Acquisition and pre-processing of reflectance factor spectra acquired using the

As mentioned before, the reflectance factor of a target is the ratio of the spectral response of the target to the spectral response of a reference sample under the same conditions of observation and illumination. Therefore the reference spectrum should be

acquired before the target measurements begin, and should be repeated every time the illumination conditions change.

The FieldSpec® Pro FR spectroradiometer samples spectra continuously but reports a time averaged spectrum. To acquire the spectral reflectance of a target, the target must be kept in the FOV of the fibre optic cable until the reported time-averaged spectrum stabilises. The practice of time-averaging many spectra is adopted to reduce measurement error when collecting spectral data. The FieldSpec® _{Pro RS}3 _{software is set up to do this averaging. For}

our measurements one acquired spectrum is the average of 25 measurements for target, dark current and white reference spectra.

The measurement protocol for acquiring reflectance factor spectra using the Spectro-CAPP is as follows. Warm up the FieldSpec® _{Pro FR for at least 90 minutes. In the}

field turn on the CAPP‟s light (which is attached to a battery) and let it warm up for 15 minutes just before beginning the spectral measurements. Attach the FieldSpec® _{Pro FR‟s}

fiber optic into the CAPP, place the white reference sample (matt white ceramic tile) under the CAPP and proceed with the standard procedures to take relative reflectance (reflectance factor) measurements using the FieldSpec® _{Pro FR (optimization, collection of dark current}

and white reference scan). Once the white reference has been collected, place the CAPP over the target (pasture sward) and collect the target spectrum. Move the CAPP to the next target and acquire new spectrum. To measure plots with dimensions greater than the spot measured within the FOV of the Spectro-CAPP (Figure 2.3), several spectra are acquired per plot and posterior averaged. When collecting a target‟s spectra there is a need to periodically check the white reference scan (by placing the CAPP over the matt white ceramic tile), and collecting a new white reference scan whenever necessary (e.g. when deviations such as steps or slopes are observed in the 100% white reference line).

In the protocol described above, since the white reference scan is acquired by placing the ceramic tile under the CAPP (0 cm height), it is assumed pasture canopy height is short and constant. The ideal would be to have a mechanism to slide the reflectance standard panel up inside the CAPP according to the PSH. Thus the irradiance of both reflectance standard and target would be the same. Because the CAPP did not have such mechanism, and to manually change the tile height during measurements in the field would be unpractical, the reflectance factors collected using the Spectro-CAPP need to be

posterior compensated for PSH. The radiance of the ceramic tile (which was previously cut to a size that could fit inside the CAPP) at different heights inside the CAPP (every centimetre between 0 to 15 cm height) was acquired (Figure 2.6). With these data - the radiance of the ceramic tile for specific heights and also measurements of PSH, the pasture reflectance factor measured with the Spectro-CAPP can be corrected. The correction is done by multiplying the pasture reflectance factor by the tile‟s radiance at 0 cm height and dividing by the tile‟s radiance at the specific PSH, for each wavelength. Another option would be to record in the field the radiance of the target instead of the reflectance factor, and posterior divide it by the tile‟s radiance at the specific PSH. However the advantage of recording the reflectance factor is that quality control of the spectra can be done visually while acquiring the data and some errors can be eliminated before the data are saved.

Following the methodology described above, the spectral reflectance factors from ten pasture plots were measured using the Spectro-CAPP. Each spectrum was the average of 10 spectra acquired within each plot. Data were obtained for the two fibre optic grip mounting positions. To remove the signal noise (mainly at the beginning and far end of the spectrum analysed, due to low signal-to-noise ratio - SNR) and step problems (steps discussed later in the paper) two data pre-processing procedures were applied to the spectra, a Savitzky-Golay smoothing filter (Savitzky and Golay 1964) with smoothing window size of 81 and polynomial order 4, using The Unscrambler® 9.7 software, and a de-step procedure (Daniel, P., Ticehurst, C., and Thulin, S., personal communication, 29 March 2007) using ENVI 4.3 plus IDL 6.3 (Figure 2.8). The de-step procedure assesses the actual differences at the steps between the three detectors in the spectrum, and then it takes the midpoint in each step as the new point for the corrected spectrum to pass through. Each of the three spectral segments is tilted up or down until they pass through the midpoints while the ends of the spectra (350 nm and 2500 nm) are held fixed. The authors later became aware of another de-step method called Splice Correction, available on the ViewSpecTM

2.2.7. Comparison between pasture spectra acquired with the Spectro-CAPP and with the FieldSpec® _{under sunlight conditions}

In addition to the measurements using the Spectro-CAPP, some pasture plots were also measured with the FieldSpec® in normal mode (under sunlight conditions) using as white references the matt white ceramic tile (295x295 mm) and a spectralon® _{disc (9 cm}

diameter). The fibre optic input when using the FieldSpec® _{under sunlight was fixed at}

nadir position at 45 cm height, so that the area measured (equal to a circle with a 20 cm diameter) would be comparable to the area measured when using the CAPP. For each plot four measurements were conducted: using 1) the FieldSpec® _{under sunlight and the}

spectralon® _{disc; 2) the FieldSpec}® _{under sunlight and the ceramic tile; 3) the Spectro-}

CAPP top-grip and the ceramic tile; and 4) the Spectro-CAPP side-grip and the ceramic tile. For the inter-comparison of spectra collected using the two different reflectance standards used, the reflectance factor spectra obtained were converted to absolute reflectance (reflectance factor spectrum multiplied by the calibrated reflectance spectrum of the spectralon® _{disc or the ceramic tile).}