3.2.1. Glasshouse Study to Prepare Soils with Different Root Densities
Topsoils (0-10 cm depth) were collected from two soil types under permanent pasture: (1) Ramiha silt loam (Andic Dystrochrept, USDA NRCS classification); (Typic Allophanic Brown soil, New Zealand classification (Hewitt 1998)) derived from mixtures of greywacke of loess and andesitic ash, and (2) Manawatu fine sandy loam (Dystric Fluventic Eutrochrept, USDA NRCS classification) (Weathered Fluvial Recent soil, New Zealand classification (Hewitt 1998)) a recent soil derived from greywacke alluvium (Cowie 1978). Soils were air dried and sieved to less than 5 mm aggregate size, and coarse pasture roots were removed and each soil was remixed thoroughly to give homogenous samples.
The soils had markedly different soil chemical and physical characteristics (Table 3.1). The Ramiha soil, containing allophane, had a higher organic matter content and soil P retention than the Manawatu soil but had a lower Olsen extractable P status and bulk density.
Sixty 450 g subsamples of each air-dried soil were weighed into plastic bags. One of 20 combinations of N and P fertilizer treatments (N0P0, N0P1 --- N3P4) were applied
and mixed thoroughly with the soil in each bag. Each treatment was replicated 3 times for each soil. N treatments were 0 (N0), 111 (N1), 222 (N2) and 444 (N3) mg N kg-1 soil
and P treatments were 0 (P0), 111 (P1), 222 (P2), 333 (P3), and 444 (P4) mg P kg-1 soil.
(containing 100 mg K and 41 mg S) was added as a basal treatment. Then, 40 seeds of Moata ryegrass (Lolium multiflorum Lam., commonly used as a winter growing pasture in New Zealand) were mixed with the surface soil of each pot. After germination, the 15 strongest plants were grown for 72 days. Pots were watered regularly to a weight that represented 80% of water capacity of the soil in the pot. At 72 days selected pots were taken for root density measurements.
Table 3.1 Chemical and physical characteristics of the two soil types. Characteristics1 Ramiha Manawatu
C (% of dry weight) 6.3 3.0 N (% of dry weight) 0.57 0.29 P (% of dry weight) 0.09 0.88 P retention (% of total P) 62 14 Ca (me 100 g-1) 1.66 6.27 Na (me 100 g-1) 0.22 0.48 K (me 100 g-1) 0.21 1.05 Mg (me 100 g-1) 0.50 1.70 CEC (me 100 g-1) 26.58 28.49 Base saturation (%) 10 33 Olsen extractable P (mg P l-1) 12 30 pH 5.26 4.86 Bulk density (g cm-3) 0.76 0.92 Texture Silt loam Fine sandy loam
[1Analytical methods in Blakemore et al. (1987)]
3.2.2. Contact Probe Modification and Spectra Acquisition
A prototype soil reflectance probe was developed which was based on a plant contact probe (ASD FieldSpecPro, Boulder, CO.); an internal light source was replaced by a greater intensity parabolic-reflector halogen lamp (4.5 watt). A round casing was developed to avoid direct contact of the quartz probe window with the soil, to exclude the ambient light, and to provide a fixed distance (30.5 mm) between the object (soil surface) and the probe window (Figure 3.1c). The object was rotated through 360o, to give a field of view of 561 mm2 (Figure 3.1c).
Soil reflectance measurements were recorded from a flat-sectioned, horizontal soil slice (1.3 cm depth). The instrument records spectra from 350-2500 nm with a sampling interval of 1.4 nm for the region 350-1000 nm and 2 nm for the region 1000-2500 nm. The data processing software associated with the FieldSpecPro spectroradiometer interpolates the 1.4- and 2-nm-spaced data to produce 1-nm spaced data.
3.5 cm slice A discarded
1.3 cm slice C 1.3 cm slice B
(a) (b)
(c) (d)
Figure 3.1 (a) Soil removed from pot, (b) sliced with a spacer ring and microtome blade, (c) spectral reflectance and (d) roots separated from soil with tap water.
3.2.3. Root Density and Spectral Measurement
Out of a total of 120 pots, 57 were selected on the basis of large shoot yield differences (0.81 to 2.91 g DM pot-1) resulting from the range of N and P treatment combinations. Thirty samples were selected from the Ramiha soil and 27 samples from the Manawatu soil. It was expected that greater shoot growth would be associated with greater root growth.
The top 3.5 cm section (slice A) of soil from each pot was sliced using a microtome blade and discarded (Figure 3.1a and 3.1b). Then the reflectance was recorded at the freshly sliced surface (slice B) using the modified contact probe attached to the portable spectroradiometer (ASD FieldSpec Pro, Boulder, CO.). From immediately below the freshly cut surface, a 1.3 cm thick soil slice (slice B) was harvested and weighed. Roots were separated from the soil slice B by washing the soil slice with tap water through a sieve stack starting at 1000 μm followed by 500 μm, 400 μm, 350 μm and 300 μm diameters (Figure 3.1d). The root mass retained on the sieves was bulked and dried in an oven at 50oC for 3 days.
A third 1.3 cm soil slice (slice C) was taken, weighed wet and then weighed after drying to constant weight in an oven at 105oC. The average wet bulk densities for the Ramiha and Manawatu potted soils were calculated from the C slices. This wet bulk density was used to convert the wet weight of slice B into a volume (cm3). The root mass in slice B was expressed as a density, mg dry root cm-3 soil.
3.2.4. Spectral Measurement of Standard Root Contents
Air-dry soil (sieved < 2mm, Ramiha) and oven dry root (50oC until constant weight) were prepared and mixed to give percentages of root content of 0, 1, 5, 10, 25 and 100%. Air-dried sieved soil was used to minimize the effect of variable water content on the spectral reflectance and to provide a standard basal root content. Surfaces of the air-dry soil/root mixtures were prepared by pressing samples (1-cm thick) into a plastic petri dish (10 cm diameter). The spectral reflectance of the flat surface was recorded using the modified contact probe attached to the FieldSpecPro spectroradiometer.
3.2.5. Continuum Removal
Continuum removal was applied to the smoothed spectral data of the standard root contents. This is a procedure that facilitates making spectral curves distinction or easier to compare (Clark and Roush 1984). The depth of an absorption band, D, is usually defined relative to the continuum, Rc:
Rc Rb D=1−
where Rb is the reflectance at the bottom (trough center point) of a band and Rc is the reflectance of the continuum at the same wavelength as Rb (Clark and Roush 1984). The depth of an absorption is related to the amount of the absorber. This approach has a powerful tool for enhancement and separation of small but often significant differences of bands of particular functional groups.
3.2.6. Spectral Pre-Processing and Data Analysis
Ten replicate reflectance spectra were acquired from each soil slice using the purpose-built contact probe (Figure 3.1c). Before statistical analysis, reflectance spectra were pre-processed using SpectraProc V 1.1 software (Hueni and Tuohy 2006). After a number of iterations the following pre-processing steps were adopted as standard: elimination of noisy data at wavebands 350-470 and 2440-2500 nm, followed by spectral smoothing using a Savitzky-Golay 4th polynomial order filter (Shepherd and Walsh 2002) with a window size of 33 nm. After smoothing, the data were reduced by taking every 10th waveband, the first derivative calculated and finally the ten replicate first derivatives were averaged. The first derivative data were imported to Minitab 14 (MINITAB Inc. 2003) for principal component analysis (PCA) and partial least squares regression (PLSR) analysis against the reference analytical data (root density). A PCA score plot was used to observe the pattern of sample scattering. During PLSR processing, samples which had a standardized residual of > 2.0 were removed as outliers (MINITAB Inc. 2003). The accuracy of the models was tested internally using a leave-one-out cross-validation method.
3.2.7. Regression Model Accuracy
The ability of the PLSR model to predict root density was assessed using the following statistics. RMSE (root mean square error), which is the standard deviation of the difference between the measured and the predicted root density was calculated: that from calibration data is called root mean square error of calibration (RMSEC) and from cross-validation is called root mean square error of cross-validation (RMSECV);
(
)
N y y RMSECV =∑
m − cv 2where ym is the measured laboratory value,ycv is the value predicted from the PLSR model, and N is the number of samples. Coefficient of determination (r2) is the proportion of variability in a data set (measured and predicted value) that is accounted for by a statistical model. RPD (ratio of prediction to deviation) is the ratio of the standard deviation of the measured soil properties to the RMSECV. RER (ratio error range) is the ratio of the range of measured values of root density to the RMSECV.
( )
RMSECV y STDEV RPD= m( )
( )
RMSECV y Min y Max RER= m − mThe best prediction model is shown by the largest RPD, RER, r2 and the smallest RMSECV.