NASA Technical Reports Server (NTRS) 19990108576: Physics Based Modeling and Rendering of Vegetation in the Thermal Infrared

5 Pages·0.39 MB·English

by  NASA Technical Reports Server (NTRS)

Physics Based Modeling and Rendering of Vegetation in the Thermal Infrared
J. A. Smith J. R. Ballard, Jr.
Laboratory for Terrestrial Physics Environmental Laboratory
NASA Goddard Space Flight Center USAE Waterways Experiment Station
Greenbelt, MD 20771 Vicksburg, MS 39180
jasmith @hemlock.gsfc.nasa.gov ballard @orion.wes.army.mil
Abstract its 1kilometer instantaneous field-of-view. However, with
the next generation, high spatial resolution satellite systems,
more attention must be focused on the three-dimensional
We outline a procedure for rendering physically-based
thermal infrared images of simple vegetation scenes. Our structure of underlying scenes. The new satellite systems
approach incorporates the biophysical processes that affect have spatial resolutions of 1to 30 meters which is compa-
the temperature distribution of the elements within ascene. rable to many intrinsic scene characteristics, e.g. crop crow
Computer graphics plays akey role in two respects. First, in structure.
computing the distribution ofscene shaded and sunlit facets Remote sensing researchers have already begun to draw
and, second, in the final image rendering once the tempera- upon traditional computer graphics techniques to aid them
tures of all the elements in the scene have been computed. in their modeling of more complex scenes, particularly in
We illustrate our approach for asimple corn scene where the optical reflective regime. Borel, et al. [1] and Goel,
the three-dimensional geometry is constructed based on et al. [3] used the radiosity method developed by Green-
measured morphological attributes of the row crop. Statis- berg and associates [4] to model the multi-spectral reflec-
tical methods are used to construct a representation of the tive response of three-dimensional plant canopies. Smith,
scene in agreement with the measured characteristics. et al. [9] also employed a statistical version of the radios-
Our results are quite good. The rendered images ex- ity method to compute short-wave absorption of a layered
hibit realistic behavior in directional properties as afunc- canopy for subsequent use in a one-dimensional thermal in-
frared model.
tion of view and sun angle. The root-mean-square error in
measured versus predicted brightness temperatures for the While progress has been made in the modeling of
scene was 2.1 deg C. the infrared response of three-dimensional scenes [6, 7],
the modeling of the thermal infrared behavior of three-
dimensional vegetation canopies is still in the beginning
1. Introduction phases [8].
In this paper, we discuss a novel technique for combin-
ing computer graphics, e.g. ray tracing, with physics-based
Remote sensing requires the interpretation of measured
energy budget modeling to predict the thermal infrared ex-
sensor response to extract information about earth surface
itance of a plant canopy. We illustrate the method for an
features. We can use remote sensing to monitor biogeo-
agricultural scene where meteorological, crop morphologi-
physical attributes and assess environmental conditions.
cal, and thermal infrared measurements were available.
The interpretation of such signals, however, is a complex
process dependent on many factors including sensor re-
2. Method
sponse function, atmospheric state, and the interplay of the
probing electromagnetic radiation (passive or active) with
the underlying scene. Figure 1 gives a flow chart of the calculation steps
Models play a strategic role in developing information required in our approach. First, we generate the three-
extraction algorithms or in sensor design studies. Tradition- dimensional geometry of the scene using simple measure-
ally, these models have been one-dimensional, plane paral- ments of the canopy including crop row spacing, height, leaf
lel abstractions. These have proven useful for the current, slope distribution, and a measure of the density of leaf area
large-scale global monitoring satellite systems such as the per unit ground area, the Leaf Area Index (LAD. The details
NOAA Advanced Very High Resolution Radiometer with are described below.
Wethenuseraytracingtodeterminaelloftheelements vapor pressure at canopy temperature Tv, e_,the vapor pres-
inthescenwehichareeithedrirectlyilluminatedi.e,.,sun- sure of air at air temperature T_, p is the air density, cp
lit,orinshadeF.oreachofthesesunlitorshadeedlements, the air specific heat, and 7 the psychromatic constant. Fi-
weseparaterluynathermamlodetlocalculatteheirenergy nally, rb is the canopy boundary layer resistance and rc is
budgeatndtosolveforthecorrespondilnegaforsoiltem- the bulk stomatal resistance. Canopy stomatal resistance
peratureWs.esubsequenatlgya,inu,seraytracingtorender and the boundary layer resistance depend upon meteorolog-
thethermainl fraredscenbeytracingthethermainl frared ical driving variables.
flux,emittedbyeachelemenattitsestimatetedmperature, Net radiation to asurface, R,_, is computed as the sum of
whichisactuallyreceivebdyasensoartaspecifievdiew the absorbed long-wave and short-wave radiation minus the
directionT.hethermaml odeilsdrivenbylocaml eteoro- outgoing long-wave emitted thermal infrared flux. Separate
logicaclonditioninscludinglocalwindspeedre,lativehu- calculations are required for sunlit versus shaded surfaces.
miditya, irtemperatuaren,dshort-waivlleuminatioanndre- For sunlit leaves, the absorbed shortwave flux is highly de-
quiresonlysimplemateriaplroperties. pendent on leaf inclination angle or the angle between the
Weusedanestedapproacthosimulattehegeometry leaf and the incident solar direction. Long-wave emission is
ofthescenebasedonmeasurecdanopyparameterWs.e a function of the unknown leaf temperature and is given by
modeleadsamplpelotof19mby19mforwhichcanopy the Stefan-Boltzmann relation.
measurememntes,teorologayn,dthermainlfraremd easure- Thus, a system of non-linear energy budget equations is
mentwsereavailableW.efirstgeneratetwdenty-fivuenique solved for individual leaf temperatures for all possible leaf
19mrowswithacanopwyidthof0.46m,heighotf0.8m, orientations corresponding to varying exposure of the indi-
totarlowspacinogf0.76mincludingfoliageandgapbe- vidual leaves to direct solar flux. For sunlit leaves, one de-
tweenrowsa, ndasinglesidedleafareaindexof3.0cor- gree increments are used for leaf inclination classes. For the
respondintogthemeasurerodwspacingh,eighta,ndLAI. case simulated, sunlit leaf temperatures temperatures range
Asphericlaelafslopedistributioannduniformleafazimuth from 24 to 32 deg C with warmer temperatures occurring
distributiownereusedtoassigonrientationtostheindivid- more frequently. For shaded leaves, the average leaf tem-
ualleaffacetsO. urprimitiveleaffacestizewas5cmby5 perature was 26.5 deg C. Average soil sunlit and shaded
cmprovidinagleafareaof0.002m52.Thus,10,48l8eaf temperatures corresponding to the measured meteorologi-
facetwsererequiretdomatchthemeasureLdAIof3.0dis- cal conditions were 55 and 46 deg C respectively.
tributeodverthe19mby0.46mrowareaT.hetwenty-five For the ray-tracing and rendering calculations we em-
rowswereinstancetodgetheartarowspacinogf0.3mto ployed the RADIANCE program developed by Ward [10].
simulataecornfoliageplotmeasurin1g9mby19mcon- Small modifications were required to handle the emission
tainingapproximate2l6y2,20f0acets. from both sides of the leaf facets.
Figure2isavisuaslimulatioonfahemispheripciaclture
ofthesceneT.hesunwas73.2degabovtehehorizonand
3. Results
solaarzimuthwas-8.8degmeasurefrdomthesouthT.he
sunlitandshadecdomponenotsfbothcanopayndsoilel-
ementasreapparenFt.romsuchimagesw,ecancompute Figure 3 displays a simulated thermal infrared corn im-
ashadme askforallelemenwtsithinthescenienordetro age corresponding to Figure 2 rendered at one-quarter me-
ter pixel size. High LAI or dense vegetation yields cooler
applytheappropriaetenergbyalanccealculatiodnisscussed
below. temperatures in the thermal image compared to the warmer
Individualelafenergbyudgetasrecalculateadsfollows soil facets which are primarily visible looking straight down
intermsofnetradiation, into the image.
Rn, sensible heat, H, and latent
The brightness temperature of asurface is defined as the
heat, LE, i.e.
equivalent black-body temperature of the surface, i.e. a sur-
R,, = LE + H (l)
face with emissivity 1.0, that would yield the same total
A standard resistance formulation [2] is used for the calcu- outgoing long-wave flux as the measured surface. Ground
lation of latent and sensible heat in terms of the unknown measurements of brightness temperature and meteorologi-
leaf temperature, T_,i.e. cal driving variables were available for the 19 m by 19 m
(To- Ta) sample area. As a check on our energy budget formula-
H_ - pep tions, we calculated net radiation for this area using Equa-
rb
tions (1) and (2) and simulated the three-dimensional ther-
LE_, = [hveg e*(Tv) - ea] pep (2) mal infrared response for the test plot corresponding to 600,
[ Jrc+ rb 7 800, 1000, 1200, 1400, 1600, 1800, and 2000 hours. Fig-
where Ta is the air temperature, hve9 is the relative humid- ure 4shows a comparison of the measured versus predicted
ity of the vegetation air space, e* is the saturation water brightness temperatures for these 8 time periods as calcu-
latedfromthehemisphericaalvlyerageodutgointghermal warranted.
infraredflux,assuminagnemissivityof 1.0.Theroot-
mean-squaerrerorwas2.1degreeCswithanr2of0.93. Acknowledgments
Theerrorbarscorrespontodtheestimatecdonfidencine-
tervalsfromtheregressiolinnefittothepredictevdersus
The research described in this paper was supported by
measurevdalues.
the Ecological Processes and Modeling Program at NASA
Headquarters. Assistance was also provided by the U. S.
Army Corps of Engineers Waterways Experiment Station,
4. Summary and Conclusions
Research, Development, Testing and Evaluation Program.
In this paper we have given a general outline for ren-
References
dering thermal infrared images of simple vegetation scenes
based on physical principles. Our approach incorporates
the biophysical processes that affect the temperature distri- [1] C.C. Borel, S. A.W.Gerstl, andB. J.Powers. The radiosity
bution of the vegetation elements within a scene. The ther- method inoptical remote sensing of structured 3-D surfaces.
Remote Sens. Environ., 36:13--44, 199I.
mal model is driven by readily available, standard meteo-
[2] D. M. Gates. Biophysical Ecology. Springer-Verlag, New
rological measurements including local wind speed, rela-
York, 1980.
tive humidity, air temperature, and short-wave illumination.
[3] N. S.Goel, I.Rozehnal, and R. L. Thompson. A computer
Computer graphics plays a key role in two respects. First, graphics based model forscattering from objects ofarbitrary
in computing the distribution of shaded and sunlit facets shapes inthe optical region. Remote Sens. Environ., 36:73-
within the scene and, second, in the final image rendering 104,1991.
once the temperatures of all the elements in the scene have [4] D.P. Greenberg, M.E Cohen, and K. E. Torrance. Radios-
been computed. ity: A method for computing global illumination. Visual
We illustrated our approach for a simple corn scene Comput., 2:291-297, 1986.
where the three-dimensional geometry was constructed [5] P. Prusinkiewicz and A. Lindenmayer. The Algorithmic
Beauty ofPlants. Springer-Verlag, New York, 1996.
based onmeasured macroscopic morphological attributes of
[6] H. E. Rushmeier and S.D. Tynor. Incorporating the BRDF
the row crop-viz crop height, row spacing, leaf-area-index, into an Infrared Scene Generation System. Proccedings of
etc. Statistical methods were then used to randomly assign
the SPIE, 1311, 1990.
individual leaf facets with the proper orientations within [7] J.R. Schott, R.Raqueno, and C.Salvaggio. Incorporation of
rows and to construct a representation of the scene in agree- time-dependent thermodynamic model and aradiation prop-
ment with the measured characteristics. agation model into infrared three-dimensional synthetic im-
Our results are quite good. The rendered images ex- agegeneration. Opt.Eng., 31(7):1505-1516, 1992.
hibited realistic behavior in the directional properties as a [8] J. A. Smith, J. R. Ballard, and J. A. Pedelty. Effect of
three-dimensional canopy architecture on thermal infrared
function of view and sun angle [8]. The root-mean-square
exitance. Opt. Eng., 36:3093-3100, 1997.
error in measured versus predicted brightness temperatures
[9] J.A. Smith and S.M. Goltz. Updated thermal model using
for the scene was 2.1 deg C.
simplified short-wave radiosity calculations. Remote Sens.
Nevertheless, there are several limitations inour method.
Environ., 47:167-175, 1994.
First, our vegetation scene only consisted of leaves with no [10] G. J. Ward and R. Shakespeare. Rendering with RADI-
woody material, e.g. trunks. The time constant for indi- ANCE TheArt and Science ofLighting Visualization. Mor-
vidual leaves to reach thermal equilibrium is less than 30 gan Kaufmann Publishers, SanFrancisco, 1997.
seconds and our equilibrium energy-budget approach was
thus warranted. In soil, we used a simple trick based on re-
sults reported in the literature, i.e. modeling conduction as a
fraction of net radiation (possibly with time lag). In forests,
however, the woody material, e.g. trunks and branches,
must be accounted for. It is more challenging to incorpo-
rate the geometry of these elements. We are experimenting
with Lindenmayer systems [5].
Finally, we imposed a static wind field on the scene, con-
sistent with average wind speed measurements and a loga-
rithmic height profile. More realistic modeling of high fre-
quency phenomena, e.g. leaf flutter, opens up the pandora's
box of turbulence, a subject in its own right. We are in-
vestigating all of these effects and further work is clearly
Crop Row Spacing
aD;d,bIL:afioSl:l:e'-....../,.I Gener_ff s3cDnG:Omatry I
Direct/Diffuse
Illum_on
I
Usa Ray-Tracing to F.stlmate
Sunlit and Shaded Surfaces I
I
I
I Meteorologlcsl
Condltlons and
Thermal Modal I j Optical Pro_l_
I
RenIdmeargsTshermal I
Figure 1. Flow chart of model calculations
Figure 2. Simulated hemispherical visual image
100
20O
30O
4O0
5OO
6O0
700
BOO
2O
9OO
I000 100 200 300 400 500 600 700 800 900 1000
Figure 3. Simulated hemispherical thermal infared image
45
2
4O
! I_ns= 2.1 deg C :: :: :: :: T
_3o
. i i i
"_25
20
15 I I I " I I I I
18 20 22 24 26 28 30 32 34 36 38
Measured Brightness Temperature--deg C
Figure 4. Comparison of predicted versus measured canopy brightness temperature