EXPERIMENT DESIGN AND SAMPLE COLLECTION PROTOCOLS
5.3 Sample Collection and Preparatory Work
5.3.3 Post Field Work Sampling
The post fieldwork sampling is defined as the discrimination of useful data and samples that were collected during the field trips. It involved:
a) drying and thin-sectioning of mineral samples b) photographing and then drying of vegetation samples c) identification of vegetation samples
d) selection of the suitable imagery scenes, and e) pre-processing the selected scenes.
5.3.3.1 Mineral Reference Samples
Reference samples of the rocks and detritus were collected with the view of identification of the origins of the material. Some were thin-sectioned. The thin-sections of the loose material revealed that it was composed of a mixture of clay, silt and composite grains in a clay matrix. This indicated that it was probably of aeolian origins. Some of the surface materials of the mineral samples were in such thin layers, that the sections failed to reveal any useful information as to their identification or origin.
5.3.3.2 Plant Reference Specimens
The vegetation reference samples were recorded and identified with reference to the Melbourne Herbarium collection. A list of plant species previously reported from the area appears as an appendix. The usual method of preservation of botanical specimens (pressing) (Bean, 2006) was not employed in the field due to the plan to use the samples for the acquisition of laboratory spectra. The unavailability of a suitable instrument and the severe degradation of the samples meant that the spectra were not acquired.
5.3.3.3 Data Selection
The quantity of data collected during the campaign was extremely large. It included meteorological data, field samples, field spectra, handheld photographs, and the aerial and satellite imagery. In order to select the most suitable imagery to be used in the research, very strict criteria were used. The selection of these criteria was to limit the effects of atmosphere, weather, operator differences, and instrument limitations. Assessment of the imagery for its suitability for calibration included using METARS weather observations from the Woomera Australian Bureau of Meteorology meteorologic station, general meteorological observations, and field observations recorded by the ASD field teams. Flight lines were excluded based on time of acquisition being too early or too late in the day; cloud cover, or poor signal to noise ratio. On this basis, of the 69 flight lines acquired during the period, only 20 were considered suitable for calibration. An additional selection criterion of pixel size, reduced this number of images to thirteen. Five of these were excluded because they did not contain imagery of the calibration panels. This excluded all but eight flight lines. Three of these were excluded based on location. The five remaining for analysis included two acquired flying in the solar plane on north-south flight lines and three perpendicular to the solar plane on an east-west heading.
The final selection criteria were: a) weather,
b) atmospheric conditions,
c) time of day at moment of capture, d) location, and
5.3.3.4 Instrument differences between the two ASDs
Once back in the laboratory, the spectral characteristics of the two instruments were again compared to determine the degree of similarity between the two. It was concluded that due to the different foreoptics used that the variety and quantity of target that fell within the fields of view of the two instruments were so spatially different as to render the spectral measurements acquired in the field, incomparable. The signals of the two instruments were also very different with respect to noise. A direct comparison between curves of Spikeycottonballs (Dissocarpus paradoxus) and Gibbers for the two instruments revealed a SNR difference between the two instruments with a smoother curve for I1. A shift in the absorption features was also noted between the two curves, (refer to Figure 32). The curve for Gibbers was compared with a USGS standard curve of Hematite, the identified dominant iron oxide (Wopfner and Twidale, 2001), and the curve of I1 returned a match with 83% confidence. In light of this, combined with the smaller field of view, data acquired by I1 was used exclusively. This reduced the number of sample points available for ground-truthing from 685 points to 309, a reduction of 55%.
5.3.3.5 Weather
The weather encountered during the HyMap® collection was uncharacteristically poor. To mitigate against weather-induced issues, imagery was selected for analysis to limit the effects of cloud cover. The basis for the decision of which imagery to include in the analysis was visible cloud in the imagery and field note comments marking cloud cover as an issue at the time of acquisition.
5.3.3.6 Atmospheric conditions
Raised dense dust clouds followed the detonation events. One day of collection was lost due to conditions adverse to the operation of the ASDs. In addition, the poor atmospheric conditions prevented some field spectra from being included in the analysis due to excessive noise. Based on the daily collection rate for the other days, it is estimated that approximately 25% of the field data collected by I1 was excluded on this basis. HyMap® imagery based on a combination of the METARS data and field notes was also excluded due to raised dust.
5.3.3.7 Time of Day at Moment of Capture
As pointed out in Chapter 2, the atmosphere selectively absorbs and scatters incoming EMR according to the wavelength of the radiation, with the most noticeable effects
occurring early or late in the day. Due to the imagery being collected for a number of end- users and purposes, collection was undertaken across most of the day. Therefore, in line with the recommendation by Salisbury (1998), imagery acquired extremely late or early in the day was excluded from the final analysis.
5.3.3.8 Pixel Size
A review of raw imagery prior to calibration revealed that the calibration panels were evident only in the imagery scenes with the highest spatial resolution. This limited the ability to calibrate the imagery using the ELC method.
Okin and Roberts (2004) reported that in arid and semi arid environments, the classification of vegetation was extremely difficult due to the sparse nature of the canopy. King et al., (2000) reported that “low concentrations” of vegetation were sufficient to provide an accurate identification on cropped lands after classifications correctly identified harvested crops in the San Luis Valley, near Summitville, Colorado, USA. They did not report the exact coverage; however, Lewis (2000) reported that if a given target contributes between 20% and 25% of a pixel signal this was sufficient to identify the target in the arid land north of Adelaide, South Australia.
The sparse nature of the vegetation cover at Woomera limited the size of the pixels considered suitable for the analysis. It was the reason for a pixel size of 2 m specified in the original experiment design. Operational and environmental constraints meant that this specification could not be met and a pixel size of 3 m was the minimum collected. The above limitations meant that only the imagery with the highest spatial resolution, 3 m, was selected for analysis.
5.4
Summary
Considerable changes and amendments to the planned experimental design were necessary during the campaign:
a) unexpected weather conditions limited the imagery available;
b) operational constraints limited the altitude of the aircraft and hence the pixel sizes were different to those operationally planned;
c) unexpected differences in the fields of view (FoV) of the two field spectrometers meant the spectra collected by each were incomparable.
The selection process for the most appropriate imagery for the research work was quite intensive. It excluded imagery that was assessed as hampering the fair comparison between the classification results. Further study and analysis of the coarser resolution imagery excluded from this study may be possible at a future date. Less than a third of the expected spectra collected were available for analysis. Mostly, this was due to the spectra collected by the two instruments being incomparable, however weather and atmospheric conditions also contributed. This would limit the availability of spectra for use in the statistical analysis of the imagery. A mitigating strategy was included in the original plan. A subsequent field trip to collect additional ground-truth data was planned, as well as the use of pan-sharpened multi-spectral imagery to determine the spread of ground-truth data points. Subsequent operational requirements for the site prevented this field trip, and the pan-sharpened imagery was used for the final assessment of classification accuracy. The next chapter, Chapter 6, presents the results of classification maps derived from the data collected. Explanations of the classification methods and techniques used in the analysis are included in the chapter.