The objectives of the Earth Observation Centre (EOC) High-Resolution Scene Brightness, or BRDF Task, were to develop guidelines for best practice outputs for the minimisation of scene brightness in high-resolution imagery such as airphotography, video, digital camera and airborne scanner data. The work reported here was principally undertaken by the CSIRO Exploration and Mining and the detailed publication on the Referencing Method, (Ong et al. 2000) is available on the web page (http://www.per.dem.csiro.au/research/MMTG/mrrp/mrrp.html). The 4 main areas of activities were:

1. Develop software for accurate registration/rectification of frames, such that fitted models are constrained and corresponding pixels in areas of scene overlap have the same geo-location. The sub-tasks were:

• Evaluate photogrammetric and image processing tool-kits for their suitability for different data types.
• Evaluate the use of auto-correlation software for the automated / semi-automated selection of ground control points.
• Produce prototype software for accurate registration / rectification of frames.

1. Compare the relative merits of all available kernel approaches on a range of data and land surface typologies, and enhance the methodology to include the constraints that the differences between brightness-corrected reflectances, in areas of scene overlap, are minimised. The sub-tasks were:

• To establish a practical current approach to BRDF modelling and scene brightness correction, including determining whether a "typology" of kernels exists.
• Develop prototype software for fitting kernel, empirical and semi-empirical BRDF functions (forward mapping) and correcting scenes using the fitted models (inverse mapping).

1. Enhance the referencing correlative technique to include the constraint that differences between brightness-corrected reflectances in areas of scene overlap are minimised,and evaluate the technique for specific applications. The sub-tasks were:

• To enhance the referencing method to include the constraint that differences between brightness-corrected reflectances in areas of scene overlap are minimised
• Model and extract brightness surfaces for terrestrial mixtures and over water surfaces (sun-glinted) for a range of data types
• Test surfaces developed where the reference bands are not overlapping with the spectral band that is to be corrected.
• Examine the effect of disparities in the time of acquisition between the high-resolution and reference images.

1. Integrate the kernel and referencing approaches to BRDF correction in order to improve the level of understanding of the effects of bi-directional reflectance. The sub-tasks were:

• An investigation of how well kernel models can be used to describe the BRDF effects calculated using the referencing method.
• A comparison of the results of the kernel approach and referencing method for specific data sets.

Referencing Method approach to Bi directional Reflectance Distribution Function (BRDF) correction

The use of remotely-sensed, bi-directional reflectance image data for quantitative analysis of the Earth’s surface is often compromised by a pronounced scan-angle-dependent BRDF effect. This BRDF effect is related to the scattering behaviour of the land surface under different illumination and viewing directions. A given pixel’s scattering behaviour can range from Lambertian (scatters equally in all directions) to specular (mirror-like) to other strongly anisotropic behaviour. Pure Lambertian scattering is rare in nature as most natural occurring surface materials develop a pronounced brightening or "hotspot" effect at the backscattered angle (Figure 1). This hotspot occurs when the sun is immediately behind the view direction such that all surfaces, from the micro-scale to the macro-scale, are in full illumination. Shading develops away from this backscattered angle causing the surface to appear increasingly dark. The backscattered effect is normally the most obvious in bi-directional imagery though pronounced forward scattering and other reflectance peaks are sometimes observed depending on the anisotropy of the surface. Therefore, a complete hemisphere of possible surface scattering behaviour is possible for every pixel for every wavelength.

Solving for this BRDF effect is complex especially as every pixel measured remotely has its own BRDF character representing the aggregate of both surface and volume scattering interactions (both single and multiple). Therefore, additive and multiplicative relationships with linear and non-linear associations may be present. Much attention to a BRDF solution has focussed on the kernel approach that involves fitting a theoretical BRDF curve (kernel), selected from a fixed number of types, to the measured data. The result is then inverted to highlight residuals and to refine the input model if required. Thus a successful kernel model usually comprises some linear combination of fixed kernel shapes weighted for each spectral band for each sun-sensor geometry for each surface type.

Other methods based on view geometry and semi-empirical functions rely on a sequence of images acquired across uniform cover under constant solar illumination condition to produce averaged bi-directional reflectance variations for each spectral band along each flight line before fitting these to pre-determined scattering angles images. In practice, it is often difficult to accumulate enough uniform images under the same solar and viewing condition to produce an appropriate bi-directional reflectance surface.

Here we describe an alternative approach to BRDF correction, called the BRDF Reference Method that uses a satellite image to help generate an appropriate BRDF model. There is no requirement for scene classification and the final product preserves the spatial and spectral integrity of the airborne data. The basis of the referencing method uses concurrent, spectrally-similar satellite data as a reference image to the airborne data, from which a pixel-based model of scene brightness variation can be generated.

The method generates a reference model using concurrent satellite imagery for each spectral band (given a wavelength-dependent BRDF effect). The assumption being that the satellite image does not suffer the same BRDF problem for the same area viewed by the airborne image. The residual product of the airborne and satellite data set is then filtered in the fourier domain to isolate and remove the lower frequency spatial content related to the BRDF variation in the airborne data set. The filtered product is then removed from the original airborne data to yield BRDF-free imagery, without compromising its spectral integrity.

The main assumption is that the satellite image has a significantly smaller view angle for the same area sensed by the airborne instrument. For example, Figure 1 shows an area imaged by an airborne instrument and by a satellite. The contrast is in solid angle for the imaged area from the aircraft sensor than that from the satellite. At a given solar elevation and across a homogenous surface of "rough" isotropic materials, the airborne instrument would record a scene brightness variation showing symmetry around the pure backscatter angle (hotspot effect).

This work is arranged to show:

• The Referencing Method applied to Airborne Video data;
• The Referencing Method applied to Airborne Scanner data;
• Constraint based on the values in the area of overlapping frames;
• Application of the Referencing Method over water surfaces;
• Testing the method when reference bands are not congruous;
• Testing the method when the reference image has time disparities; and,
• Testing the method over different cover types.

Figure 1: Shows the principles underlying the referencing method and the effect that slight aircraft roll can have on the hot spot position.

brdf-jan2000-1.htm

brdf-jan2000-2.htm

brdf-jan2000-3.htm

brdf-jan2000-4.htm

brdf-jan2000-5.htm

brdf-jan2000-6.htm

brdf-jan2000-7.htm

brdf-jan2000-8.htm

References

Lyon, R.J.P., Honey, F.R. and Hick, P.T. Second Generation Airborne Digital Multispectral Video: Evaluation of a DMSV for Environmental and Vegetation Assessment. Proc. of the First International Airborne Remote Sensing Conference and Exhibition. Vol 2, 105-116

Jernakoff, P., Hick, P., Ong, C., Hosja, V. and Grigo, S. (1997) Mapping algal blooms using airborne multispectral digital video and the importance of bloom dynamics in the collection of in-water data. Marine Technology Society Journal Vol. 30 No. 4 pp36-45

David L B Jupp & Alan Strahler (1996) Scene Brightness and BRDF: Background document; Image Brightness & BRDF Workshop Issues. on http://www.eoc.csiro.au

Hick P. and Ong, C. (1995) Final report for the processing of DMSV data and production of Vegetation Maps of the Weipa and Andoom Minesites. Report from CSIRO Minesite Rehabilitation Research Program for Comalco Minerals and Alumina. Pp 1-16Report No???

Pickup, G., Chewings, V.H. and Pearce, G. (1995). Procedures for correcting high resolution airborne video imagery. International J. Rem. Sens., 16, 1647-1662.

Roujean, J.L., Latoy, M. and Deschamps, P.Y. (1992). A (bi-directional reflectance model of the earth's surface for the correction of remote sensing data. J. Geophys. Res., 97, 20455-20468.

Tucker, C.J. (1979) Red and photographic infrared combinations for monitoring vegetation. Remote Sensing of the Environment 8, 127-150.

Ong, C., Craig, M., Hick, P.T. and Jupp D.L.B. ( 20--) The REFERENCING METHOD: correcting variable illumination effects (BI-DIRECTIONAL REFLECTANCE Distribution Function) in remote sensing. Prep.????????