358
Campaign (EISAC’89) which took place at a site in the Somerset Levels during May 1989. The specific focus
of interest was an experimental grassland site on Tadham Moor managed by the Institute of Grassland and
Animal Production (IGAP) and the Institute of Terrestrial Ecology (ITE). The site comprised 20 ha of
unimproved species-rich hay meadows located within a Site of Special Scientific Interest (SSSI). The
experiment design comprised three replicate blocks, each of which contained four fields to which levels of
nitrogen fertiliser between 25 and 200 kg per hectare had been added annually and a single reference field
which had no application of nitrogen fertiliser.
During the EISAC’89 experiment, image data were obtained with the Compact Airborne
Spectrographic Imager (CASI) (see Jones, 1992), and a large body of ground data was collected. Reflectance
data were collected using several spectroradiometers from the Natural Environment Research Council
Equipment Pool for Field Spectroscopy (Rollin, 1991). This paper discusses bi-conical spectral reflectance
data (0.35trm-23/tm) collected with a dual-beam GER IRIS MKIV spectroradiometer over Replicate Block
3 on 21 May 1989. Time constraints limited the number of spectra collected at each sample location to a
single spectrum at each of 6 randomly located quadrats in all 5 nitrogen treatments in Replicate Block 3,
resulting in a data set of 30 spectra. At each sampling point, herbage was harvested from a 0.306m 2 quadrat
and the following variables extracted: fresh weight, dry weight, percentage gravimetric water loss on drying
and chlorophyll a and b. A measure of canopy water status was calculated as the gravimetric water content
expressed as a percentage of the fresh biomass weight per square metre area of canopy. This is referred to
here as wet% and is calculated as follows.
wet weight per m 2 - dry weight per m 2
wet weight per m 2
* 100
This analysis attempts to establish if useful relationships between spectral parameters and leaf water
content can be identified in field data focusing primarily on the correlations between the spectral parameters
and wet%. The contribution of canopy geometry and soil background is ignored on the grounds that grass
canopy studied was complete and relative unif orm across all the sample plots, so the effect is likely to be
insignificant.
The analysis is confined to the 400-1360nm wavelength region which includes two water absorption
features at 940nm and 1150nm. Other, major water absorption features occur in the SWIR; between 14001500nm
14001500nm and at 1920nm but outside the atmospheric windows ie. within the wavelength regions affected by
atmospheric water vapour. This study focuses mainly on the 1150nm water feature.
3. SPECTRAL DATA PROCESSING
3.1. Pre-processing
Pre-processing of the spectra required instrument specific calibrations to correct for dark current, gain,
gratings overlap, target-reference detector function and wavelength. A further calibration was applied to
correct for the reflectance of the BaS0 4 reference panel used as the reflectance standard. At this stage, data
were output in a standard format comprising paired wavelength and absolute reflectance values as xy co
ordinates. Efficient manipulation of high dimensionality data sets on a personal computer is beyond the
capabilities of many general purpose statistical packages and was performed in this case using the signal
processing package ASYST. This greatly simplified the process of correlating spectral data at all wavelengths
with each of the biophysical parameters in turn and also facilitated advanced smoothing and data
manipulations.
3.2. Resampling
Spectra measured with the GER IRIS MK IV have different sampling resolutions over the visible and nearinfrared
nearinfrared (NIR) wavelength ranges (400-1 lOOnm) and the shortwave-infrared (SWIR) region. In the region
400-1100nm the resolution is just under 2nm compared with almost 4nm for wavelengths above HOOnm.
Data within each wavelength region were dealt with separately and were initially resampled to a regular
interval by fitting a cubic spline through the data points. For the visible/NIR wavelength range the spline
was fitted through 440 points, resulting a modified spectral resolution of approximately 1.5nm. For the longer
wavelength range a spline of 134 points was fitted used, resulting in a resolution of approximately 2.5nm.
