Pre-processing of Imagery

6.2.1

Methods selected for use for imagery collected over the

Woomera Explosives Test (WET)

The HyMap® imagery was received as raw radiance. Good calibration of the imagery is crucial to a satisfactory classification outcome (Mather, 1991; Clark, 2002; and Liang, 2004). As the literature presented contradictory reports as to the efficacy of the commonly used calibration routines, two calibration methods were selected to calibrate the imagery on this occasion:

a) Empirical Line Calibration (ELC) method, and

b) a combination of ATREM, a radiative transfer model based on the 5S model (Center for the Study of Earth from Space (CSES), 2004), to remove the atmospheric contribution, and then running the ELC to fine tune the calibration, as suggested by Clark et al (2002).

The two methods were selected on the basis that one utilised spectra acquired from in- scene calibration panels to determine atmospheric contributions to the signal and then remove them, while the other used a model to remove atmospheric effects from the imagery, then fine-tuned the result with in-scene calibration panels. Theoretically, the

methods were independent of any changes that occurred between the capture of the individual scenes.

6.2.1.1 ELC Method

The calibration panels used for the ELC for the test site were 15 x 15 m and were laid out in open areas clear of vegetation. Indigenous rocks pegged down the panels to minimize any introduced signal contamination. Panels of 15 x 15m square were felt to be of sufficient size for a pixel size of approximately 2 m, which was the operationally planned size of the collected data pixels. This gave an approximate number of pixels for calibration as 49 (7 x 7). Clark et al (2002) point out that a larger number of pixels improves the noise in the signal, according to the square root of the number of pixels averaged, and for the AVIRIS sensor, they recommend at least 25 pixels. Using this number as a guide for the HyMap sensor, even losing two pixels in each row and column (one from each end), and the minimum number of pixels for a good average was satisfied. Unfortunately, the minimum flying altitude (1400 m) resulted in a pixel size of around 3 m. At best, there could only be a 5 x 5 pixel grid (25 pixels). Contamination from the adjacent material on the edge pixels meant losing one pixel from each end of the rows and columns (3 x 3 pixels). These factors, combined with the misalignment of the pixels over the panel, meant that only one pixel for each panel had a spectral curve suitable for use in the calibration. Additionally, as has already been pointed out, the calibration panels were too small to be detected on imagery collected at higher altitudes. Distortion of the pixels due to aircraft pitch and yaw made them very difficult to detect even in some of the low altitude scenes. Once the imagery was geo-rectified, the pixels were evident in a given scene; however, in order to preserve the integrity of the data and limit any misalignment of pixels the standard procedure is that imagery is classified with a minimum of pre- processing, including geo-rectification (Research Systems Inc, 2005).

Although shown not to be ideal, this procedure (ELC) was adopted as one method of atmospheric calibration for the imagery associated with this research using the ASD- acquired library panel reference spectra. This was used to define the apparent reflectance according to Equation 2.7, and applied across all pixels in the scene. This is discussed further in Chapter 7.

The radiative transfer model algorithm, ATREM, is based on the 5S code (Center for the Study of Earth from Space (CSES), 2004); however, it includes more robust methods for determining the water absorption features as well as some of the atmospheric gases. Situation-dependent factors are included in the processing via the small routine presented in Figure 34. The routine allows the inclusion of location (LINES 4-7), and date/time (LINE 3) to be included in the calculations. This is used to calculate solar azimuth and zenith. A flat land surface is assumed, and in the instance of Woomera, this was a valid assumption. The flying altitude of the aircraft (LINE 2) and land surface elevation (LINE 16) are used to calculate the atmospheric pressure. The concentration of some of the gases of interest is assumed. However, the inclusion of the water vapour absorption bands (LINE 10), and the visibility data (LINE 15) allow an atmospheric model, in this case continental, mid-latitude summer, to calculate the water and ozone (H2O and O3) content of the atmospheric column. The optical thickness of that column is also calculated. ATREM calculates the water concentration by ratioing the depth of the water vapour absorption bands (Center for the Study of Earth from Space (CSES), 2004), and outputs the result as an image (LINE 23).

LINE 19 indicates the size of the image cube to be analysed and corrected. The output is an atmosphere-corrected image (LINE 20); appropriately named and placed in a designated directory along with the water vapour image (LINE 23).

Figure 34 Macro used to run the ATREM atmosphere removal algorithm LINE 1 HYMAP LINE 2 1.532 LINE 3 10 07 2002 03 24 07 LINE 4 31 00 38.8 LINE 5 S LINE 6 136 46 46.8 LINE 7 E LINE 8 0. LINE 8 C:\ATREM_WET_ANALYSIS\tape7R2B1400_rad\ATREM_files\wavs_kerryn.txt LINE 9 1 LINE 10 0.865 3 1.030 3 0.940 7 LINE 11 1.050 3 1.235 3 1.1375 7 LINE 12 2 LINE 13 1 1 1 1 1 1 1 LINE 14 0.34 LINE 15 1 50 LINE 16 0.139 LINE 17 C:\ATREM_WET_ANALYSIS\tape7R2B1400_rad\imagery\tape7R2B140052_rad.bil LINE 18 1 LINE 19 0 256 256 126 2 LINE 20 C:\ATREM_WET_ANALYSIS\tape7R2B1400_rad\imagery\tape7R2B140052_ref.bil LINE 21 0. LINE 22 2000. LINE 23 C:\ATREM_WET_ANALYSIS\tape7R2B1400_rad\imagery\tape7R2B140052_rad_h2o.img LINE 24 C:\ATREM_WET_ANALYSIS\tape7R2B1400_rad\imagery\tape7R2B140052_rad_sub_trans.lib

The advantage of a radiative transfer model for the correction for atmospheric effects is the accommodation of the bidirectional reflectance when applying the atmospheric correction to the imagery. Referring back to Chapter 2, Figure 2, the various parameters that define the BRDF are illustrated. When preparing to run the ATREM process the short routine listed in Figure 34 provides the factors used to define the BRDF parameters. Table 8 summarises the BRDF parameters and which factors define them. Vermote (1997) Table 8 Table summarising the relationship between BRDF parameters and the factors required for their definition, as used in the ATREM model

Parameter Description Defining Factors

θ0

Angle between the incident radiation and the normal to the surface

• Date • Time • Location

θ1

Angle between the reflected radiation and the normal to the surface

• Flying altitude • Surface elevation • Sensor type

φ0

Angle between the solar plane and the sun

• Date • Time • Location

• Surface elevation

φ1

Angle between the solar plane and the sensor

• Date • Time • Location • Surface elevation • Sensor altitude • Sensor type

Incident defuse EMR

• Atmospheric model selected • Aerosols type • Visibility Surface tangent vector • Location • Surface elevation

points out that the 5S code assumes a Lambertian surface and uses approximations to calculate the BRDF, to mitigate the expensive computational load on the less efficient computers available when the code was first designed. As ATREM is based on this code, it is likely that those approximations remain. Once the ATREM was complete, the ELC was applied in cascade as per section 6.2.1.1.

Once the atmospheric correction of the imagery was complete, a correction for the sensitivity of the sensor array, as supplied by HyVista, was applied to the imagery. This concluded the pre-processing of the imagery. The atmospheric corrections are further discussed in Chapter 7.

6.3

Imagery Classification Algorithms