Predicting herbage mass of Phleum pratense L . pastures with a disk meter

A simple disk meter was calibrated for predicting the herbage mass (HM) of rotationally grazed pastures dominated by Phleum pratense L. in 1991-1992. A total of 696 paired observations were made on disk height (DH) and HM > 4 cm. The samples were classified for three different statuses: spring growth, post grazing and aftermath. In 1991 three different disk weights (c. 3.5, 5.0 and 6.5 kg/m 2 ) were compared and the lightest disk was chosen for further studies in 1992. The variation in HM was adequately explained by linear models. There were only small differences between the predictive ability of different disk weights, the lightest disk having the highest r 2 values. Post grazing and aftermath samples could be pooled, whilst spring growth samples needed a separate model. Year had no significant effect on the parameters of any model. The models chosen were: spring growth HM = -406.7 + 113.4(DH); r 2 = 0.95, post grazing and aftermath HM = -629.1 + 122.1(DH); r 2 = 0.88. The disk meter is a potential tool for predicting the HM of rotationally grazed timothy pastures.

ntroduction Herbage mass (HM; kg ha -1 ) or an expression reflecting it is an essential factor in interpreting the animal response in grazing experiments.HM is also needed for understanding grassland re- sponses to different management practices.Due to the considerable variability within a grazed sward, a large number of direct samples are of- ten needed to get reliable HM estimates.To over- come the need for large sample numbers and to find nondestructive methods for HM estimation several procedures have been developed, e.g.visual estimation, determination of sward height with a ruler or a disk meter, measurements with a capacitance meter and spectral analysis (Frame  1981, Burns et al. 1990).The disk meter is cheap and both simple and quick to use.A number of studies have indicated good relationship between HM and disk meter readings (e.g.Powell 1974,  Castle 1976, Bransby et al. 1977, Griggs and  Stringer 1988, Mould 1990).Since the disk meter is easy to construct (e.g.Castle 1976), its price will depend on the components available; costs are minimal in any case.According to Castle (1976)  and Bransby et al. (1977), 50 readings can be taken in 10-15 minutes from a paddock 1.6-2.5 ha in size.Even including the time for calibration, the disk meter is still preferred to the clipping method when estimating the HM of large areas (e.g.Vartha and Matches 1977,  Earle and McGowan 1979, Griggs and Stringer   1988).
Finnish pastures differ from those reported in previous disk meter experiments in terms of species composition, sward structure and man- agement practices.The most common species are timothy (Phleum pratense L.) and meadow fes- cue (F.pratensis Huds.); cocksfoot (Dactylis glomerata L.) is used to some extent.Due to species composition, rapid spring growth, rapid generative development during the long days of summer and the rotational grazing system, tiller number remains low, from 2000-2500 (Huoku- na 1964) to 8000 tiller nr 2 , with large variations within a single paddock.The leaf area index (LAI) also remains relatively low (Virkajärvi, manuscript).The height of the stand of pregrazed swards is usually higher than that reported in most previous disk meter studies.Thus, it was necessary to clarify the prediction ability of the disk meter on rotationally grazed pastures dom- inated by timothy, Phleum pratense L., under Nordic conditions.First, different disk weights were compared to establish the most suitable downward pressure for timothy.The stability of calibration in the course of the grazing season and between grazing seasons was studied in the following year.

Material and methods
The disk meter was calibrated by developing re- gression equations between the meter readings (cm) and herbage mass (kg ha~' dm > 4.0 cm) at the Karelia Research Station in 1991 and 1992.The pastures, which were rotationally grazed by suckler cows, contained 80-90% timothy and 0-20% meadow fescue.The amount of dicotyledonous weeds was insignificant.The pastures were fertilized with 170-180 N, 20-40 P and 140-210 K depending on the soil properties.Climatic data were recorded at the Karelia Research Station.In all, 18 series of calibrations were made between 30th May 1991 and 15th September 1992.The calibration included spring growth, post-grazed swards and aftermath, later referred to as field status.
Comparison of three disk weights with accuracy of the method The disk meter was constructed as described in Castle (1976).It consisted of a plastic rod (1100 mm long, 22 mm diameter) and two free- ly sliding aluminium plates (1 mm thick) joined together.The diameter of the lower aluminium disk was 300 mm.Additional brass weights were used to give weight of 247, 354 and 458 g to correspond to weights ofc. 3.5, 5.0 and 6.5 kg/m 2 .
The rod was graduated in 1 cm intervals.For measurements of HM, the meter was held upright and then pushed into the vegetation.The horizontal plate was raised by the vegetation, and the settling height was read from the rod with 0.5 cm accuracy.
The available HM was determined by cutting 0.225 m 2 quadrats to a height of 4.0 cm with elec- tric garden shares.Before the cutting, three disk meter readings (DH) were taken ofeach quadrat at each disk weight and the mean was pooled against HM.In addition, extended height (SH)  was measured with a ruler, and sward density (DEN) was determined visually as percentages of full tiller density.The dry matter content (DM) was determined by force-drying samples at 100°C for 20 hours.In order to describe the ma- terial more precisely, a representative sample was collected on each occasion, and crude protein (CP) and crude fibre (CF) were determined at the Central Laboratory, Agricultural Research Centre, Jokioinen, according to standard procedures.
The data were split according to Neter et al.  1989, i.e. about two-thirds of the observations were used to develop estimation equations (estimation data set) and the remaining third were used for validation (validation data set).For the comparison of disk weights made in 1991, HM prediction equations were developed from the data of calibrations 1,3, 4,5, 7 and 9.The re- maining calibrations (2, 6 and 8), one of each field status, were used for validation.
The HM prediction models were developed by SAS GLM and REG procedures (SAS Insti- tute 1985) on the basis of residual diagnostics (Henderson and Velleman 1981), r 2 and residual standard deviation (RSD).For the comparison of disk meter weights, the original HM model had field status (ST), calibration (CAL) and DH and all possible interactions of these as varia- bles.The terms which were not significant in the regression model at the 0.05 level were excluded stepwise.Quadratic models (Y = DH + DH 2 ) were studied to test curvilinearity.Logarithmic (log (Y) = DH) models were also studied to sat- isfy the equal variance assumption.
The prediction equations developed were applied to the validation data, i.e. the remaining third of observations.The accuracy of prediction models was studied by comparing the predicted values of HM with their corresponding clipped HM values in terms of r 2, and the stand- ard error of validation (SEV).SEV is defined as: [X(Y.-'£) 2 /n] l/2 (Griggs and Stringer 1988).

Accuracy of light disk in 1991-1992
To determine the accuracy of the disk meter over the years, measurements were continued in 1992 with the lightest disk only.To ensure the representativeness of the estimation data set, the samples were classified in the field into weak, aver- age and strong vegetation, including extreme sites in estimation data set (Bransby and Clarke  1988).The proportions of each class in the esti- mation data set were approximately the same.The original HM model had year (YR), ST, CAL and DH and all possible interactions of these as variables.The procedure continued as described previously.

Results
The weather conditions are presented in Table 1.There were marked differences in growing sea- sons: 1991 was extremely wet but in 1992 the pastures suffered from drought.
HM and SH varied a great deal during the Field status: SPGR = spring growth, POSTG = post grazing, AFT = aftermath.

Effect of disk weight on accuracy of the method
The variation in HM was adequately explained by linear models.The data on after grazing and aftermath samples could be pooled, but those on first-growth samples had to be analysed separately.The results of the comparison were published as a poster at European Grassland Feder- ation's XIV General Meeting 1992 (Virkajärvi  et al. 1992).The differences in accuracy of the 400 AGRICULTURAL SCIENCE IN FINLAND 'Model terms: HM = herbage mass (dry matter kg/ha > 4 cm), DH = disk height (cm), n = number of observations used for calibration and validation, SE = standard error of estimate, RSD = residual standard deviation, SEV = standard error of validation (defined as: [Z(Y -Y.) 2 /n] l/2 (Griggs and Stringer 1988).
method between three disk weights were small (Table 3).The effect of disk weight was strongest in spring growth.The light disk (L) was slightly the most reliable in terms of r 2 and SEV on all occasions and was therefore chosen for sampling in 1992.
od was good in spring growth and slightly poorer in post grazing and aftermath.

Discussion
Accuracy of light disk in 1991-1992   The original HM model had YR, ST, CAL and DH and all possible interactions of these as var- iables.Year and calibration or their interactions were not significant (p > 0.05) terms in the mod- el.As in the case of different disk weights, the variation in HM was adequately explained by linear models.Use of quadratic instead of the linear models did not markedly improve r 2 val- ues (from 0.948 to 0.952).The data on after graz- ing and aftermath samples could be pooled, but the first growth samples had to be analysed separately.The HM estimation models chosen are presented in Table 4.The accuracy of the meth- Effect of disk weight on accuracy of the method The downward pressures used, 3.5-6.5 kg/m 2 , were rather average when compared with those reported in other studies, in which pressures have varied from 2.9 (Castle 1976, Mould 1992) to 15 kg nr 2 (Bransby et al. 1977).Downward pres- sure did not affect the accuracy of the method but only the slope of the regression line.Here, timothy pasture resembles tall fescue (Bransby  et al. 1977).The light disk gave a slightly more accurate result than the other disks.Thus it would seem sensible to study lighter rather than heavi- er disks, as also suggested by Mould (1992).'Model terms: HM = herbage mass (dry matter kg/ha > 4 cm), DH = disk height (cm), n = number of observation, SE = stadard error of estimate, RSD = residual standard deviation, SEV = standard error of validation (defined as: [I(Y-Y) 2 /n] l/2 (Griggs & Stringer 1988).

Accuracy of the light disk meter
In terms of accuracy of the method or residual deviation, the variation in HM was adequately described as a linear function of DH, although the residuals tended to increase as DH increased (Fig. 1).Logarithmic models were not suitable.
Although there was a slight tendency to curvi- linearity in the validation set of spring growth samples (Fig. la.), the quadratic model did not markedly improve the r 2 value (less than 1% unit).This is probably due to the robustness of the linear model and the sensitivity of the quadratic model to differences between the estimation and the validation data set.The relationship between DH and HM remained linear even though the height of the stand was above the normal range for pastures.From the literature it can be concluded that linear functions have been used when the DH values have been relatively low, but at high DH values (> 40 cm) a polynomial function has been more accurate (Bransby  et al. 1977, Baker et al. 1981).It is clear that when the average height increases, the risk of lodging also increases, thus causing considera- ble error.The method may not, therefore, be suitable for estimating, say, silage yield if lodging is abundant.The accuracy of the method is also strongly impaired if the grass is trampled or if the soil surface is uneven.
The r 2 values found here were relatively high, which may be attributed to the uniform botanical composition (up to 90% of timothy).SEV, too, was low, 345 and 240 kg ha -1 DM (Table 4), but because oflow HM, the coefficient of variation was high.This may restrict the usefulness of the method, especially when measuring HM after grazing.However, the cutting height, 4 cm, lowered the HM values observed compared with those for clipping to ground level.Density correlated only very weakly with HM, This could be partly due to the fact it was estimated visually, but the correlation was low with more accurate methods, too (Urioste 1984,  Griggs and Stringer 1988).It is interesting to note that DH correlated clearly better with SH (r = 0.92, p < 0.0001) than with DEN (r = 0.14, p<0.0001).The inclusion of density in the HM prediction models did not improve the accuracy of prediction.
According to several researchers, the disk meter should be calibrated several times during the growing season and again in different years (e.g.Bransby 1977 et al., Vartha and Matches  1977, Bryan et al. 1989, Mould 1990) because of changes in the botanical composition of the sward and differences in vegetative and genera- tive growth type.Here, the parameters of the spring growth model differed from those of post grazing and aftermath.Other differences in the course of the growing season were not found, possibly because botanical composition was uni- form throughout the growing season.Moreover, the cutting height of4 cm may have affected the results,as herbage often accumulates under the grazing horizon in the course of the grazing sea- son.The nearer the sampling height and the low- er limit of the actual grazing horizon are to each other, the less effect this accumulation should have.
Uniform botanical composition is the most probable reason for there being no statistical dif- ferences between years, even though 1991 was extremely moist whilst 1992 was dry, and swards clearly suffered from drought.Thus turgor pressure of the leaves or plants appears to have had only a slight effect on the parameters of the re- gression line.Many researhers have concluded that there is no uniform calibration equation (Griggs and Stringer 1988, Bryan et al. 1989,  Gonzalez et al. 1990).Although in this study sta- bility was good under very different moisture conditions, the disk meter should still be calibrated for each occasion if the results are to be expressed as HM.Depending on the purpose of the determinations, the calibration could be done according to Bransby and Clarke (1988) or us- ing the simplified version described by Stock- dale (1984).

Conclusions
The disk meter is accurate enough for predicting the HM of timothy pastures of uniform bo- tanical composition for specific purposes when large paddocks are to be evaluated.Such purposes might be estimation of the effects of over- wintering damage on growth in different man- agement practices or estimation of sward growth rate; it can also be used as a tool for management decisions in grazing trials.The disk meter cannot be used to predict the HM of a sward in the event of disturbances in canopy height, e.g.lodging or the presence of old stubble or tram- pled grass.Sward density is ofminor importance to meter readings.

Table 1 .
Monthly mean temperatures and precipitation during growing seasons 1991 and 1992 and average long-term values.

Table 2 .
Date, number of paired observations, field status, yields of clipped samples, extended sward height and crude fibre content for 18 series of calibrations.

Table 3 .
Selected available HM prediction models' for three disk meters, light (L), medium(M)and heavy(H).