You are on page 1of 19

Parameteriza)on

of Forest Canopies with the PROSAIL Model


Michael J. Austerberry, Creighton University Dr. Susan Us)n, UC Davis Shane Grigsby, UC Santa Barbara

Project Overview
Goals
Determine important vegeta)ve parameters Observe eects of varying each one Accurately model reectance from forest canopy Determine physical characteris)cs of deciduous canopies from reectance spectra

Atmospheric and terrestrial modeling depend on accurate knowledge of radia)ve transfer Hand measurements of trees are cumbersome

Data Sources
Terrestrial
Hand measurements of tree dimensions Destruc)vely sampled branches Georeferenced data sites

Airborne Imaging
NSERC DC-8
MASTER: 14m ground resolu)on, 25 Near IR spectral bands

NASA ER-2
AVIRIS: 35m ground resolu)on, 224 Near IR spectral bands

PROSAIL Model
Arboreal simula)on, outputs reec)vity spectrum

Vegeta)ve Parameters
Leaf Area Index
LAI: Single-sided area of leaves per unit area of ground surface (m2/ m2)
Typical range = 1 (grass) to 16 (dense conifer)

Leaf Inclina)on Angle/Leaf Angle Distribu)on


Average angle of incidence with at plane LAD: Width of Gaussian distribu)on around mean
LAD does not aect PROSAIL output

This is just two of MANY important characteris)cs!

Loca)ng Tree Pixels in AVIRIS


Sensor and Solar geometry are recorded as input for PROSAIL model Red box = tree pixel

(Sensor Azimuth) (Sensor Zenith) (Solar Azimuth) (Solar Zenith)


Modeling with PROSPECT + SAIL


PROSPECT
Receives chemical input Outputs specically LEAF spectral data

SAIL
Receives LAI, Leaf Angle, Solar Geometry input Outputs general CANOPOY reectance spectra

PROSAIL
Combines PROSPECT and SAIL inputs Returns a higher-resolu)on reectance spectra

Modeling with PROSAIL: BRDF


Chemical data combines with structural parameters and input for Bidirec)onal Reectance Distribu)on Func)on AVIRIS records geometry between sensor, ground target, and sun
Point Long Lat SolAz SolZen SenAz SenZen Cab 1 -119.7313 37.107312 208.62364 23.716497 86.023109 6.24447 2 -119.7311 37.10679 208.61858 23.714979 86.28775 6.547404 3 -119.7316 37.107436 208.62012 23.715893 86.392708 6.645448 Car 48.7 67.2 66.6 10.01 12.62 9.9 Cbrown 0.1 0.1 0.1 Cw 0.01398 0.01647 0.01513

Iterated PROSAIL Modeling


For each of the three trees, PROSAIL was run 25 )mes with incremented input values LAI varied from 2 to 4, in steps of .5 Leaf Angle varied from 15 to 75, in steps of 15
2, 15 2.5, 15 3, 15 3.5, 15 4, 15 2, 30 2.5, 30 3, 30 3.5, 30 4, 30 2, 45 2.5, 45 3, 45 3.5, 45 4, 45 (LAI, LEAF ANGLE) 2, 60 2.5, 60 3, 60 3.5, 60 4, 60 2, 75 2.5, 75 3, 75 3.5, 75 4, 75

Iterated PROSAIL Modeling


75 2.0

Reectance

Reectance

15
Wavelength (nm)

4.0

Wavelength (nm)

Varied Leaf Angle (Blue = 15, Black = 75)


LAI = 2

Leaf Angle = 15 Varied LAI (Blue = 2, Black = 4)

Linear Regression
Python func)on stats.linregress relates each trees water-masked tree spectra to all 25 outputs
Reectance (*10000)

Tree 1

Tree 2

Tree 3

Wavelengths (nm) 2, 15, 2.5, 15, 3, 15, 3.5, 1.5, 4, 15, 2, 30, 2.5, 30, 3, 30, 3.5, 30, 4, 30, 2, 45, 2.5, 45, 3, 45, 3.5, 45, 4, 45, 2, 60, TREE 3 2.5, 60, TREE 2 3, 60, TREE 1 3.5, 60, 4, 60, 2, 75, 2.5, 75, 3, 75, 3.5, 75, 4, 75,

Tree 1
2, 15, .8401 2.5, 15, .8204 3, 15, .8091 3.5, 1.5, .8027 4, 15, .7992 2, 15, .7486 2.5, 15, .7241 3, 15, .7103 3.5, 1.5, .7026 4, 15, .6984 2, 15, .6437 2.5, 15, .6181 3, 15, .6040 4, 15, .5923 2, 30, .8807 3, 30, .8262 3.5, 30, .8115 4, 30, .8021 2, 30, .7995 2.5, 30, .7587 3, 30, .7317 3.5, 30, .7141 2.5, 30, .8482

R-Squared 2.5, 45, .9138 Values 2.5, 60, .9669


2, 45, .9478 3, 45, .8824 3.5, 45, .8565 4, 45, .8362 2, 60, .9488 3, 60, .9684 3.5, 60, .9602 4, 60, .9466 2, 60, .9375 2.5, 60, .9414 3, 60, .9311 3.5, 60, .9131 4, 60, .8913

2, 75, .6783 (worst) 2.5, 75, .7034 3, 75, .7221 3.5, 75, .7370 4, 75, .7492 2, 75, .7106 2.5, 75, .7312 3, 75, .7464 3.5, 75, .7582 4, 75, .7680 2, 75, .8113 2.5, 75, .8279 3, 75, .8397 3.5, 75, .8488 4, 75, .8561

Tree 2
2, 45, .8899 2.5, 45, .8425 3, 45, .8020 3.5, 45, .7697

4, 30, .7029 (worst) 4, 45, .7450 2, 30, .7026 2.5, 30, .6570 3, 30, .6278 4, 30, .5977 2, 45, .8196 2.5, 45, .7581 3, 45, .7094 3.5, 45, .6725 4, 45, .6450

Tree 3
2, 60, .9452 2.5, 60, .9262 3, 60, .8976 3.5, 60, .8648 4, 60, .8313

3.5, 1.5, .5964 (worst) 3.5, 30, .6093

Tree 1 Best Correla)on


Reectance (x10000)

LAI = 3.0 Leaf Angle = 60 R-Squared = .9684

Wavelength (nm)

Tree 2 Best Correla)on


Reectance (x10000)

LAI = 2.5 Leaf Angle = 60 R-Squared = .9414

Wavelength (nm)

Tree 3 Best Correla)on


Reectance (x10000)

LAI = 2.0 Leaf Angle = 60 R-Squared = .9452

Wavelength (nm)

Final Result
Extracted LAI and Leaf Angle Verica)on provided by NDVI-Extracted LAI and airborne LiDAR Beer-Lambert law Es)mates do not match perfectly, but they do correspond
R-Squared TREE 1 TREE 2 TREE 3 .968425 .9414192 .9451759 LEAF ANGLE 60 60 60 LAI 3.0 2.5 2.0 NDVI- Beer- Extracted LAI Lambert LAI 2.8-3.5 3.0 2.8-3.1 3.5 3.1 2.8

Conclusion
Provides verica)on of the PROSAIL model Advances understanding of surface radia)on budget under vegeta)ve cover Future research may enable parameteriza)on of canopy structure exclusively from airborne measurement

Acknowledgements
NSERC SARP Emily Schaller Rick Sheoer Susan Us)n Shane Grigsby The Us)n Land Research Group University of California Kayla Ryan Rachel Glade Dion Kucera

Ques)ons?

Sources/Works Cited
Jacquemond, S. et al, 2000: Comparison of Four Radia)ve Transfer Models to Simulate Plant Canopoies Reectance: Direct and Inverse Mode. Remote Sensing of Environment, 74, 471-481. Jacquemond et al, 2009: PROSPECT + SAIL models: A review of use for vegeta)on characteriza)on. Remote Sensing of Environment, 113, 556-566. Houborg, R., Anderson, M., Daughtry, C., 2009: U)lity of an image-based canopy reectance modeling tool for remote es)ma)on of LAI and leaf chlorophyll content at the eld scale. Remote Sensing of Environment, 113, 259-274. Zheng, G., Moskal, M., 2012: Leaf Orienta)on Retrieval from Terrestrial Laser Scanning (TLS) Data. In Press.

You might also like