An objective of the High Resolution Scene Brightness or BRDF Task for the CSIRO Earth Observation Centre is the development of guidelines for best practice outputs (including prototype software) for high resolution Aerial Photography, Video, Digital Camera and Airborne Scanner data to minimise the effects of scene brightness. A main outcome of the task has been the development of the reference method and software by CSIRO Exploration and Mining (CEM). This report describes initial testing of the reference method approach by CSIRO Wildlife and Ecology (CWE) to June 2000.

The following two BRDF approaches were used:

• The Reference Method (Ong et al., (2000a)) which uses a satellite image to help generate an appropriate BRDF model. The procedures were run by CEM as the software and documentation were not ready for release when testing started (1/2000).
• A semi-empirical procedure developed by CWE (Pickup et al. (1995a), Pickup et al. (1995b)) which uses viewing geometry and a sequence of images acquired across a fairly uniform landscape type.

Data and Methods

Although the CWE DMSV aerial video archive contains imagery acquired over a range of landscapes and flight conditions, for the initial test we used transects for which corresponding Landsat TM data had been purchased previously and where ground differential GPS data had already been collected. Both transects are in the Kunoth Paddock test site (23^o 32' S, 133^o 34' E) and flight details are given in Table 1. Corrections were applied for the effects of aircraft motion, camera misalignment and lens distortion (Pickup et al. 1995a, Pickup et al. 1995b). Details of the Landsat TM data used in the Referencing Method correction are given in Table 2.

Three overlapping video frames from each transect were spatially registered to UTM together with Landsat TM imagery for the same areas. These files were then sent to CEM where Mike Caccetta applied the BRDF Reference Method correction procedures using TM data from two different dates for each frame.

The raw video data were also corrected using the largely automated CWE BRDF procedures. All 36 frames in each transect were used to calculate average bi-directional reflectance parameters.

Two approaches were used to check the consistency of spectral values in the frame overlap areas for three frames before and after correction for differential illumination. Correlation coefficients were calculated using the spectral values of identical individual pixels located manually from the overlap portion of each pair of frames. The images were also classified using a supervised Maximum Likelihood procedure in microBRIAN. Class statistics were calculated using a consistent set of training patches (>50) from one frame of each pair which were applied to both frames to classify the imagery into broad categories such as soil, litter, ‘green trees’, ‘grey trees’ and shadow. The proportion of pixels in each category was then calculated for a portion of the overlap area.

Results and Discussion

Table 3 shows correlation coefficients for a series of points in the overlap between Greentree frames G16 and G18 for uncorrected data and after corrections using the CWE method, the Reference Method using the 12/94 TM reference image and the Reference Method using the 12/95 TM reference image. Table 4 shows similar results for the Highway frames T16 and T18. In both cases, the uncorrected data has slightly higher correlations at some wavelengths. Tables 5 and 6 show the proportion of classified pixels in each of the broad soil and vegetation categories in the overlap regions. Both BRDF correction procedures resulted in an improvement in classification consistency.

For operational use, a major consideration is the time required to apply BRDF corrections to each dataset. We usually spend up to 2 hours to apply the CWE correction to a transect of up to 36 frames. Implementing the Reference Method takes considerably longer as each frame needs to be individually registered to a satellite image.

Table 1. Details of DMSV aerial video data used.

Video data	Highway transect (open woodland)	Greentree transect (mulga Acacia aneura)
Date acquired	26/3/97	26/9/95
Flying height AGL	1420 m	1420 m
Resolution	1 m	1 m
Transect bearing	78 degrees magnetic	124 degrees magnetic
Solar zenith	40 degrees	47.8 degrees
Solar azimuth	50.9 degrees	287.8 degrees
Local time	10:35	15:21
Coords AMG Zone 53	N 7398500 E 353500	N 73930000 E 353800

Table 2. Details of Landsat TM data used in the Reference Method corrections.

TM imagery	Highway transect		Greentree transect
Date acquired	5/3/95	27/4/97	31/12/94	1/12/95
Solar elevation	43.76o	37.99 o	51.79 o	51.51 o
Solar azimuth	75.8 o	48.69 o	97.83 o	94.78 o

Table 3. Correlation coefficients for spectral values of points selected from the overlap of Greentree frames G16 and G18. (n=25).

	Uncorrected	Correction method
Nm	RAW	CWE	Ref 12/94	Ref 12/95
450	0.850	0.826	0.827	0.811
550	0.951	0.952	0.947	0.937
650	0.916	0.924	0.915	0.922
770	0.791	0.788	0.784	0.783

Table 4. Correlation coefficients for spectral values of points selected from the overlap of Highway transect frames K16 and K18. (n=24)

	Uncorrected	Correction method
nm	RAW	CWE	Ref 3/95	Ref 4/97
450	0.961	0.952	0.924	0.922
550	0.985	0.977	0.962	0.959
650	0.949	0.973	0.955	0.949
770	0.961	0.957	0.954	0.950

Table 5. Percent of pixels classified into broad categories for an area on Greentree frames G16 and G18. (n=25).

	Uncorrected		Correction method
	RAW		CWE		Ref 12/94		Ref 12/95
Frame	G16	G18	G16	G18	G16	G18	G16	G18
Soil	4.8	0.8	6.4	5.8	8.4	7.4	8.5	7.7
Soil&litter	61.1	58.1	59.2	58.6	56.3	56.1	54.6	54.4
Litter	18.2	20	18.3	19	18.6	19.3	20.2	20.8
Whitewood	3.9	4.5	4.5	4.9	5.1	5.6	6.5	7
Shadow	2.2	3.5	1.9	2	1.5	1.4	1.3	1.2
Mulga	9.2	12.5	9.3	9.4	9.7	9.7	8.5	8.4
Unclassified	0.5	0.7	0.3	0.3	0.5	0.5	0.4	0.4
(n=200900)	100.0	100.0	100.0	100.0	100.0	100.0	100.0	100.0

Table 6. Percent of pixels classified into broad categories for an area on Highway frames K16 and K18.

	Uncorrected		Correction method
	RAW		CWE		Ref 3/95		Ref 4/97
Frame	T16	T18	T16	T18	T16	T18	T16	T18
Soil	39.8	45.3	41.5	41.3	43.2	47.1	42.7	44.3
Grass	4.5	6.8	4.8	4.8	4.9	4.8	5.3	5
Litter	30.6	24.8	28.5	28.5	30.2	27.9	29.3	29.2
Mulga	17.7	16.5	16.4	17.7	15.3	15.2	15.8	16
Shadow	4.6	1.2	5.4	4.2	4.5	3.4	5.5	4.3
Unclassified	2.9	5.3	3.3	3.5	1.8	1.6	1.4	1.2
(n=108300)	100.0	100.0	100.0	100.0	100.0	100.0	100.0	100.0

Further testing

We are keen to try the Reference Method on DMSV data which could not be corrected satisfactorily using the CWE approach. A suitable dataset might be one acquired for CSIRO TAg over Eucalypt woodland in the Charters Towers area in Queensland (Dotswood) if suitable Landsat TM data can be located within CSIRO.

Figure 1. Frame G16 from the Greentree transect. The yellow rectangle indicates approximately the portion also on frame G18.

References

Ong, C., Hick, P. and Caccetta, M. (2000). An Integrated Approach to Correction of Scene Brightness Variation in High-resolution Airborne Imagery. Final Report of Stage One of the BRDF Task 3.1.

Pickup, G.,Chewings, V.H. and Pearce, G. (1995a). Procedures for correcting high resolution airborne video imagery. International Journal of Remote Sensing 16, no. 9, 1647-1662.

Pickup, G., Bastin, G.N., Chewings, V.H. and Pearce, G. (1995b). Correction and classification procedures for assessing rangeland vegetation cover with airborne video data. Procs. 15th Biennial Workshop on Videography and Colour Photography in Resource Assessment, Terre Haute, Indiana, 305-314.

Vanessa Chewings, Graham Pearce and Gary Bastin
Centre for Arid Zone Research
CSIRO Wildlife and Ecology

12/6/2000