Bidirectional Radiance Data Processing for Forest Canopy BRDF Model Studies

Bidirectional Radiance Data Processing for Forest Canopy BRDF Model Studies

Wang Jindi,Li Xiaowern
Institute of Remote Sensing Application,
Chinese Academy of Sciences, Beijing 100101, P.R. China

Alan H. Strahler
Certer for Remote Sensing , Boston University, Boston, MA 02215, USA

Abstract
The directional anisotropy of solar radiance reflected from terrestrial surfaces is by now almost as well recognized as spectral variation. Off-nadir measurements of a surface are becoming possible with the development of new satellite sensors ( e.g. HRV, MISR ) and airborne sensors ) e.g. ASAS). This paper presents an the study of modeling the Bidirectional Reflectance Distribution Function ( BRDF) of forest canopy.

With the Geometric – Optical BRDF model of Li and Strahler, biological and structural parameters of forest canopy can be deduced from the multiangle remote sensing data. The ASAS data were obtained from forest remote sensing experiment field in Maine, USA. The ASAS data set consists of 29 spectral bands and 14 viewing zenith angles for the same target. The field measurement data involve the structure parameters of crowns which were obtained at the same time. The data processing includes: 1) Data compression and color image displays for 14 different viewing angles. 2 ) Using the K – L Transformation, the first three principal components of 29 bands were extracted. By statistical analysis, the mean value and variance of the first three principal components on the BRDF principal plane were calculated and plotted. These plots show the “Hotspot” and “Bowl-Shape”, which can be explained by the overlaping function and the mutual shadowing of crowns in the Geometric – Optical model. 3) The data were processed to obtain the image sebset for which synchronous field measurement data were obtained. With the statistics data of these subset image and field measurement data, the model can be verified by running the Monte Carlo simulation program of Geometric-Optical model. As a result, the Geometric – Optical model fits measurement data very well.

Introduction
For remote sensing data application, many kinds of mathematics model are used to describe the relationship between multiangle and multispectrum remote sensing data and the feature of objects, and to estimate the geometric structure and biological parameters of objects.

The bidirectional reflectance distribution function (BRDF) describes the rate of change of the directional radiance of the surface as a function of illuminination zenith angle and azimuth, and the downwelling irradiance. BRDF models are used in the direct mode to simulate reflection and absorption of solar radiation as a function of canopy parameters, and in the inverse mode in order to estimate canopy parameters from measured reflectance data. In the Li-Strahler geometric – optical canopy model, the forest is treated as a collected on regular geometric shapes that cast shadows on a background and are viewed and illuminated from different directions. The directional radiance of the forest is a function of the sizes, shapes, orientations and placement of the objects ( i.e. individual tree crowns ) within the scene [1]. The model has been tested in woodland using Landsat Multispectral Scanner (MSS), Thematic Mapper (TM) , and SPOT High Resolution Visible (HRV) imagery [2]

The ASAS is an airborne, off-nadir-pointing imaging spectroradiometer used to acquire bidirectional radiance data for terrestrial targets. ASAS will permit measurement of radiance at a sufficient variety of angle to estimate the BRDF. ASAS also is being used to develop understanding of fine scale differences between directional reflectance and hemispherical reflectance

[3] As an example the ASAS data demonstrate combined effects of reflectance anisotropy and increased atmospheric path length on off-nadir observation, the result o these effects is a variation in vegetation indices as a function of view direction [4].

The objective of the study was to estimate the BRDF on principal plane from the processed ASAS images, and to test the geometric – optical canopy model.

ASAS data processing
ASAS employs detector array technology to acquire digital image data for visible and near – infrared spectral bands with a spectral resolution of 15nm. The spectrum range is 455nm—873nm. ASAS is able to track and image a target site through a sequence of fore –to-aft view angle which is up to 45 degree either fore or aft of the nadir view. And its spatial resolution at nadir was approximately 7.0 meter in the along-track, and 5.0 meter in the cross-track when the height of flight is 5,000 meter.

The ASAS data processed in this study were obtained from forest remote sensing experiment field in Maine, USA on Sept. 8, 1990. When the data acquiring, the solzr zenith angle is near +45 degree. The data consist of 7 images with different viewing direction on the principal plane and other 7 images on the cross plane. The viewing angle in each plan are from – 45 degree to + 45 degree with the incremence 15 degree. The original data were recorded in 12 bits, And after radiometric calibration the data were stored as a 16 bits interge in digital image files.

1 Data compression and color image display
In order to know the quality of the total 29 bands images, the 16 bits original data were compressed to 8 bits. The compression ratio depends on the statistical parameters of each band. Then the data can be displayed. By the display images, most bands of data are of high ratio of signal to noise and can be used to estimate the reflectance properties of the target.

Furthermore we take three bands of every image as a color combination image to display. The band No. 8(560nm), band No. 15(650nm) and band No.20(720nm) are coded as blue, green and red respectively to make a color combination image. In order to give a better color image and remain their different reflectance as well, we use the same compression ratio and enhancement method to process every color image. So the relative variation of brightness in these images can e seen clearly. When the viewing angle is equal to +45 degree, the image looks bright. As the viewing direction approach to the nadir, images look clearly and the average brightness of image decreases. The viewing zenith angle is equal to – 45 degree. These color images show that the reflectance of object is changing with the reflectance of object is changing with the viewing direction.

2 Statistics of ASAS data
For every image, in order to use all information of 29 bands the principal component transformation (KL-Trans.) was used to process every scene image. First, to calculate the maximum value, minimum value, mean value, standard deviation, eigen vector and eigen value for every band, and to transform every 29 bands data into its PC plane using its own eigen values and eigen vectors independent with the original 29 bands, and include 99 percent information of total 29 bands. Second, to group PC-1 values of each image into 30 intervals from its mean PC-1 value minus twice of its standard deviation to its mean value plus twice of its standard deviation to its mean value plus twice of its standard deviation. The two ending groups include all of others which have PC-1 valie beyond the range. Then the PC-2 mean value is plotted versus PC-1 mean value for each group, with an error bar showing standard deviation of PC-2 in that group. Third, to put the processing result of every image into the same (PC-1, PC-2) plane. The relative position of different image in the new ) pc-1, pc-2) plane can be calculated using its original mean value. Then they become comparable each other in the plane. Fig. 1. shows the scattering of the surface in principal plane. It is obvious that Hotspot and Bowl-shape of the forest directional reflectance on principal plane is similar to the geomertric – opitcal model described.

Fig. 1. Scatter of different viewing images on principal plane
Gemoetric-optical canopy model verification
As preparing for the model test, some forest structural parameters measured in the field wee made in order Based on these images after KL-trans., subset image with polygon boundary was taken form every whole scene image in where some field measuring had been done. Then mean value and standard deviation of every subset image was calculated. The results are plotted in Fig. 2.,

Fig. 2. Statistical parameters of subset images.
Where the curve show the variation of mean value and error bar show standard deviation.

When the ASAS data were obtained, some field measurement works were doing at the forest remote sensing experiment in Maine of USA. 228 trees were measured. The measurement data consist of height of trees and size of crowns for each tree. To make these data in order, we obtained some average forest structural parameters which is shown in table 1.

Table 1. Forest Canopy Structural Parameters

Mean height of tree:	19.232(m.)
Mean height of crown:	11.054(m.)
Mean diameter of crown:	4.287 (m.)

An Monte Carlo Simulation program of Geometric- Optical model was developed. The program can be run at SUN-station and IBM-PC microcomputer. Running the program, we entered the forest canopy structural parameters shown in Table1. , and the coverage was 59.1%, the reflectance values for scene components in Near-Infrared were 0.2 for illuminated ground, 0.55 for illuminated crown, 0.05 for shaded ground and crown. The program output the forest canopy BRDF calculated with geometric – optical model and measurement data. Comparing Fig. 2 with fig. 3., we get some conclusion in the model test.

In Li-Strahler canopy model, the geometric – optical effects of crown shape and mutual shadowing play an important role. By these effects, the shape of hotspot in BRDF of forest canopies is better related to the canopies structure, and the bowl-shape in it’s BRDF is explained too.

At he hotspot, when illumination and viewing positions coincide, shadows are hidden behind crowns and the scene appears bright. As the viewing position diverges from that of illumination, the shadows behind the crowns are progressively revealed and the scene darkens. This divergence depends on canopy structure [5] as Fig. 3 shown. IN Fig. 2 we can see the feature of the forest directional reflectance taken from remote sensing data (ASAS), though the

Data only cover the principle plane of BRDF surface. The hotspot is near+45 degree viewing zenith angle, the mean value and standard deviation get the maximum in here. In color image, we can see the most of shadows are hidden behind the crowns, so the scene looks bright

Forest BRDFs as observed from real data often show a shape rise at the edge of 3-D display surface, especially at large zenith angle opposite to the solar position, called “bowl-shape” of BRDF. Fig. 1 and Fig. 2 all shows this behavior at-45 degree viewing zenith. This is first explained by a simple assumption that all illuminated shadows will preferentially occupy the lower surface of crowns, These shadow will be more likely obscured at large view zenith since adjacent crowns will also tend to obscure the lower portions of other crowns or ground. We can see the bowl-shape in Fig.3, ( but it seems not as well as in Fig. 2) Further the BRDF model was modified, the spatial correlation of such surface is taken into account.

Conclusion
Our work shows that the processing method for ASAS data is usable to describe the directional reflectance of object, and is usable to the BRDF mathematics model test Within the data covered area, the geometric – optical model fits the measurement data very well.

Reference

X. Li and A. H. Strahler, “Geometric – optical bidirectional reflectance modeling of a coniferous forest canopy,” IEEE Trnas. Geoscience and Remote Sensing, Vol. GE-24, pp 281 – 293, 1986.
J. Frankilin and D.L. Turner, “The Application of a Geometric Optical Canopy Reflectance Model to Semiarid Shrub Vegetation,” IEEE Trans. Geoscience and Remote Sensing , Vol. 30, No.2, Mar. 1992.
G.H. Asrar,”Theory and Application o Optical Remote Sensing ,” A Wiley-interscience publication, John Wiley anf Sons, 1989.
J.R. Iron and etc., “An off-nadir-pointing imaging spectroradiometer for terrestrial ecosystem studies,” IEEE Trans, on Geosciences and Remote Sensing, Vol. 29, No. 1 , Jan 1991.
X. Li and A. H Strahler, “Geometric – Optical bidirectional reflectance modeling of the discrete crown vegetation canopy: effect of crown shape and mutual shadowing, “IEEE Trans. Geosciences and Remote Sensing, Vol. 30, No.2., Mar. 1992.