3 Description of field-measured data

The data used in the experimental chapters of this thesis were collected during the growing season of 1997 from a study site based around Barton Bendish Farm, Norfolk (52" 37.2' N 0" 32.0' E). The farm covers some 9 km2 and produces winter wheat, several varieties of spring barley, peas and sugar beet. Due to the initial work done here (Disney et al., 1997) and subsequent field experiments, this site and the surrounding area has been taken on as a MODIS core validation site.

Three barley fields, and one wheat field were selected at the start of the fieldwork campaign. These fields are identified as follows (and will be referred to as such throughout):

barley 24th AprilFields ba_104, ba_112 and ba_119.

barley 13th MayCombines measurements made from the 13th - 15th May 1997 (due to bad weather) in fields ba_112 and ba_119.

barley 4th JuneFields ba_104, ba_112 and ba_119

barley 24th JuneFields ba_112 and ba_119

wheat 23rd MarchField ww_109

wheat 23rd AprilField ww_109

Figure 3.1 shows the fieldwork site at three different scales: an AVHRR image of the UK (9/8/95) showing the East Anglia region; a false-colour composite SPOT image (4/4/97) of the NIR, visible red and green bands highlighting vegetation as red. Also shown is a scanned aerial photograph of the Barton Bendish farm and surrounding area (6/8/97), obtained by the Natural Environmental Research Council (NERC) Airborne Research System (ARS) Facility aircraft. The Barton Bendish farm area and the fields selected for this study are marked on the image. RAF Marham air-base, clearly visible in the SPOT image, is just visible at the top of the aerial photograph. The cereal crops have been harvested at this stage, and the remaining green areas are fields of sugar beet.

It can be seen that there are quite large variations in the soil brightness throughout the site, caused by differences in soil composition and drainage efficiency. To compensate as far as possible for any effect this variability may have on crop density and/or structure, as many samples as possible were taken from within each field. Data were also collected from several fields during each visit to minimise the possibility of selecting an atypical field of any particular crop. The planned temporal sampling (a full set of complementary measurements every two weeks) was not achieved due to the inclemency of the weather during the growing season of 1997. This resulted in a lack of sampling between early May and early June.

3.1 Summary of data, collection methods and validation

The field data, collected on eight visits to the Barton Bendish field site between 24th April and 6th August 1997, consist of measurements of a number of properties including:

  • Ground spectro-radiometric measurements.
  • Canopy coverage estimates.
  • LAI measurements.
  • Canopy height, and plant and row spacings.
  • 3D plant structure, for BPMS modelling, characterisation of LAD.
  • Airborne data (NERC ARS ATM, CASI data and aerial photography).

Data collection methods are described in detail below, in addition to validation studies performed in order to provide confidence in the measured data. This is primarily a comparison of 3D barley and wheat canopies reconstructed from measured plant parameter data using the BPMS, with properties measured in the field such as LAI, LAD and BRDF. The ability to characterise the primary structural properties (and the consequent radiometric behaviour) of the measured canopies within the BPMS is fundamental to this thesis. Much of the following work deals with analysis of BPMS-simulated canopy reflectance. Establishing agreement between reflectance behaviour simulated using measured plant parameters within the BPMS, and the reflectance behaviour of the real canopies, permits conclusions drawn from analysis of BPMS results to be applied to real data. The canopy structural data collected through manual measurement are first described, followed by various other measurements collected for validation of the BPMS data and for subsequent comparison with information derived from airborne data.

3.1.1 Plant structural measurements

Plant and row spacings for each field were recorded along with the planting densities of the various crops i.e. the number of plants per metre[1]. In addition, detailed descriptions of the topology and structure of individual plants were made for use within the BPMS. These measurement techniques are discussed in detail by Lewis and Muller (1990), Lewis and Disney (1997) and Lewis (1999). Typically five or six plants per field per date were measured. A set of ten measurements were made of each plant, with the 'leaf' subset of these measurements being made for every leaf of a plant. Brief details of the measurements are as follows:

For each tiller:

  • tiller azimuth angle
  • tiller zenith angle

For each stem section (emergent from ground, and between leaf nodes):

  • stem length
  • stem width

For each leaf:

  • leaf base zenith angle
  • leaf tip zenith angle
  • leaf length
  • % leaf length to maximum width
  • (maximum) leaf width

Individual leaf angles (base, tip) and stem section angles were measured with a protractor to an estimated precision of 5o. Leaf length, maximum width and stem lengths were measured to an estimated precision of 1 mm. Figure 3.2 shows a measured barley plant. Sensitivity analysis was carried out to test whether simulated BRDF is sensitive to any particular BPMS measurement. The measured parameters were perturbed randomly according to a Gaussian distribution of specified mean and standard deviation and BRDF was simulated for each 'perturbed' canopy. Simulated BRDF was found to be more sensitive in shape to row and plant spacing than any of the individual structural measurements, and more sensitive in magnitude to the specified leaf and soil reflectance values. Figures 3.3 and 3.4 show simulated canopies using BPMS measurements.


3.1.2 Ground spectro-radiometric measurements

Measuring BRDF in the field is not trivial, largely because of variable atmospheric conditions and hence varying quantities of diffuse and direct illumination (Sandmeier and Itten, 1999). Other uncertainties are caused by the difficulty of determining precisely what the instrument IFOV is seeing, movement of crops due to wind, lack of a stable instrument base etc. Although no atmospheric scattering data were available during this field campaign (sun photometer measurements are now made at Barton Bendish to augment field measurements), measurements were recorded rapidly and repeated often to try and minimise the effects of changing atmospheric conditions. In addition, BRDF measurements were restricted to 'blue sky' days. However, these can be few and far between in a typical UK summer and 1997 transpired to be particularly wet. However, samples of directional reflectance were made with a boom-mounted ASD PS-II radiometer (www[3.1]) in the visible to NIR wavelength range (due to limitations in the instrument, effectively, 450-900 nm). Measurements were made at 10o intervals in view zenith (v) from –70o to +70o in both the principal and cross-principal solar planes[2]. The relative azimuth angle, , between v and the illumination zenith, i, is then 0o or 90o for viewing in or across the principal plane. This viewing and illumination geometry is illustrated in figure 3.5. In addition to the target radiance measurements, near-simultaneous radiance measurements of a polythene (assumed) Lambertian reference panel were made. Reflectance is then calculated as the ratio of the target radiance to that of the reference panel (see equation 2.2). For each reference panel measurement, the so-called dark current response of the instrument was also recorded. This is a measurement of the signal recorded when no light is incident on the radiometer, caused by background (thermal) noise in the silicon detector. The dark current is subtracted from the measured radiance to provide an accurate measurement of target radiance. An example of measured barley BRDF is given in figure 3.6, along with reflectance simulated using the manually measured canopy structural parameters within the BPMS.

Figure 3.6 Measured and simulated canopy reflectance at four bands.

Agreement between the modelled and measured reflectance data is reasonable at visible wavelengths, but not so in the NIR (850nm). Whilst the directional variation of simulated reflectance may be reasonable, the magnitude is generally too low. This is a result of reflectance being simulated without multiple scattering. As discussed in section 2, the multiple scattered component of canopy reflectance tends to be isotropic, and will therefore increase the magnitude of simulated canopy reflectance without significantly changing the shape). In addition, the level of noise in the measurements can be seen. This is particularly obvious at nadir where three measurements were made, one at the start of the string of measurements, one in the middle and one at the end. The variability of these measurements (up to 10% reflectance) is caused largely by changes in atmospheric conditions during the measurement period, even on a 'blue sky' day with little or no cloud. This is a typical problem when performing field radiometric measurements and is discussed at the end of this chapter, as well as in chapter 8. Values of r2 for linear regression of the measured against modelled reflectances are given in table 3.1.

Table 3.1Values of r2 for comparison of measured to modelled reflectances.

r2 values
canopy / 450nm / 550nm / 650nm / 850nm
ba_112 18/04 / 0.82 / 0.84 / 0.87 / 0.68
ba_119 13/05 / 0.94 / 0.95 / 0.89 / 0.74
ba_112 04/06 / 0.90 / 0.96 / 0.95 / 0.80
ba_119 24/06 / 0.79 / 0.84 / 0.80 / 0.71
ww_109 23/03 / 0.78 / 0.80 / 0.78 / 0.64
ww_109 23/04 / 0.92 / 0.95 / 0.89 / 0.83

The figures in table 3.1 demonstrate the generally good agreement between the measured and modelled canopy reflectance values. The values of r2 are high enough to conclude that, given the uncertainty in the measured reflectance values due to changing atmospheric conditions and the small IFOV, the BPMS-simulated reflectances are an accurate representation of the real canopy reflectance, both in shape and in magnitude. The cases where agreement is not so good are likely to be due to the canopy being very sparse (LAI <1). The agreement between the angular variation is higher than the spectral variation. Figure 3.7 illustrates this. Overall agreement between the BPMS and measured reflectance values is generally good in the visible but the variance of the NIR reflectance values is clear. The signal-to-noise ratio (SNR) of the detector in the region beyond 900nm is very much lower than in the visible part of the spectrum.

Figure 3.7 Scatter of BPMS against measured canopy, barley 13th May 1997.

3.1.3 Canopy coverage estimates

Nadir-pointing photographs were taken (from a height of 1m above the ground) of the various canopies throughout their development. No photographs were taken during the later stages when the canopy heights exceed 1m as there was total coverage in these cases. A series of canopy cover photographs showing the progression of cover in barley field 119 are shown in figure 3.8. The variability of local planting density (even in such a small region) is clear in the 13th May image. The camera tripod is visible at the lower extent of the first two images.

Canopy cover was derived from the photographic data by estimating the proportion of visible soil per unit ground area (Disney et al., 1998). Only the central portion of each image was used to avoid image plane distortions caused by the relatively wide field of view (115o). RGB to HSI transformations were performed on the scanned images, which were then thresholded manually on hue. The mid-range hue values corresponded closely to soil, while the higher values corresponded to vegetation. This allows percent cover to be simply estimated. Values of % cover as estimated from the photography are presented in figure 3.9 in the form of a scatter plot with corresponding BPMS derived values. The error bars in figure 3.9 represent the variability of % cover estimates made from photography (one SD). The regression of all samples together shows reasonably high correlation, with a slight offset (overestimation) of around 10% in the values derived from photography. This is due to inaccuracy in the thresholding of the HSI-transformed photography. Relatively small changes in the value of the selected threshold cause significant changes in % cover. Conservative threshold values were chosen deliberately to make sure all vegetation was included.

Figure 3.9 %cover estimated from photography and BPMS simulations.

3.1.4 LAI measurements

LAI measurements were made with a LI-COR LAI 2000 instrument, which works by comparing the intensity of (diffuse) incident illumination measured at the bottom of the canopy with that arriving at the top (LI-COR technical report; Welles and Norman, 1990). Incident light is recorded over five concentric angular rings, each of approximately 15o in width (giving a nearly hemispherical field of view). LAI is estimated by calculating the probability of a photon penetrating to a depth z in the canopy (under various assumptions regarding the arrangement and radiometric properties of scattering elements in the canopy), and comparing this with the measured radiance at the bottom of the canopy. The probability of non-interceptance at depth z within the canopy is described by Beer’s law (equation 2.19). The angular integral of this property (over all zenith angles) is approximated as a weighted summation over the five concentric angular rings of the instrument.

The manufacturer's recommendations were followed in deciding a measurement plan. One measurement was made at the top of the canopy followed by ten measurements at different locations below the canopy. This pattern was repeated three times per field, and the resulting thirty-three samples comprise one full set of measurements. Measurements were made following transects at an angle of between 30o and 45o to the row direction, and at intervals suitably spaced to avoid over or under-estimation of LAI caused by measuring at integer numbers of rows or inter-row gaps. Care was also taken with regard to the illumination conditions under which measurements were made. The instrument operates under the assumption that the canopy is illuminated by a diffuse source, and therefore should be shadowed from any direct sunlight. Due to the weather during the fieldwork campaign it was possible to make the majority of the measurements under diffuse conditions. On days with significant direct sunlight a 270o view cap was used to mask out both the operator and sun from the viewing hemisphere. Measurements were also made as late or early as possible in the day when penetration of the canopy by direct sunlight is minimised. Measured values of LAI for the barley and wheat canopies are plotted as a function of day of year (doy) in figure 3.10. Also shown are estimates of LAI derived from BPMS canopies generated from the manually measured canopy structural data. These values are presented explicitly in table 3.2.

Figure 3.10 Measured and modelled LAI for barley and wheat.

It can be seen that there is a wide spread of measured LAI values for some canopies during development, resulting in a relatively poor agreement between the measured and modelled values in some cases. The likeliest cause of this divergence is the intra-field variability of measured LAI: the coverage of the barley canopy of 13th May (ba_119) varied from 10% to 70% across the field.

Table 3.2 Values of LAI derived from BPMS canopies.

canopy / LAI
ba_112 18/04 / 0.70
ba_119 13/05 / 3.89
ba_119 04/06 / 3.16
ba_119 24/06 / 3.19
ww_23/03 / 0.09
ww_23/04 / 2.25

3.1.5 Airborne data

A variety of airborne data were collected by the NERC ARS aircraft over the Barton Bendish field site during the 1997 field campaign. Due to the weather, the dates on which data could be collected were not ideally spaced through the growing season (see below). In addition, atmospheric conditions were not ideal when airborne data could be obtained (clouds and/or haze). However, contemporaneous ground measurements of LAI and BRDF described above were made. The nature of the data collected by the NERC aircraft is as follows:

  • Daedalus 1268 Airborne Thematic Mapper (ATM) data – radiometer measurements in 11 broadband channels in the visible/NIR, SWIR, and TIR. Parallel flight lines were flown over the scene to provide overlap which gives BRDF sampling in the central region of overlap due to the wide (70o) FOV of the sensor (Barnsley et al., 1997). The data were recorded at an altitude of 3000m giving a spatial resolution of around 5m. Forty three flight lines were flown in total, on three dates: 5/6/97, 2/8/97 and 6/8/97, each of which was accompanied by contemporaneous ground measurements of the properties described above. An ATM image of the field site is shown in figure 3.11.
  • Compact Airborne Spectrographic Imager (CASI) data - hyperspectral radiometer data, consisting of 288 channels in the visible/NIR, obtained concurrently with the ATM data. Forty two flight lines were flown in total, on the dates given above, with a spatial resolution of approximately 3 m.
  • Colour aerial photography - obtained concurrently with the ATM and CASI data. The resolution of the photography is approximately 2m (see figure 3.1).

3.2 Summary of field measurements and validation

The fieldwork described above was designed to provide information permitting validation of simulations of canopy. In particular, measurements of structural and radiometric canopy properties were designed for comparison with modelled properties derived from a full 3D (BPMS) structural model of the canopy. A number of general difficulties of conducting such fieldwork were encountered, in addition to specific problems caused largely by the weather conditions of summer 1997, which was an unusually wet summer. In general, the BPMS estimates of canopy structural parameters agreed relatively well with the measured parameters. Given the inter- and intra-field variability of the crops agreement between measured and modelled values of %cover and LAI were generally good. Agreement of the measured and modelled values of canopy were not always so good but this is largely a function of the changing atmospheric conditions and the uncertainty of what is being viewed through the radiometer IFOV (to better characterise this variability, simulations should be performed using an equivalent size IFOV). Even a small wind can move the canopy enough to change this dramatically, particularly at low canopy LAI when soil can be very visible. Given these constraints, it was concluded that the BPMS characterisation of canopy structure was more than sufficient for the experimental work described in succeeding chapters.