Directional Reflectance Distrbutions of a Hardwood and Pine Forest
Transcription
Directional Reflectance Distrbutions of a Hardwood and Pine Forest
281 IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. GE-24, NO. 2, MARCH 1986 Directional Reflectance Distrbutions of a Hardwood and Pine Forest Canopy DANIEL S. KIMES, W. WAYNE NEWCOMB, ROSS F. NELSON, AND JOHN B. SCHUTT T HE COMPLEX radiant energy interactions that take place in various Earth surface scenes must be understood for future advancement in remote-sensing technology. Forests represent an important component of the Earth's surface that have a unique heterogeneous structure relative to homogeneous grasslands and agricultural crops. Few directional measurements of forest canopies have been reported in the literature [1]-[3]. The only study to report the directional reflectance of a forest canopy over the entire exitance hemisphere as a function of solar zenith angle was Kriebal [4]. Furthermore, no studies have physically analyzed the unique radiant transfers that take place in forest canopies. Understanding the scattering behavior of forest canopies is of importance to both regional and global remote-sensing studies. In this study, the directional reflectance distributions in AVHRR bands 1 (0.58-0.68 ,um) and 2 (0.73-1.1 ,um) were measured as a function of sun angle for both a hardwood and pine forest canopy at Beltsville, Maryland, in June from a helicopter platform. The reflectance distributions are reported and compared to the scattering behavior of agricultural and natural grassland canopies as measured and modeled by Kimes et al. [5]-[8]. In addition, we used the unique radiative transfer model of Kimes [6], [9] to extend our understanding of the physical principles causing the scattering behavior in forest canopies. Only a few physical analyses have been performed to quantitatively understand the physical mechanisms that cause the observed dynamics of directional reflectance distributions as a function of solar zenith angle, geometric structure (leaf orientation and plant spacing), and leaf and soil spectral properties [6], [7]. No such studies have been performed for forest canopies. Most radiative transfer models of vegetation assume infinitely extended horizontal layers of spatially homogeneous vegetation components over a Lambertian soil as reviewed by Smith [1], [2]. Several of these models have been modified to treat heterogeneous row crops as reviewed by Smith [1], [2]. There are several geometric optics models that have been developed for a single type of heteorogeneous scene. For example, Strahler and Li [10] modeled forest canopies using cones and Jackson et al. [11] used rectangular solids to model row crops. Other similar models are reviewed by Smith [1], [2]. In contrast, two models have been developed that have the capability of modeling radiant transfers in any 3-D heteorgeneous scenes of vegetation canopies (e.g., row crops, orchards, open forest, etc.) [6], [12]. The 3-D radiant transfer model of Kimes [6], [9] was applied to document the unique radiant transfers that take place in forest canopies due to their special geometric structure. Manuscript received April 3, 1985; revised September 13, 1985. D. S. Kimes, R. F. Nelson, and J. B. Schutt are with the Earth Resources Branch, NASA Goddard Space Flight Center; Greenbelt, MD 20771. W. W. Newcomb is with Republic Management Systems, Inc., Applications Project, Landover, MD 20785. IEEE Log Number 8407043. II. EXPERIMENT A. Radiometric Measurements All field data were collected at the Beltsville Agricultural Research Center, U.S.D.A., in June 1984 (Table I). All spectral radiance measurements were taken from a Abstract-The directional reflectance distributions for both a hardwood and pine forest canopy at Beltsville, Maryland, were measured in June as a function of sun angle from a helicopter platform using a hand-held radiometer with AVHRR band 1 (0.58-0.68 Mm) and band 2 (0.73-1.1 Mm). Canopy characteristics were measured on the ground. The reflectance distributions are reported and compared to the scattering behavior of agricultural and natural grassland canopies. In addition, the three-dimensional radiative transfer model of Kimes was used to document the unique radiant transfers that take place in forest canopies due to their special geometric structure. Measurements and model simulations showed that the scattering behavior of relatively dense forest canopies is similar to the scattering behavior of agricultural crops and natural grasslands. Only in more sparse forest canopies with significant spacing between the tree crowns (or clumps of tree crowns) does the scattering behavior deviate from homogeneous agricultural and natural grassland canopies. This clumping of vegetation material has two effects on the radiant transfers within the canopy: A) it increases the probability of gap to the understory and/or soil layers that increases the influence of the scattering properties of these lower layers; and B) it increases the number of low transmitting clumps of vegetation within the scene causing increased backscatter and decreased forward scatter to occur relative to the homogeneous case. Both effects, referred to as phenomenon A and B, respectively, tend to increase backscatter relative to forward scatter. For typical forest canopies, the peak backscatter reflectance can be increased as much as 30 percent relative to an equivalent homogeneous canopy due to phenomenon A and 35 percent due to phenomenon B. The combined effect of phenomenon A and B can cause typical increases of 65 percent or higher. It is hypothesized that these phenomena are especially important in sparse conifer forests, such as boreal forests, that account for 50 percent of the world's forest area. I. INTRODUCTION U. S. Government work not protected by U.S. copyright 282 IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. GE-24, NO. 2, MARCH 1986 TABLE I FOREST CANOPY MEASUREMENT TIMES AND CORRESPONDING SOLAR ZENITH ANGLES CANOPY DATE EASTERN STANDARD TIME SOLAR ZENITH ANGLE Hardwood 6/1/84 5:50 a.m. 790 6/7/84 7:15 a.m. 630 Pine 8:50 a.m. 450 10:50 a.m. 250 6/1/84 6:20 a.m. 740 6/7/84 7:35 a.m. 590 9:10 a.m. 410 10:50 a.m. 230 Xz. SENSOR =00 (a) -J z LLi 0. Bell-Jet Ranger helicopter at approximately 90-m altitude above the top of the canopy. Spectral directional radiances were taken in NOAA satellite 6-9 AVHRR bands 1 (0.58-0.68 ,um) and 2 (0.73-1.1 ,um) using a Mark-III three-band radiometer with a restricted 120 field of view. The circular footprint of the nadir looking radiometer at the top of the canopy had a diameter of 19 m, assuming a mean altitude of 90 m. For each measurement period, 41 directions were measured located at nadir and at 150 increments of off-nadir angle (15°, 300, 450, 600, and 750) and 450 increments of azimuth angle (00, 450, 900, 1350, 1800, 2250, 2700, and 3150). The 00 azimuth corresponds to the direction of the sensor looking toward the sun. Thus, an azimuth of 00 and 1800 represents forward scattering and backscattering, respectively. The coordinate system used is shown in Fig. 1. For each measurement period, two or three complete directional radiance distributions were taken at different sampling points within the middle of a homogeneous surface. This sampling procedure took less than 20 min. All directional radiance values were divided by the corrected radiance from a horizontal barium sulfate panel. The resulting values are reflectance factors [13]. The corrected radiance from a barium sulfate panel refers to corrections made for the non-Lambertian behavior of the reference panel for the specific irradiance conditions as described by Kimes and Kirchner [13]. For these corrections, the distribution of diffuse sky radiance was taken from the simulated data sets of Dave [14]. One mean distribution was calculated for each measurement period. All of the reflectance distributions were essentially symmetric about the principle plane of the sun. Therefore, corresponding data points on either side of the principle plane of the sun (Fig. 1) were averaged (e.g. azimuths 45° and 315°, 90° and 2700, and 1350 and 2250 and averaged for equal offnadir view angles). The coefficient of variation of reflectance for the various view directions and wavelength bands was on the order of 0.4. This statistic is reported to give the reader a feel for data dispersion. Keep in mind, 0 (b) Fig. 1. (a) Coordinate system defining solar and sensor angles and (b) polar plot showing scheme for plotting directional reflectance factors. The solar azimuth is always 1800. The sensors azimuth and off-nadir angles are shown as 1 and 0, respectively. A sensor with a 00 azimuth looks into the sun. Thus, an azimuth of 0° and 1800 represents forward scattering and backscattering, respectively. The spectral directional reflectance factors were plotted in a polar plot, where the distance from the origin represents the off-nadir view angle of the sensor and the angle from 0 = 00 represents the sensor's azimuth. The points show the directional measurements plotted. Lines of equal percent reflectance were contoured as presented in Fig. 7. Only 0- 180° azimuth is shown because azimuthal symmetry about the principle plane of the sun is assumed. The principle plane is defined as the plane perpendicular to the horizontal ground and contains the solar azimuth. however, that in many cases only four sample points were taken for each view direction. The method of plotting the directional reflectance factor distributions is described in Fig. 1. The radiometric data were collected for various solar zenith angles as reported in Table I. B. Site Description Ground data were collected in the hardwood and pine forest sites as shown in Fig. 2 to provide a general description of each site. Sample plots were selected by transect at each site. Ten plots were chosen in the larger hardwood tract; eight plots were chosen in the pine tract. Circular plots incorporating 0.0263 ha (9.14-m radius) were located in the hardwood tract. Due to significantly higher stem counts encountered on the pine site, two different plot sizes were used in order to reduce the number of trees tallied. Plots of 0.01 17 ha (6. 10-m radius) were used when 10 or more trees could be tallied within that sample area; plots of 0.0263 ha were used on less dense samples. Within each sample plot, the diameter at breast height (dbh) of all trees greater than 10.2 cm (4 in) were mea- 283 KIMES et al.: DIRECTIONAL REFLECTANCE DISTRIBUTIONS TABLE II CHARACTERISTICS OF THE HARDWOOD AND PINE TRACTS (The mean plus or minus one standard deviation is given where appropriate.) Hardwood Tract Predominate Tree Species Basal Area + (m2/hectare) Number of Stems' (Stems/hectare) Average Height (m) 26 ± 12 18 t 9 377 ± 198 22 ± 6 987 480 ± 11 ± 4 8 93 92 7 by Stem Count % Conifer 8 93 % hardwood 92 7 * Basal area is the area of the plane passed through the stem of the tree at right angles to the longitudinal axis of the tree. The sum of these wooden cross sectional areas per unit land area is given (in meter2 per hectare). These numvbers denote counts of stems with diameters greater than to 10.16 cm (4") at breast height. or equal available to them [15]. The hardwood tract is over 90 percent hardwood species based on basal area and stem counts. The Virginia pine is found for the most part on the periphery of the hardwood tract on the drier soils surrounding the bottomlands. The pine tract is significantly younger and smaller (in terms of basal area and height) than the hardwood tract. Though the trees on this pine tract are physically smaller, the stand is, like the hardwood tract, fully stocked. The number of stems found per unit area are over 2.5 times the counts found in the hardwood forest. Height uniformity, the purity of the stands found throughout this sandy, well-drained pine tract, and proximity to on-going agricultural activity suggest old field succession of Virginia pine. Other species include shortleaf pine (Pinus enchinata Mill.), sweet gum, northern red oak, and red maple. The pine tract is over 90-percent softwood species; 83 percent of those softwoods are Virginia pine, the remainder is shortleaf pine. Other important characteristics of each forest canopy were measured. At each sample site, the probability of gap (PGAP) through the forest canopy to the ground was measured as a function of off-nadir view angle (0). It was assumed that this function was constant with azimuthal orientation. Color photographs were taken at each view direction. The center portion of each photograph was projected onto a dot grid consisting of 200 dots. The number of dots laying on a vegetation component or a gap were tallied. The mean proportion of gap for each offnadir view direction was calculated for all sample points of each forest canopy. The bound on the errors of estimation of the proportion of gap (g) can be calculated as two times the square root of the estimated variance of g. The estimated variance is calculated as space sured and the tree species were recorded. The average height estimate was obtained for each sample plot Suunto Hypsometer. These data were used to estimate basal area per hectare, percent conifer/hardwood within the tract, average height, and number of stems per hectare. The results were reported in Table II. The hardwood tract includes lowland and upland sites, each with characteristic eastern U.S. hardwood species mixes. The lowland species growing on a poorly drained floodplain are predominatly red maple (Acer rubrum L) and black gum (Nyssa sylvatica Marsh). Other lowland species include swamp chestnut oak (Quercus prinus L.), American hornbeam (Carpinus caroliniana Walt.), green ash (Fraxinus pennsylvania var lanceolata (Berkh) Sarg.), holly (llex opaca Ait.), and willow oak (Quercus phellos L.). The well-drained areas adjacent to the floodplain support American beech (Fagus grandifolia Ehrh.) black gum, sweet gum (Liquidamber styraciflua L.), and tulip poplar (Liriodendron tulipifera L.). Other upland species include nothern red oak (Quercus rubra L.), white oak (Quercus alba L.), red maple, and Virginia pine (Pinus virginiana Mill.). The data in Table II describe fully stocked stands occupying over 95 percent of the growing Virginia pine % hardwood + canopy using a Lowland: red maple. blackgum Upland: american beech, tulip popular Stand Compositi on: by Basal Area t Conifer (a) (b) Fig. 2. Aerial photographs of the (a) pine and (b) hardwood forests in Beltsville, MD, on June 1984. Pine Tract IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. GE-24, NO. 2, MARCH 1986 284 1.0 TABLE III PROBABILITY OF GAP (p.ap(G)) THROUGH THE PINE AND HARDWOOD FOREST CANOPIES TO THE GROUND AS A FUNCTION OF OFF NADIR VIEW ANGLE (0) .8 c) OFF-NADIR ANGLE (8) 0° 300 15° 450 600 z 750 Pine Forest .25 .24 .23 .14 .096 .018 Hardwood Forest .21 .15 .14 .14 .082 .026 .6 0 .4 gq N-n (n-1) N where q is the estimated proportion of vegetation material, n is the number of dots sampled, and N is the population size. For this specific case (N - n)/N is considered to be 1.0 [16]. Thus, in this study the maximum bound on the error of estimation for any photograph would be +0.071 for a 50-percent probability of gap. The results are shown in Table III. Using a plumb-bobed protractor as described by Ranson et al. [17], the leaf inclination of 1000 leaves were measured on four isolated trees of the major hardwood species: red maple, black gum, American beech, and tulip popular. Each tree was under four meters in height and occurred on the edge of the forest clearing. The mean frequency distribution of leaf inclination is shown in Fig. 3. Azimuthal symmetry in leaf orientation was assumed in this study. Hemispherical leaf reflectance and transmittance in AVHRR bands 1 and 2 were sampled using a Beckman DK2 Spectroradiometer for the four major hardwood species and Virginia pine. The leaf optical properties for each species is presented in Table IV. The mean reflectance and transmittance values for the four hardwood species was 0.056 and 0.051, respectively, for band 1 and 0.46 and 0. 50, respectively, for band 2. These values were used in model simulations. C. Model Simulations The upgraded three-dimensional radiative transfer model of Kimes [6] was used to explore some unique radiative properties of forest canopies. The upgraded anisotropic soil algorithm was used in this study. Only a brief description of the model is presented here as a more detailed description may be found in [6], [7], [9]. The conceptual framework of the model is a retangular solid of any dimension that is subdivided into cubical cells of unit dimensions. Individual cells are identified by their x, y, and z coordinates (Fig. 4). Each cell is associated with information about the elements (e.g., leaves) within the cell. This information is used to define the manner in which the elements interact with radiation. For example, in the case of vegetation canopies, each cell has specific information about the elements constituting the scene (leaves, stems, reproductive structures, and soil), expressed as the element-area indexes, the angular distribution of the elements, their spatial dispersion, and optical properties. The information content of a cell can apply .2 .0 0° 15' 45' 300 75' 60' 90' LEAF INCLINATION ANGLE Fig. 3. Mean cumulative frequency distribution of leaf inclination distributions of the hardwood forest canopy (curve a), compared with the classical distributions of an erectophile (curve b), planophile (curve c), and spherical (curve d) canopies. Z y '--CELL (1, 1, 1) L:x Fig. 4. Three-dimensional framework of the model showing the rectangular cell matrix, cell coordinate system, diffuse solar sources (only one is illustrated), and the direct solar source. The diffuse and direct solar sources are extended down to the surface of each cell on the top surface of the cell matrix. TABLE IV HEMISPHERICAL REFLECTANCE AND TRANSMITTANCE OF LEAVES IN AVHRR BANDS 1 AND 2 FOR THE MAJOR SPECIES FOUND WITHIN THE HARDWOOD AND PINE FOREST CANOPIES. AVHRR BAND 2 - 1.1 pm) AVHRR BAND 1 (0.58 Red Maple - 0.68 vm) (0.73 Reflectance Transmittance Reflectance Transmittance 0.052 0.045 0.47 0.50 Black Gum 0.051 0.068 0.44 0.52 American Beech 0.058 0.051 0.43 0.50 Tulip Popular D.U64 0.061 0.U39 0.48 0.49 -- -- *Viryinia Pine *A mask of needles was used to measure reflectance. to any object, whether it is man-made or plant-canopy element. The spatial variation of cell contents among the cells determines the nature of the 3-D scene as shown in Fig. 5. KIMES et al.: DIRECTIONAL REFLECTANCE DISTRIBUTIONS 285 CORRESPONDING MODULE ENTIRE SCENE 44% CROWN COVERAGE 25% CROWN COVERAGE II gmP ROW CROP &M (B) (A) 79% CROWN 59% CROWN COVERAGE COVERAGE (D) (C) ORCHARD 4 TOTAL HOMOGENEOUS SCENE 2 (J) - FOREST 10 URBAN Fig. 5. Four scenes and their cooresponding modules. Within this framework, all radiant flux is transferred in defined by the divided into a number of contiguous sectors defined by an azimuth (q) and off-nadir (0) interval. The midvectors of all contiguous sectors in the 47r-sr region define the possible directions of radiant flux sources. These midvectors also define all possible orientations of leaf normals. The direction and the magnitude of the flux within each sector are defined by the midvector. The diffuse and direct solar sources are extended down to the target (Fig. 4) and then radiant transfer and scattering processes are simulated within and between the individual cells. The model has been recently upgraded by calculating a realistic anisotropic phase function of leaforientation distribution (both azimuth and zenith angle modes), leaf optical properties, and source direction [7]. (The general concept of a phase function is defined by Chandrasekar [18].) The phase function of a canopy is a very important component in controlling the scattering behavior of a vegetated surface and may be defined as the anistropic scattering that takes place at all interaction points throughout the canopy. The model may also sima finite number of discrete directions as user. The spherical coordinate system is Fig. 6. Nadir view of various forest simulation patterns. Six modules [8] having 9 x 9 cells each are shown for each pattern. Shaded areas represent cells containing tree vegetation in layers 2-6. The bottom-most layer of cells (layer 1) had homogeneous understory vegetation in all cells and is not shown in this representation. The patterns range from a sparse forest canopy (a) to a homogeneous canopy (e). ulate the non-Lambertian scattering by the soil [6] using an algorithm developed by Walthall et al. [19]. Multiple directional scattering between the cells is then simulated interactively until all the flux is absorbed, escaped from the canopy, or reaches a zero threshold. The model is unique in that it can predict 1) the hemispherical reflectance of the scene, 2) the directional spectral reflectance factors of a three-dimensional scene as a function of the sensor's azimuth and zenith angles and the sensor's position above the canopy, and 3) the directional spectral radiance as a function of the location of a sensor placed anywhere within the scene. The model was further upgraded by treating radiant transfers in zenith and azimuth intervals of 100 and 30°, respectively, for a total of 158 possible source directions in 4-r sr as opposed to only 74 source directions in past studies. A number of simulations were performed to try to show the unique radiative properties of forest canopies as opposed to homogeneous agricultural crops and natural grasslands as studied by Kimes [5]-[8]. Band 1 was treated in this study. Both dense and sparse forest canopies were simulated. The forest characteristics used for all simulations were those measured for the deciduous forest canopy in band 1 as reported above unless stated otherwise. The leaf reflectance and transmittance values were 0.056 and 0.051, respectively. The leaf inclination distribution is reported in Fig. 3. The soil reflectance was 0.20. Six layers were simulated, with the bottom layer (representing the understory-ground vegetation) always having an LAI of 1.5. Leaf densities were varied according to tree spacing. A number of tree crown (or clumps of tree crowns) spacings were simulated as shown in Fig. 6. Each tree crown or clump of trees is five layers high above the understory layer 1. In the nadir direction, the 286 IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. GE-24, NO. 2, MARCH 1986 (b) 180 90, 90 75 75 60 60v 45° 45' 30' 30" 15½ 15' canopy in the visible band for four different su 30- pi ae 30' 30' 45' 45 180 45 150 135' 75' W. 1235'75 7. 180 150 Directional percent reflectance distribution of the hardwood forest canopy in the visible band for four different sun angles. The system is described in Fig. 1. The solar position is shown as starred circle on each plot. plotting a small probability of gap through the canopy to the soil for simulation A and D was 0.21 and 0.096, respectively. RESULTS AND DISCUSSION Figs. 7-10 show the measured directional reflectance distributions of the hardwood and pine forest canopies in the visible and near-infrared (IR) bands, respectively. When comparing these distributions to the measured and simulated distributions of homogeneous agricultural crops and natural vegetation canopies reported by Kimes et al. [5]-[8], the reflectance distribution trends and dynamics are very similar. All of the same significant radiant transfers that take place in these homogeneous canopies take place in the forest canopies. These include 1) strong ansotropic scattering properties of the soil, 2) the geometric effect of the vegetation probability of gap function on the soil anisotropy, and 3) the anisotropic scattering of vegetation that is controlled by the phase function, leaf optical properties, and geometric "effect 1. " All of these are described by Kimes et al. [6], [7]. These phenomena account for the major scattering behavior of homogeneous vegetation canopies and apparently of very dense forest canopies such as measured in this study. The model simulations support this claim. For examIII. Fi.875' 90, 900 180, Fig. 30 452 51 75 90, 180" 15 Fig. 8. Directional percent reflectance distribution of the hardwood fo.rest canopy in the near infrared band. Symbols follow Fig. 7. ple, Fig. 11 shows a comparison between two vegetation canopies with the same total leaf area index (LAI = 4.2). Both canopies were identical, in that the canopy characteristics (leaf and soil optical properties, LAI, and leaforientation distribution) were the same as that measured for the hardwood canopy. The only difference between the two simulations is that one canopy was completely homogeneous (Fig. 6, simulation pattern E) and one had large regions of clumped vegetation (Fig. 6, simulation pattern D) that one might expect in dense forest canopies such as measured in this study. We can see that the two simulation patterns D and E, representing a dense deciduous forest canopy and the equivalent homogeneous canopy, respectively, are very similar in their reflectance distributions. Fig. 12 shows a quantitative comparison between the reflectance in the principal plane of the sun for the forest canopy simulation and the equivalent homogeneous case. For most purposes there is an insignificant difference between the dense forest canopy and the equivalent homogeneous scene. The small difference that does occur at each sun angle (Fig. 10) is due to the clumping of vegetation in individual tree crowns or groupings of tree crowns in forest canopies. This clumping has two effects on the radiant transfers within the canopy: A) it increases the probability of gap to the understory and/or soil layers, 287 KIMES et al.: DIRECTIONAL REFLECTANCE DISTRIBUTIONS 90 / 75 135 755 / 135s 90, 90, 1s0, 180 can- Fig. 10. Directional percent reflectance distributions of the pine forest canopy in the near infrared band. Symols follow Fig. 7. which increases the influence of these lower layer's scattering properties, and B) it increases the number of low transmitting clumps of vegetation within the scene causing increased backscatter and decreased forward scatter to occur relative to the homogeneous case. These two effects are terrned phenomenon A and B, respectively, throughout the paper. The physics of phenomenon B is similar to the physics involved in the increased backscatter and decreased forward scatter observed in organic soils and wintering deciduous forest canopies with no leaves. These canopies have opaque vertical components that cause large azimuthal variations in scattering. For example, soils have vertical components that have very low transmittance, and thus dark shadowing of scene components occurs. In the antisolar direction (backscatter toward the sun) only those surfaces that are in direct sunlight are viewed by the sensor, and thus the reflectance is maximum in this direction. As the sensor direction moves away from the antisolar direction, the following two mechanisms cause the reflectance to decrease. 1) In the sensor's field of view, the relative proportion of shadowed surfaces increases. 2) In the sensor's field of view, the proportion of particle facets with normals that deviate from the solar direction increase, causing decreased solar irradiance on these facets (cosine function). Thus, organic soils exhibit strong back- scatter and weak forward scatter. These trends are discussed in more detail and supported by data in [20]-[22]. The physics of phenomenon B are similar to soil canopies in that the clumping of vegetation into individual tree crowns (or contiguous tree crown groupings) create vertical structures that have relatively low transmission causing strong backscatter and weak forward scatter. The magnitude of phenomenon B is not as dramatic in forest canopies as it is in soils, however, since the vertical clumps of vegetation are not opaque as in the case of soil particles. Phenomena A and B become important in sparse forest canopies as discussed in detail later. The measured data of the hardwood canopy in band 1 (Fig. 7) compares relatively well in trend and magnitude with the simulated pattern D (Fig. 11) if one compares near equal solar zenith angles. However, the reflectance in the backscattered direction (00 azimuth-away from the sun) does not increase as rapidly in the simulated data as compared to the measured data. For these canopies the LAI is sufficiently high that the anisotropic soil has little effect on the reflectance distribution above the canopy. Canopies ranging between erectophiles and spherical leaforientation distributions with leaf reflectance approximately equal to leaf transmittance have the minimum reflectance more toward nadir at all sun angles, and a greater increase in reflectance with increasing off-nadir view an- Fig. 9. Directional percent reflectance distribution of the pine forest opy in the visible band. Symbols follow Fig. 7. IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. GE-24, NO. 2, MARCH 1986 288 SIMULATION PATTERN D (a) (c) (b) 90' s90 90' L 180' SIMULATION PATTERN E (d) (e) 9W W (f) 80' 70' 60' 50o 4W 30' 20' 10' 0' 10 20" 30' 40 50 ( 60 70 80 90 180 L._. 190 90 180 Fig. 11. Comparison of reflectance distributions in the visible band between two vegetation canopies with the same total LAI (4.2). For each canopy, three plots with different sun angles are shown. Both canopies were identical in that the canopy characteristics (leaf and soil optical properties, LAI, and leaf-orientation distribution) were the same as that measured for the hardwood canopy. The only difference between the two simulations is that one canopy was completely homogeneous (simulation pattern E, Fig. 6) and one had large regions of clumped vegetation (simulation pattern D, Fig. 6), which might be expected in dense forest canopies. gle for all azimuth view directions and all sun angles relative to a more planophile canopy. A planophile canopy has a minimum reflectance region that is shifted further away from nadir in the forward scatter direction (00 azimuth direction) relative to erectophile-spherical canopies [7]. The data and findings of Kimes [7] suggest that the hardwood forest canopy has a more spherical-erectophile average leaf-orientation distribution rather than the near planophile distribution measured on a few isolated trees under 4 m in this study. In fact, Hutchinson et al. [23] 289 KIMES et al.: DIRECTIONAL REFLECTANCE DISTRIBUTIONS 12 ta) " ) U) 11 10 w~~~~~~~~~~~~~~~~~~~~5 z .0 W -c 1 0, 1S0 ° 180, C. z Fig. 13. Directional percent reflectance distribution of a simulated sparse forest using simulation pattern A in the visible band. Total LAI was 2.3. The bottom most layer (#1) had an LAI of 1.5 representing understory vegetation and the area containing tree material had an LAI of 3.4. All other canopy characteristics were the same as those measured for the hardwood forest. 20° 400 -80° -600 -400 -200 -o0 OFF NADIR VIEW ANGLE IN PRINCIPAL PLANE OF SUN 600 80° Fig. 12. Directional percent reflectance in the principal plane of the sun for simulation pattern D with the hardwood canopy characteristics and the equivalent homogeneous case in the visibie band. Off-nadir view angles with negative values represent the backscatter direction and positive values the forward scatter direction. have shown leaf-orientation distribution of their east Tennessee deciduous forest canopy to be plagiophile, which is a distribution between planophile and erectophile. In summary, analyses of the simulated and measured data suggest that the same major physicai phenomena operating in agricultural crops and natural vegetation communities operate in dense forest canopies. The clumping of vegetation in tree crowns of very dense forest canopies as studied here, is insignificant in terms of directional reflectance when the field of view covers several tree crowns. As the density of the forest canopy decreases one would expect major differences in the scattering behavior of forest canopies relative to homogeneous agricultural and nat- ural vegetation communities. As the density of the canopy decreases, large openings begin to occur between individual tree crowns and/or clumps of tree crowns. With these changes in the geometrical structure, one would expect to see significant changes in the scattering behavior of the canopy. This change in scattering behavior was simulated by the model simulations. For example, Fig. 13 shows simulated data of a very sparse forest canopy with large openings between individual tree crowns. Simulation pattern A (Fig. 6) was used with a total LAI of 2.3. The bottom layer had an LAI of 1.5 representing understory vegetation and the area (cells) containing tree material had an LAI of 3.4. The leaf-orientation distribution, leaf and soil optical properties were the same as those measured for the hardwood forest. Fig. 13 shows an increased backscattering and decreased forward scatter component relative to the dense forest canopy (Fig.- 11, simulation pattern D). At first glance one would expect this difference to be due to the large natural openings between the tree crownsphenomenon B as discussed previously. However, further analysis shows the following. Fig. 14(a) shows the reflectance in the principal plane of the sun of various simulations for a solar zenith angle of 20°. The figure clearly shows strong backscatter and weak forward scatter of the sparse forest canopy (simulation pattern A) relative to the dense forest canopy (simulation pattern D, Fig. 11(a)). The leaf area (LAI = 3.4) of the tree crowns in simulation pattern A was redistributed evenly in all cells (layers 2-6, layer 1 in all cases has a homogeneous LAI of 1.5) to simulate the equivalent homogeneous canopy (Fig. 14(a)). It is clear that the "vegetation clumping" that occurs in sparse forest canopies causes a significant increase in backscatter relative to forward scatter. To explore the cause of this effect we turned the soil scattering off by making the soil black (absorption of 1.0). Comparing simulation A with the equivalent homogeneous case (both cases with black soil, Fig. 14(a)), we see that the "clumping of vegetation" causes a modest increase in vegetation backscatter relative to forward scatter. Thus, phenomenon B is responsible for a small portion of the increased backscatter in simulation A. The remaining increase in backscatter in simulation A as compared to the homogeneous case is due to the increase in probability of gap to the soil as a result of vegetation clumping (referred to as phenomenon A previously). For example, the probability of gap throughout the canopy to the soil at 200 is 0.21. for simulation A and 0.13 for the equivalent homogeneous case. This causes an increase in backscatter with a peak at the hot spot (-20° off nadir view angle) due to the increase in directly viewed, highly reflective, and directly illuminated soil. Thus, in these particular simulations, the clumping of vegetation in sparse forest canopies significantly alters the probability of gap function that in turn permits the scattering properties of the substrate to be expressed to a larger degree. Relative to these changes, the change in the scattering properties of the vegetation was secondary. It is interesting to note that the peak reflectance of sim- IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. GE-24, NO. 2, MARCH 1986 290 U z a: CD z J C-) r UJ UJ LL z v z C-) a. -80° -600 -400 -200 40° 200 00 600 2.0 800 -80° -600 -400 -200 00 200 4Q0 600 80° OFF-NADIR VIEW ANGLE IN PRINCIPAL PLANE OF THE SUN OFF-NADIR VIEW ANGLE IN PRINCIPAL PLANE OF THE SUN (b) (a) 9.0 8.0 7.0 z J aK: 6.0 LL cr: z C-) 5.0 4.0 3.0 2.0 -800 -600 -400 -200 00 200 400 600 80° OFF NADIR VIEW ANGLE IN PRINCIPAL PLANE OF THE SUN (c) Fig. 14. Directional percent reflectance in the principal plane of the sun for simulation pattern A and the equivalent homogeneous case for a solar zenith angle of (a) 200, (b) 50°, and (c) 80° in the visible band. Simulation pattern A characteristics are described in Fig. 13. The figure also shows simulations where the soil was made black (absorption of 1.0) for both simulation pattern A and the equivalent homogeneous case. Finally simulation A is shown with Lambertian soil. Fig. 14(b) shows the analysis of simulation A and the ulation A with black soil occurs at -40° (Fig. 14(a)). The position of this peak is due to a balancing between the equivalent homogeneous case for a solar zenith angle of geometrical effect 1 and the vegetation phase function of 50. The same principles as discussed for the 200 solar zenith simulation (Fig. 14(a)) apply for the 500 solar zethe canopy as discussed by Kimes et al. [6], [7]. 291 KIMES et al.: DIRECTIONAL REFLECTANCE DISTRIBUTIONS nith angle simulation. Fig. 14(c) shows the 80° solar zenith angle simulation. Because of the low sun angle the soil has little influence on the scattering behavior of the canopy. The canopy scattering behavior for all simulations in Fig. 14(c) is characteristic of any moderately dense, homogeneous canopy at a large solar zenith angle-minimum reflectance near nadir and large increase in reflectance with increasing off-nadir view angle for all azimuth view directions. The effect that vegetation clumping of sparse forest canopies has on the scattering behavior of the canopy (phenomenon A and B) is greatly diminished at large solar zenith angles. Fig. 14(a) and (b) also shows simulation A with Lambertian soil. The comparison between the Lambertian and non-Lambertian soil cases shows the importance of the non-Lambertian soil reflectance function in influencing the reflectance of the canopy as a whole. The non-Lambertian reflectance function of the soil becomes more non-Lambertian as the solar zenith angle increases [2], [8], [201[231. However, as the solar zenith angle increases, the probability of gap to the soil decreases causing a decrease in the contribution of scattered flux from the soil to the sensor. One would expect that as the vegetation density of the understory increased, the probability of gap to the soil would become very small; phenomenon A would become insignificant and only phenomenon B would be expressed. Furthermore, as the leaf density increases in individual tree crowns, the magnitude of phenomenon B would increase. These trends are shown in Fig. 15 for a solar zenith angle of 50°. The simulations were the same as simulation A discussed above except that layer 1 LAI was increased to 4.0 and the LAI within individual tree crowns was increased to 10.0 (one-sided projection), which means an LAI of 2.0 for individual cells within the tree crown. The high LAI in the individual tree crowns would be typical for sparse conifer stands. In general, conifer stands have a much higher leaf area index as compared to broadleaved forests. Tadaki [24] reports that a reasonable range of LAI for evergreen forests is 15 to 20 and for deciduous broad-leaved forest 4 to 6 where the leaf area is reported on a one-sided basis for broad-leaved species and on allsides basis for needle-leaved species. The leaf-orientation distribution as well as the leaf reflectance and transmittance values measured in this study on the hardwood canopy are close to those measured for lodgepole pine by Kimes et al. [25], [26]. There is evidence presented by Kimes et al. [25] that this leaf distribution may be characteristic of a large class of needle bearing species. So Fig. 15 is reasonable for a sparse conifer canopy with a dense understory. The probability of gap to the soil is small and, thus, the scattering properties of the soil are insignificant in both simulation pattern A and the equivalent homogeneous case. Fig. 15 shows that by making the soil black there is essentially no change in the scattering behavior of the canopy. However, a relatively large increase in backscatter relative to forward scatter by the vegetation itself due to the clumping of vegetation into 5.0 4.5 C-J 4.0 z H C-C 3.5 *L. z H UL 3.0 2.5 2.0 200 400 oo -800 -600 -400 -200 OFF-NADIR VIEW ANGLE 600 800 IN PRINCIPAL PLANE OF THE SUN Fig. 15. Directional percent reflectance in the principal plane of the sun for simulation pattern A and the equivalent homogeneous case for a solar zenith angle of 500 in the visible band. Canopy characteristics as described in Fig. 13 were used except that layer I LAI was increased to 4.0 and the LAI within individual tree crowns was increased to 10.0. This canopy is typical for a sparse conifer canopy with a dense under- story. tree crowns is apparent by comparing simulation pattern A with the equivalent homogeneous case. Thus, in such canopies phenomenon A is insignificant and phenomenon B becomes very significant. Fig. 16 shows how phenomenon A and B combined cause a significant increase in backscatter relative to forward scatter with decreasing forest density. The total LAI for each simulation pattern is reported in Fig. 16. In all cases, the leaf-orientation distribution used were those measured for the hardwood canopy. Furthermore, in all simulation cases the LAI in layer 1 was 1.5-representing a homogeneous understory-and the LAI in each tree crown was always 3.4. Fig. 16 clearly shows the progressive increase in backscatter relative to forward scatter as a result of increased vegetation clumping and de- creased total leaf density. IV. CONCLUSIONS AND IMPLICATIONS Within the remote sensing community the authors have heard several researchers hypothesize that forest canopies may behave very differently from agricultural crops because of their unique structure-tall canopy height and open spacings between the upper crowns of individual trees. Measurements and model simulations in this study show that the directional scattering behavior of relatively dense forest canopies is very similar to the directional scattering behavior of agricultural crops and natural grasslands. The most significant physical phenomena involved in these dense canopies where the soil contribution is minimal is the anisotropic scattering of vegetation, 292 IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. GE-24, NO. 2, MARCH 1986 5.0 the probability of gap to the soil is small. In such cases, only phenomenon B is responsible for the increased backscatter relative to the forward scatter when compared with the equivalent homogeneous canopy. In this study, the peak backscatter was increased due to phenomenon B as much as 35 percent relative to the equivalent homogeneous case. The magnitude of phenomenon B increases as the leaf density in the tree crowns increases. The magnitude of this phenomenon would be generally larger in conifer forests as opposed to broad-leaved forests since individual conifer trees generally have a higher leaf density. In sparse forest canopies the combined effect of phenomenon A and B can typically cause a 65-percent or higher increase in the peak backscatter relative to the equivalent homogeneous canopy. It is hypothesized that these phenomena are important in sparse conifer forests, such as the boreal forests, which account for 50 percent of the LU H. 2.) cr 3.0 IN P Fig.16.Dirctinalperentrefetneih 2.5 W~~~~~~F-AI VIE ANL world's forest area. 2 0 -800 -600 -400 -200 00 200 400 600 800 OFF-NADIR VIEW ANGLE IN PRINCIPAL PLANE OF THE SUN Fig. 16. Directional percent reflectance in the principal plane of the sun for simulation patterns A, B, C, and D. The visible band is reported for a 200 solar zenith angle. The total LAI and percent crown cover for simulation patterns A, B, C, D were 2.3, 3.0, 3.5, 4.2, and 25, 44, 59, 79 percent, respectively. The leaf orientation distribution were the same as measured for the harwood forest. In all simulations, the LAI in the lower layer was 1.5-representing a homogeneous understory-and the LAI in each tree crown was always 3.4. which is controlled by the phase function of the leaf orientation distribution, the leaf optical properties, and the geometric effect 1 as discussed by Kimes et al. [6], [7]. Only in more sparse forest canopies with significant spacing between the tree crowns (or clumps of tree crowns) does the general scattering behavior deviate from homogeneous agricultural and natural grassland canopies. This clumping has two effects on the radiant transfers within the canopy: A) it increases the probability of gap to the understory and/or soil layers causing an increase in the influence of the scattering properties of these lower layers and B) it increases the number of low transmitting clumps of vegetation within the scene causing increased backscatter and decreased forward scatter to occur relative to the homogeneous case. In sparse forest canopies, phenomenon A is clearly dominant in forest canopies where the clumping of tree crowns is such that a large probability of gap occurs to substrate layers (e.g., soil, litter, snow, or understory vegetation cover) that have scattering properties significantly different from forest vegetation. In the case where the substrate is soil (as presented in this study), the backscatter is greatly increased in relation to the forward scatter. In this study, the peak backscatter was increased as much as 30 percent relative to the equivalent homogeneous canopy case. Phenomenon B adds to this effect by further increasing the backscattering relative to the forward scatter. Phenomenon A becomes insignificant when the understory vegetation has similar scattering properties as the forest vegetation and is significantly dense so that REFERENCES [1] J. A. Smith and K. J. Ranson, "Bidirectional reflectance studies literature review," NASA/GSFC, prepared by ORI, Inc., Silver Spring, MD 20910, 1979. [2] J. A. Smith, "Matter-energy interaction in the optical region," in ASP Manual of Remote Sensing, 2nd ed., ch. 3, 1983. [3] D. L. Williams, "Characterization of key remote sensing variables associated with a forest under stress due to acid rain deposition," Task 3 of NASA RTOP, NASA Headquarters, Washington, DC, 1984. [4] K. T. Kriebel, "Measured spectral bidirectional reflection properties of four vegetated surfaces," Appl. Opt., vol. 17, pp. 253-259, 1978. [5] D. S. Kimes, "Dynamics of directional reflectance factor distributions for vegetation canopies," Appl. Opt., vol. 22, pp. 1364-1372, 1983. 16] D. S. Kimes, J. M. Norman, and C. L. Walthall, "Modeling the radiant transfers of sparse vegetation canopies,'" IEEE Trans. Geosci. Remote Sensing, vol. GE-23, no. 5, pp. 695-704, Sept. 1985. [71 D. S. Kimes, "Modeling the directional reflectance from complete homogeneous vegetation canopies with various leaf-orientation distributions," J. Opt. Soc. Amer., vol. 1, pp. 725-737, 1984. [8] D. S. Kimes, W. W. Newcomb, C. J. Tucker, I. S. Zonneveld, W. van Wijngaarden, J. de Leeuw, and G. F. Epema, "Directional reflectance factor distributions for cover types of nothern africa," Remote Sensing Environ., 1985. [9] D. S. Kimes and J. A. Kirchner, "Radiative transfer model for heterogeneous 3-D scenes," Appl. Opt., vol. 21, pp. 4119-4129, 1982. [10] A. H. Strahler and X. Li, "An invertible coniferous forest canopy reflectance model," in Proc. 15th Int. Symp. Remote Sensing Environ. (Ann Arbor, MI), 1981. [11] R. D. Jackson, R. J. Reginato, P. J. Pinter, Jr., and S. B. Idso, "Plant canopy information extraction from composite scene reflectance of row crops," Appl. Opt., vol. 18, pp. 3775-3782, 1979. [12] J. M. Norman and J. M. Welles, "Radiative transfer in an array of canopies," Agron. J., vol. 74, pp. 481-488, 1983. [13] D. S. Kimes and J. A. Kirchner, "Irradiance measurement errors due to the assumption of a Lambertian reference panel," Remote Sensing Environ., vol. 12, pp. 141-149, 1982. [14] J. V. Dave, "Extensive datasets of the diffuse radiation in realistic atmospheric models with aerosols and common absorbing gases," Sol. Energy, vol. 21, pp. 361-369, 1978. [15] s. F. Gringrich, "Criteria for measuring stocking in forest stands," Proc. Soc. Amer. Foresters, vol. 23, pp. 198-201, 1964. [16] W. Mendenhall, L. Ott, and R. L. Scheaffer, Elementary Survey Sampling. Belmont, CA: Wadsworth Publishing, 1971. [17] K. J. Ranson, V. C. Vanderbilt, L. L. Biehl, B. F. Robinson, and M. E. Bauer, "Soybean canopy reflectance as a function of view and illumination geometry," in Proc. 15th Int. Symp. Remote Environ. (Ann Arbor, MI), May 1981. [18] S. Chandrasekhar, Radiative Transfer. New York: Dover, 1960. [19] C. L. Walthall, J. M. Norman, J. M. Welles, G. C. Cambell, and 293 KIMES et al.: DIRECTIONAL REFLECTANCE DISTRIBUTIONS [19] C. L. Walthall, J. M. Norman, J. M. Welles, G. C. Cambell, and B. L. Blad, "A simple equation to approximate bidirectional reflectance from vegetative canopies and bare soil surfaces," Appl. Opt., vol. 24, pp. 383-387, 1985. [20] K. L. Coulson, "Effects of reflection properties of natural surfaces in aerial reconnaissance," Appl. Opt., vol. 5, pp. 905-917, 1966. [211 F. D. Eaton and I. Dirmhirn, "Reflected irradiance indicatrices of natural surfaces and their effect on albedo," Appl. Opt., vol. 18, pp. 994-1008,1979. [22] G. H. Suits, "The cause of azimuthal variations in directional reflectance of vegetative canopies," Remote Sensing Environ., vol. 2, pp. 175-182, 1972. [23] B. A. Hutchinson, D. R. Matt, R. T. McMillen, L. J. Gross, S. T. Tajchman, and J. M. Norman, "The architecture of an east Tennessee deciduous forest canopy," J. Appl. Eco., 1985. [24] Y. Tadaki, "Some discussions on the leaf biomass of forest stands and trees," Bull. Gov. Exp. Stn., vol. 184, pp. 135-161, Tokyo, Japan, 1966. [25] D. S. Kimes, J. A. Smith, and J. K. Berry, "Extension of the optical diffraction analysis technique for estimating forest canopy geometry," Aust. J. Bot., vol. 27, pp. 575-588, 1979. [26] D. S. Kimes and J. A. Smith, "Simulations of solar radiation absorption in vegetation canopies," Appi. Opt., vol. 19, pp. 2801-2811, 1980. * W. Wayne Newcomb received the B.S. degree in agronomy from the University of Maryland in 1974. Since 1978, he has been a data analyst for RMS Technologies at the NASA Goddard Space Flight Center, primarily working with NASA scientists evaluating Thematic Mapper Simulator data for agricultural use. He also has worked with AVHRR data on the GIMMS Project to monitor African grasslands. K Ross F. Nelson received the B.S. degree in forest j management from the University of Maine, Orono, in 1974, and the M.S. degree in forestry/ remote sensing from Purdue University, West Lafayette, IN, in 1979. Since 1979, he has been a Physical Scientist in m, the Earth Resources Branch at the NASA Goddard Space Flight Center, involved with the use of digital remotely sensed data for assessing the forest canopy. He has worked with Landsat MSS data to evaluate gypsy moth damage, Thematic Mapper Simulator data to evaluate spruce budworm damage, and AVHRR and MSS data to assess deforestation in South America. He has worked with geobotanical scientists to determine the effects of mineralization on forest canopy condition, and is investigating the utility of laser-induced fluorescence for tree species identification. R= * Daniel S. Kimes was born in Columbus, OH. He received the B.S. degree in wildlife biology from Colorado State University, Fort Collins, in 1975, the M.S. degree in remote sensing from the University of Michigan, Ann Arbor, in 1976, and the Ph.D. degree in earth resources from Colorado State University, in 1979. Since 1979, he has been employed at the NASA Goddard Space Flight Center, Earth Resources Branch, Greenbelt, MD, where he has been engaged in mathematical modeling of visible, near infrared, and thermal infrared radiation interactions with vegetation. , _~ John B. Schutt was born in New Haven, CT. He received the B.S. degree in chemistry for the Massachusetts Institute of Technology, Cambridge, in 1952, M.S. degrees in physical chemistry and chemical engineering in 1956, and the Ph.D. degree in chemical engineering from the University of Rochester, Rochester, NY, in 1958. Since 1973, he has been employed at NASA Goddard Space Flight Center, Earth Resources Branch, Greenbelt, MD, where he has been engaged in studying the dynamic behavior of vegetation and its radiometric manifestations.