopy  high-­‐spa5al  and  spectral  resolu5on  spectroscopy  to  the  rescue?  

(a)  Fresh   (b)  Dry  

Acer  saccharum  

Imaging  spectroscopy  has  a  wealth  of  underu5lized  observa5ons  for  ecosystem  models!  

nm

nm

TAFKAA  

Raw  DN  

Cal.  Radiance  

Reflectance  

ER-­‐2  AVIRIS   AVIRIS  Image  Cube  

il  menu  del  giorno  

•  Two  possible  pathways:  – Data  products  assimila5on  – Direct  op5cal  proper5es  assimila5on  

•  Based  on  results  from  two  field  projects:  – ChEAS  Ameriflux  Cluster  – NASA  HyspIRI  prep  mission  

•  With  one  assimila5on  system  /  model:  – PEcAn  with  ED2  

PEcAn:  Predic5ve  Ecosystem  Analayzer  is  a  workflow  for  model-­‐data  assimila5on  

Ecosystem  C  /  H20  /  E  

ED2  is  a  dynamic  ecosystem  model  and  already  includes  broadband  radia5ve  transfer  

Medvigy  et  al  2009  

PEcAn  variance  decomposi5on  provides  informa5on  on  sensi5vity  of  a  model  output  variable  (e.g.,  NPP)  to  uncertainty  in  input  data  

(CV),  model  sensi5vity  (elas5city),  and  joint  variance  

165  Plots    120+  AVIRIS  Scenes   NASA  FFT  Project   NASA  HyspIRI  Campaign  

Two  study  regions  in  US  

2013-­‐2014  Spring    Summer    Fall  

Singh,  Serbin,    McNeil,  Townsend.  (in  prep)  Eco.  Apps.  

ChEAS  

Of  a  total  of  145  scenes,  26  in  ChEAS:  All  midsummer  (July/August)  images  7.0m  -­‐  16.8m  resolu?on  (low/high  alt.  ER-­‐2)  

ChEAS:  AVIRIS  over  4  Ameriflux  sites  

DATA  PRODUCT  ASSIMILATION  

MODIS  LAI  +  Flux  tower  ver5cal  PAR  profile  tames  model  phenology  

Viskari  et  al.,  (in  prep)    

Prior  Posterior  

Filled  dot  =  MODIS  LAI,  open  dot  =  LAI  from  flux  tower  profile  FaPAR  

2002  

2003  

2004  

2005  

SPRING   FALL  

AVIRIS  products  generated  with  PLSR  technique  and  leaf-­‐level  spectroscopy  calibra5on  

Serbin  et  al.,  2012  J  Exp  Botany  

Where  we  are  today:    

Standardized  algorithms  to  predict  foliar  cons5tuents  (%C,%N,LMA,…)  using  spectroscopy  across  diverse  forest  types  w/  uncertainty  es5amtes.    

•  Our  scaling  methods  propagate  the  uncertain5es  at  the  leaf-­‐level  through  the  canopy  PLSR  modeling  to  produce  es5mates  of  foliar  traits  (i.e.  trait  maps)  with  an  es5mate  of  the  associated  retrieval  uncertainty  (pixel  by  pixel).      •  Thus,  we  can  u5lize  these  products  in  a  model  DA  framework  given  that  we  have  quan5fied  uncertainty  in  retrievals.  

Model-­‐data  Assimila5on:  AVIRIS  Willow  Creek  EC  Tower  Site,  Wisconsin    

www.pecanproject.org  

Working  toward  the  assimila5on  of  AVIRIS-­‐derived  products.    These  include  foliar  chemistry  (e.g.  N,  C,  CN,  lignin)  and  morphology  (SLA).        

Model-­‐data  Assimila5on:  AVIRIS  Willow  Creek  EC  Tower  Site,  Wisconsin    

www.pecanproject.org  

• Of  course  this  requires  uncertainty  for  proper  DA.  Otherwise  too  much  weight  given  to  the  RS  es5mates  causing  overconfidence.    Therefore,  our  methods  u5lize  the  genera5on  of  AVIRIS  retrieval  uncertainty  to  properly  assimilate  datasets!  

DIRECT  OPTICAL  ASSIMILATION  

Scalering  elements,  LADF,  LAI  

Absorp5on  

Transmilance  

Reflectance  

Canopy Reflectance Variability 241

and transmittance properties, quantification of the rela- 1400 nm and 1900 nm were due to water vapor, pre-venting measurements and subsequent modeling of thesetive importance of water, carbon, and nitrogen continues

to be difficult (Curran et al., 1992; Fourty et al., 1996). spectral regions. Soil reflectance variability was highestin the SWIR2 (c.v.!18–20%) and lowest in the VISStanding litter and woody stem optical properties

were generally more variable compared to leaves (Fig. (c.v.!14–16%).2). Litter reflectance and transmittance c.v.’s were 8–35% (lowest in VIS, highest in SWIR2) and 23–51% Leaf Optical Variability at Canopy Scales(lowest in NIR, highest in SWIR2), respectively. These For analyses in this and the following two sections, leafresults were similar to those found for the Texas data set and NPV optical properties were taken from Figures 1(Asner et al., 1998b). The wide-ranging litter optical val- and 2 unless noted otherwise. The darkest soil spectrumues are primarily due to differences in residual water from the field data (Fig. 3) was used to better isolate thecontent as well as species-specific differences in carbon effects of changing leaf and canopy structural conditions(e.g., lignin, cellulose) concentration (Murray and Wil- (to minimize soil effects). Figure 4 shows the role of leafliams, 1987; Asner, unpub. data). optical variability (4 s.d. in Fig. 1) on nadir-viewed canopy

Coefficients of variation for woody stem reflectance reflectance as simulated for a canopy of low and high LAIranged from 20% to 45% (highest in NIR region near 900 (1.5 and 6.0). The former is a common LAI scenario fornm, lowest in VIS). This high variability likely occurred grasslands, while the latter is typical of forest canopiesfrom differences in surface moisture and carbon constit- (Table 4). In the low LAI scenario, leaf optical variabilityuents of the outermost bark (Asner and Wessman, 1997); played a relatively small role in driving canopy reflectancehowever; stem spectral–biochemical relationships have not variation (Fig. 4a). The total range in canopy reflectancebeen quantitatively established. In general, woody stem resulting from leaf optical variation was "1% in the VISand litter spectra showed fewer water absorption features and a maximum of 4% in the NIR (Fig. 4c). At high can-than green leaf material, allowing the features associated opy LAI, the effects of leaf optical variability were morewith lignin and other organic compounds (at roughly pronounced, causing maximum variation in canopy re-1700 nm, 2000 nm, and 2200 nm) to emerge in the spec- flectance in the NIR (14–18%; Fig. 4b). In the VIS andtra (Curran et al., 1992; Verdebout et al., 1994; Jacque- along the “red edge” (!700 nm), effects of leaf-levelmoud et al., 1996). variation were still extremely small ("0.5%; Figs. 4b,c).

Mean (#1 s.d.) soil reflectance is shown in Figure 3. Leaf effects at canopy scales were greater in the SWIR1Strong atmospheric absorption features centered near than in the SWIR2 because the single scattering albedo

(reflectance$transmittance) of fresh green leaves ishigher in the SWIR1.Figure 2. A) Mean (#1 s.d.) standing litter reflec-

tance (solid lines) and transmittance (dotted lines)Structural Effects on Canopy Reflectance Propertiesproperties collected from sites in the United States

and Brazil; B) mean (#1 s.d.) woody stem reflec- Canopy LAI variation had a strong influence on reflec-tance. tance signatures (Fig. 5a), with the most pronounced ef-

fect in the NIR and the smallest effect in the VIS region.Structure enhances canopy reflectance in spectral regionswhere the scatterers are “bright” (e.g., NIR for green

Figure 3. Mean (#1 s.d.) soil reflectance spectracollected at sites in the United States and Brazil.Strong atmospheric water absorption near 1400nm and 1900 nm prevented measurements inthose spectral regions.

Asner  et  al.  (1998)  RSE  

Canopy Reflectance Variability 241

and transmittance properties, quantification of the rela- 1400 nm and 1900 nm were due to water vapor, pre-venting measurements and subsequent modeling of thesetive importance of water, carbon, and nitrogen continues

to be difficult (Curran et al., 1992; Fourty et al., 1996). spectral regions. Soil reflectance variability was highestin the SWIR2 (c.v.!18–20%) and lowest in the VISStanding litter and woody stem optical properties

were generally more variable compared to leaves (Fig. (c.v.!14–16%).2). Litter reflectance and transmittance c.v.’s were 8–35% (lowest in VIS, highest in SWIR2) and 23–51% Leaf Optical Variability at Canopy Scales(lowest in NIR, highest in SWIR2), respectively. These For analyses in this and the following two sections, leafresults were similar to those found for the Texas data set and NPV optical properties were taken from Figures 1(Asner et al., 1998b). The wide-ranging litter optical val- and 2 unless noted otherwise. The darkest soil spectrumues are primarily due to differences in residual water from the field data (Fig. 3) was used to better isolate thecontent as well as species-specific differences in carbon effects of changing leaf and canopy structural conditions(e.g., lignin, cellulose) concentration (Murray and Wil- (to minimize soil effects). Figure 4 shows the role of leafliams, 1987; Asner, unpub. data). optical variability (4 s.d. in Fig. 1) on nadir-viewed canopy

Coefficients of variation for woody stem reflectance reflectance as simulated for a canopy of low and high LAIranged from 20% to 45% (highest in NIR region near 900 (1.5 and 6.0). The former is a common LAI scenario fornm, lowest in VIS). This high variability likely occurred grasslands, while the latter is typical of forest canopiesfrom differences in surface moisture and carbon constit- (Table 4). In the low LAI scenario, leaf optical variabilityuents of the outermost bark (Asner and Wessman, 1997); played a relatively small role in driving canopy reflectancehowever; stem spectral–biochemical relationships have not variation (Fig. 4a). The total range in canopy reflectancebeen quantitatively established. In general, woody stem resulting from leaf optical variation was "1% in the VISand litter spectra showed fewer water absorption features and a maximum of 4% in the NIR (Fig. 4c). At high can-than green leaf material, allowing the features associated opy LAI, the effects of leaf optical variability were morewith lignin and other organic compounds (at roughly pronounced, causing maximum variation in canopy re-1700 nm, 2000 nm, and 2200 nm) to emerge in the spec- flectance in the NIR (14–18%; Fig. 4b). In the VIS andtra (Curran et al., 1992; Verdebout et al., 1994; Jacque- along the “red edge” (!700 nm), effects of leaf-levelmoud et al., 1996). variation were still extremely small ("0.5%; Figs. 4b,c).

Mean (#1 s.d.) soil reflectance is shown in Figure 3. Leaf effects at canopy scales were greater in the SWIR1Strong atmospheric absorption features centered near than in the SWIR2 because the single scattering albedo

(reflectance$transmittance) of fresh green leaves ishigher in the SWIR1.Figure 2. A) Mean (#1 s.d.) standing litter reflec-

tance (solid lines) and transmittance (dotted lines)Structural Effects on Canopy Reflectance Propertiesproperties collected from sites in the United States

and Brazil; B) mean (#1 s.d.) woody stem reflec- Canopy LAI variation had a strong influence on reflec-tance. tance signatures (Fig. 5a), with the most pronounced ef-

fect in the NIR and the smallest effect in the VIS region.Structure enhances canopy reflectance in spectral regionswhere the scatterers are “bright” (e.g., NIR for green

Figure 3. Mean (#1 s.d.) soil reflectance spectracollected at sites in the United States and Brazil.Strong atmospheric water absorption near 1400nm and 1900 nm prevented measurements inthose spectral regions.

Canopy  Refl:  f(Incident  rad,  leaf  op5cs,  stem/soil/liler  op5cs,  stem  density,  crown  structure,  LADF,  LAI,  dispersion  of  leaves,  [e.g.  random  or  clumped])  

Asner  et  al.  (1998)  RSE  

Diffuse  rad  penetra5on  

Direct  rad  penetra5on  

Soil  Reflectance  

NPV  

Canopy  RTM  Inversion    RMSE  =  0.015  LAI  =  4.89  

PROSPECT  Inversion  (Modeled  Op5cs)    Cab  =  58.3    EWT  =  0.020  cm  SLA  =    18.4  m2  /  kgC    

Single  Scalering  Albedo    Update  Eco.  Model  leaf  op5cs  (refl.  /  trans)    

AVIRIS  

Albedo  

*constrained  by  data  

*  

*  *  

*  

*  

*  

*  

Tsoil  

*  

*  *  

*  

*  

*  

*  

*constrained  by  data  

GPP  

*  

*  *  

*  

*  

*  

*  

*constrained  by  data  

Pros  and  Cons:  Product  assimila5on  

•  Product  assimila5on  typically  requires  less  modifica5on  of  most  ecosystem  models  –  Tradeoff  is  uncertainty  of  product  directly  propagates  into  uncertainty  in  model  

– Model  and  product  make  different  assump5on  about  canopy  architecture  –  bias  is  likely  if  the  two  are  fundamentally  different  (also  scale  dependent)  

–  Computa5onal  cost  for  radia5ve  transfer  based  parameter  inversion  is  done  at  product  stage  instead  of  during  model  execu5on  

–  Characterizing  product  uncertainty  as  important  as  actual  value  for  data  assimila5on  approaches  

Pros  and  Cons:  Direct  op5cal  assimila5on  

•  Op5cal  assimila5on  requires  iden5fica5on  of  proper  canopy  radia5ve  transfer  model  –  Increases  parameters,  but  possibly  allows  for  op5cs  to  directly  guide  model  improvement  without  a  priori  assump5ons  of  what  spectral  signatures  mean  

–  Similarly,  no  bias  from  difference  in  assump5on  of  canopy  architecture  

–  Easily  extendable  to  many  remote  sensing  plasorms  –  Ini5al  model  investment  is  high  and  model  canopy  may  not  be  well  suited  for  radia5ve  transfer  

A  look  forward  •  HyspIRI  (hlp://hyspiri.jpl.nasa.gov/)  or  similar  future  

satellites  (EnMAP;  hlp://www.enmap.org/)  along  with  con5nuous  canopy  spectral  measurements  (SpecNet)  will  drama5cally  increase  the  volume  of  spectral  informa5on  in  Visible,  near  IR,  and  thermal  wavelengths  

•  We  can  do  more  than  just  make  prely  pictures  and  poorly  validated  “products”  -­‐  need  to  move  away  from  exclusively  using  vegeta5on  indices  

•  The  need  to  reduce  terrestrial  carbon  cycle  model  parameters  is  urgent  and  methods  to  assimilate  spectral  informa5on  directly  into  models  is  limited  to  date  

•  Spectral  databases  (e.g.  EcoSIS,  SPECCHIO)  can  be  mined  for  key  PFT-­‐level  informa5on  useful  for  constraining  model  projec5ons  

!"#$ %&'( )#*!%+*%! , #* %-( #./(. #0 123 !2 4-( !"#$ %&'(5.(!6#*!( #0 %-( !(*!#.! 7!(/ 89 :(!;+./&*! < =('#*>?@ABC 6.#'6%(/ D+..+%% >?@A3C %# ).&%&E7( %-#!( '(+!7.(5'(*%! +*/ )#*)"7/( %-+% %-( FGH 0"7I(! !700(.(/ 0.#'"+.J( (..#.! >!B1KC 8()+7!( %-(9 $(.( 7*+8"( %# !+'6"(%-( -&J- 0.(E7(*)9 6#.%&#* #0 %-( 0"7I )#!6()%.7'24-( *(I% $+L( #0 %()-*#"#J&)+" &'6.#L('(*%! )+'(

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`7**&*J !" #$2M ?@@@P =+$ !" #$2M!78'&%%(/C2 F7..(*%"9M %-( (//9 )#L+.&+*)( '(%-#/ &!8(&*J 7!(/ +% #L(. ?O1 !&%(! $#."/$&/(M +! 6+.% #0 %-(\=ab]_4 6.#J.+' >N+"/#))-& !" #$2M H11?+C +*/ &*L#"L(!*($ .(J&#*+" *(%$#.R! &* ]#.%- V'(.&)+ >\"7I*(%M

F+*+/+CM N.+c&"M V!&+ >V!&+\"7ICM V7!%.+"&+ >Gc\"7IC +*/V0.&)+2

!"#$%#&'() *$+ %,$%(&-)

4-( +%'#!6-(.( )#*%+&*! %7.87"(*% '#%&#*! #0 76$+./+*/ /#$*$+./ '#L&*J +&. %-+% %.+*!6#.% %.+)( J+!(!!7)- +! FGH2 4-( (//9 )#L+.&+*)( %()-*&E7( !+'6"(!%-(!( %7.87"(*% '#%&#*! %# /(%(.'&*( %-( *(% /&00(.(*)(#0 '+%(.&+" '#L&*J +).#!! %-( )+*#695+%'#!6-(.( &*%(.50+)(2 d* 6.+)%&)(M %-&! %+!R &! +))#'6"&!-(/ 89 !%+%&!%&)+"+*+"9!&! #0 %-( &*!%+*%+*(#7! L(.%&)+" '+!! 0"7I /(*!&%9>%"&!)M !'#"'#H !#?CM 7!&*J `(9*#"/!e .7"(! #0 +L(.5+J&*J >`(9*#"/!M ?O@3CM $-&)- +.( /(!).&8(/ 8("#$2 4-(6.#/7)% #0 %-&! #6(.+%&#* &! + .("+%&#*!-&6 %-+% (I6.(!!(!%-( '(+* 0"7I /(*!&%9 #0 FGH +L(.+J(/ #L(. !#'( %&'(!6+* >!7)- +! +* -#7.C +! %-( )#L+.&+*)( 8(%$((* 0"7)%75+%&#*! &* L(.%&)+" L("#)&%9 >&C +*/ %-( FGH '&I&*J .+%&#>'" !)f!+ $-(.( !+ &! +&. /(*!&%9 +*/ !) &! FGH /(*!&%9Cg

% " !!!+ $ &%'%" &?'

d* _E* ?M %-( #L(.8+.! /(*#%( %&'( +L(.+J&*J +*/ 6.&'(!.(6.(!(*% 0"7)%7+%&#*! 0.#' %-( '(+* >(2J2 '"" '# !'' C2V 6#!&%&L("9 !&J*(/ )#L+.&+*)( .(6.(!(*%! *(% FGH %.+*!50(. &*%# %-( +%'#!6-(.( +*/ + *(J+%&L( L+"7( /(*#%(! %-(.(L(.!(2

()"!*+*!",)- !../ '01#*,#)'! 2!#34*!2!)"3

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h#$ %-&! 6.#)(!! &! +))#'6"&!-(/ &! /&!)7!!(/*(I%2\#. %-( )+!( #0 FGHM %-( )#*!(.L+%&#* #0 '+!! !%+%(!

%-+% %-( !7' #0 %-( "#)+" %&'( .+%( #0 )-+*J( #0 %-( FGH

'&I&*J .+%&# >%(.' dCM 'M +*/ +/L()%&#* >%(.' ddC &! 8+"5+*)(/ 89 %-( !7' #0 %-( 0"7I /&L(.J(*)( #0 FGH &* %-(L(.%&)+" >6CM "+%(.+" >/C +*/ "#*J&%7/&*+" >7C /&.()%&#*!>%(.' dddC +*/ %-( 8&#"#J&)+" !#7.)( !&*R5!%.(*J%- >8NC>%(.' dWCg

! H11Z N"+)R$("" T78"&!-&*J =%/M 9$0:#$ ;<#)-! =,0$0-/M ./ BA@,B@H

012 : 2 : 2 N V =:GFFh d

Net CO2 Flux (mmol m-2 s-1)

Param  Est.  

Es5mates  of  Vcmax  and  Jmax  derived  from  imaging  spectroscopy  will  be  compared  with  those  inverted  from  flux  tower  data  at  each  site  using  a  coupled  2-­‐layer  (sunlit-­‐shaded  canopy)  FvCB  ecosystem  model.  Uncertainty  will  be  assessed  using  Bayesian  parameter  inversion.  

NASA  HyspIRI  California  Campaign  Eddy  Covariance  

HyspIRI  overflies  a  range  of  Mediterranean  and  Western  Pine  ecosystem  flux  tower  sites  

•  Clima5c  &  topographic  varia5on  in  fluxes    •  Water  availability  is  key  

Subalpine  Forest  (high  eleva?on)   Mixed  Conifer  Forest  (high  eleva?on)  

Ponderosa  Pine  Forest  (mid  eleva?on)   Oak-­‐pine  Woodland  (low  eleva?on)  

Coastal  Sage  (low  eleva?on)   Coastal  Grassland  (low  eleva?on)  

Desert  Chaparral  (low  eleva?on)  

Thank  you  

•  Many  collaborators  in  the  field  in  Wisconsin  and  California,  ED2  model  developers,  PEcAn  data  assimila5on  crew,  NASA  AVIRIS  team  

•  More:  hlp://pecanproject.org/    •  Funding:  NASA  ROSES  Hyspiri  Preparatory  NNX12AQ28G,  NSF  Advances  in  Biological  Informa5cs  DBI-­‐1062204.