4 Airborne Measurements and Preprocessing of the Data
4.2 Data Preprocessing
4.2.4 Validation of the Preprocessed Data
The preprocessing sequence provides atmospherically corrected spectral ground reflectance data. This enables the quantitative analysis as well as the comparison with reflectance data derived at different flight dates, because the signal contains only the part of the information which is directly related to the observed pixel on-ground and is not influenced by the atmosphere, sensor characteristics or by radiation of pixels surrounding it.
During preprocessing it turned out that the reflectance spectra of the unfocussed data showed an insufficient quality for parameter derivation. Therefore the AVIS data measured on 29th and 30th April as well as on 25th May could not be included in analysis.
The results of the preprocessing must be validated for their quality. This was done using measurements of a ground based spectrometer, in this case GER SIRIS (see section 3.4.4.2), which were carried out nearly simultaneously to the image acquisition with AVIS, namely on 3rd and 18th July 1999. The results will be discussed in the following sections. Figure 4-10: Modelled total at-sensor radiance using LOWTRAN-7 for the 18th July 2000 with a spectral
4.2.4.1 1999
For 1999 the preprocessed image data show a high consistency in the VIS region, but in the NIR region the spectra show increasing variations towards longer wavelengths. The problems, which occurred, will be described on the basis of a typical example provided in Figure 4-11.
The graph on the left shows reflectance spectra of five individual pixels spread over a maize field. The course of the curves is very similar until 740nm. The NIR plateau up to 850nm is also characterised by a similar trend of the curves, but with higher variations than in the VIS. The maximum amplitude of these variations reaches 5% reflectance. In contrast to the variations, the different level of the curves at the NIR plateau is a common phenomenon. The reflectance in the NIR region is dependent on the leaf structure of the plants (Gausman, 1974), which may vary within a canopy, or their leaf angle (Price, 1994) as well as on variations in the viewing geometry such as the view angle (Kennedy et al., 1997), the multiple scattering or shadowing effects in the canopy (Kuusk, 1991). Price (1994) measured NIR reflectances in the range of 29 to 43% during a sequence of measurements within a maize canopy. The variability of spectra is described more detailed in section 5.3. The differences in the overall reflectance level of the AVIS single spectra are most probably due to the spectral variability of a maize canopy and cannot be ascribed to the preprocessing. 0 5 10 15 20 25 30 35 40 550 600 650 700 750 800 850 900 950 1000 Wavelength [nm] Refleczance [%] 0 5 10 15 20 25 30 35 40 45 550 600 650 700 750 800 850 900 950 1000 Wavelength [nm] Ref lect ance [ % ] SIRIS AVIS
The variations between two bands in the 850–1000nm wavelength region are in the magnitude of 10% reflectance. These variations are too high for a quantitative or even qualitative analysis of the single reflectance spectra. Figure 4-11 also shows that the variations are not systematically, but appear to be random. This is confirmed when looking at the average field spectrum, which is shown in the right graph in Figure 4-11, where most of the variations are averaged. These facts led to the conclusion that the variations Figure 4-11: AVIS reflectance spectra for 1999 for individual pixels (left) and field average (right),
are noise. The SNR of the system decreases steeply from a value of 65 at 760nm towards the longer wavelengths (see section 2.4). At 850nm the SNR is about 30 and decreases further to 20 at 1000nm. The signal in the 850–1000nm wavelength region is strongly influenced by system noise, and therefore at least the single spectra cannot be analysed in that wavelength range. The quality of the averaged spectra depends strongly on the number of pixels that are included in the averaging. This number of pixels varies with the field size and the flight altitude. Both the size of the fields being investigated and the observation altitude differ, but should be comparable within the analysis. Therefore the averaged spectra were also discarded for wavelengths above 850nm.
The comparison of the averaged AVIS field spectrum with the SIRIS spectrum (Figure 4-11) shows good results considering the fact, that point measurements are compared with two-dimensional measurements. The differences in the NIR are of the magnitude 3.5% reflectance, in the VIS they range from 1 to 1.4%. In the region of maximum chlorophyll absorption at 680nm the SIRIS measured 1.1% reflectance while AVIS measured 2.1%.
4.2.4.2 2000
0 10 20 30 40 50 60 550 600 650 700 750 800 850 900 950 1000 Wavelength [nm] Reflectan ce [% ] 0 10 20 30 40 50 60 550 600 650 700 750 800 850 900 950 1000 Wavelength [nm] Reflectan ce [% ]The results of preprocessing for the year 2000 will also be discussed on the basis of an example, which is shown in Figure 4-12. The acquisition date of the measurements is the 22nd July. Compared to the 1999 data, the individual pixel spectra, which are spread over a field of wheat, show a great improvement, especially in the NIR. The course of the curve is smooth up to 900nm with variations of at most 4% reflectance. The water absorption feature at 940nm can be reconstructed, but in this wavelength region the variations increase up to 8% reflectance. The reflectance values beyond 980nm drop down both in the individual pixel spectra and the averaged field spectrum. The comparison of reflectance spectra of different land use types and other acquisition dates confirmed this Figure 4-12: AVIS reflectance spectra for 2000 for individual pixels (left) and mean field spectrum (right),
observation. The reason for this artefact is most probably the decrease in system sensitivity combined with the decreased SNR that cannot be compensated during preprocessing. The 2000 data are analysed in combination with those of 1999, therefore the reflectances beyond 850nm will not be included in the further analysis.
For 2000 no field spectrometer data are available for validation. The mean field spectrum of the maize canopy therefore cannot be validated with ground-based measurements, except with the DLR hangar spectrum. This spectrum is used as a calibration target during preprocessing, thus it cannot be used as a validation target.