Effect of botanical composition calibration on the accuracy of undisturbed sward height and comparative yield method techniques for herbage mass estimation in tropical heterogeneous pastures

This study aimed to evaluate the influence of botanical composition calibration on the accuracy of undisturbed sward height and comparative yield method (CYM) techniques for herbage mass estimation in tropical heterogeneous pastures. Two studies were conducted using two grazing systems based on heterogeneous pastures. Herbage mass was estimated using CYM and undisturbed sward height techniques within quadrats (SHQ) or at a set of standard points in the paddocks (SHP). SHQ had a higher adjusted R2 when calibrated with the Urochloa brizantha cv. Marandú (U. brizantha) proportion compared to the simple SHQ model (0.75 vs 0.68), while RSE was lower (0.18 vs 0.21). The R2 increased (0.63 to 0.68), while both residual means and RSE decreased (−2.30 to –0.05 and 0.22 to 0.20, respectively) when SHP was calibrated with U. brizantha. It also resulted in a reduction of mean squared prediction error (MSPE) to CYM, SHQ, and SHP, respectively. The CYM allows for higher accuracy of herbage mass estimation in tropical heterogeneous pastures than undisturbed sward height, irrespective of botanical calibration. However, the botanical composition calibration had positive effects on the models from undisturbed sward height, and the calibration with U. brizantha increased the accuracy of herbage mass prediction.

Herbage mass is defined as the total weight of dry matter per unit area at a specific time.It is considered one of the most important variables in determining the appropriate stocking rate that pastures can support at a given time without resulting in adverse effects related to overgrazing and soil degradation (Li et al. 2016).Regular herbage mass evaluation is recommended to obtain a higher utilisation efficiency of pastures and to allow optimal animal performance (Klootwijk et al. 2019).
During the last 70 years, several methods aimed at reducing labour requirements and increasing suitability on a farm scale have been developed to estimate herbage mass more accurately (Edvan et al. 2016).Therefore, different methods have been developed that involve mathematical relationships between indirect or non-destructive techniques and herbage mass harvesting (Gonzalez et al. 1990).For example, the comparative yield method (CYM) described by Haydock and Shaw (1975) estimates the herbage mass from the relationship between standard and visual points ranked relative to the standard points using a linear regression.The CYM has been demonstrated to increase the accuracy of herbage mass estimates and reduce labour and is widely recommended for both homogeneous and heterogeneous grasslands based on tropical and temperate pasture species (Martin et al. 2005).However, this technique requires trained observers to obtain accurate herbage mass estimates.In contrast, indirect methods based on determining undisturbed sward height may be less laborious and more practical in farm conditions.However, these methods require frequent calibration, and their accuracy in obtaining good herbage mass estimates may be affected by botanical composition (George et al. 2006).Therefore, botanical composition calibration could be important for herbage mass estimation in tropical livestock systems, where pastures predominate with high heterogeneity (Boval and Dixon 2012).Thus, this study aimed to evaluate the influence of botanical composition calibration on the accuracy of undisturbed sward height and comparative yield method (CYM) techniques for herbage mass estimation in tropical heterogeneous pastures.
The study was conducted at San Roque, Colombia (6°29′8.38″N, 74°50′9.95″W), which has an average annual rainfall of 2 200 mm, average temperature of 24.1 °C and an altitude of 801 m above sea level.The study was carried out in 2019 (from March to October) and 2020 (from May to November) on two grazing systems of 6.1 (grazing system 1) and 2.0 (grazing system 2) ha, each consisting of eight (4 253 m 2 ) and four (1 332 m 2 ) paddocks, respectively.Both grazing systems were based on heterogeneous pastures (Table 1), characterised by sandy loam soil with a clay texture, pH of 5.3, 2.3% organic matter and an average terrain slope of 10.3% (grazing system 1) and 31.6%(grazing system 2).During April and October (early wet season) of each evaluation year, dolomitic lime (1 000 kg ha −1 year −1 ) and urea fertiliser (equivalent to 125 kg N −1 ha −1 year) were applied to each grazing system.The grazing management was based on rotational stocking, with a 26-and 28-day rest period, average pre-grazing sward height of 26.6 ± 8.8 cm and 23.0 ± 9.4 cm and stocking rate of 2.5 and 1.4 AU ha −1 for grazing system 1 and grazing system 2, respectively.
Four observers were trained on the herbage mass estimation techniques and management of the grazing systems for one month before the commencement of the trial measurements.Herbage mass estimation for each paddock was carried out pre-grazing and post-grazing using the undisturbed sward height (O'Donovan et al. 2002) and the comparative yield method (CYM) described by Haydock and Shaw (1975).Undisturbed sward height measurements were performed using a sward stick.The sward height was measured inside a set of 10 quadrats (SHQ), or 150 undisturbed sward heights and random measurements were performed to obtain the average undisturbed sward height per paddock (SHP).For the CYM technique, five 0.25 m 2 quadrats were placed in the paddock and treated as standard points along the herbage mass scale as follows: (1) standard points 1 and 5 represent the lowest and highest herbage mass concentrations, respectively, and (2) intermediate standard points (2, 3 and 4) were then defined to avoid extreme and/or non-representative herbage mass.These standard points were then utilised to rate 40 additional quadrats or visual points on a predefined scale (Balehegn and Berhe 2016).At the conclusion of visual estimation or measurements, the standard point quadrats were cut to ground level and used to estimate the herbage mass (y) through the average of the visual points (x), using linear regression: y = a + bx (Despain and Smith 1997).To calculate the actual herbage mass (HHM) and calibrate both techniques, inside each paddock and grazing time (pre-grazing and post-grazing), a set of 10 quadrats (0.25 m 2 ) of herbage mass were cut to ground level and averaged, resulting in 176 HHM data.All forage sample cuts were weighed and dried at 60 °C for 72 h to express herbage mass in dry matter per unit area (g DM m −2 ).
Within the HHM (10) and CYM (40) quadrats, botanical composition was evaluated using the dry weight rank method (Mannetje and Haydock 1963).Inside each quadrat, ranges 1 (0.7019), 2 (0.2108), and 3 (0.0873) were defined according to the most representative species relative to the dry weight in each quadrat.The botanical composition was expressed as the proportion of species per paddock.
A simple linear regression (n = 176) was used to predict herbage mass based on a single predictor using the lm function in R software (v.4.1.1).Botanical composition was added to each model using multiple linear regression.Thus, if both the slope and intercept were similar across the tested variables, data were combined, and a common regression equation was reported.Model residual assumptions were checked using the Jarque-Bera test and Durbin-Watson tests.
Prediction precision was assessed using the Bland-Altman method, mean squared prediction error (MSPE), and Lin's concordance correlation coefficient (CCC).A paired t-test was used to compare the herbage mass predicted between actual herbage mass (HHM) and the tested technique.Differences were considered statistically significant when p ≤ 0.05, and trends were declared at 0.05 ≤ p < 0.1.
The tested models to predict herbage mass were significant in the evaluated techniques (CYM, SHQ, and SHP), with a higher R 2 and lower residual means in herbage mass predicted through comparative yield method (CYM) than those undisturbed sward height techniques (Table 2).These results may be because the CYM technique includes direct herbage mass cuts, integrating visual ratings of both the height and density of herbage when selecting each range of quadrats.This finding is particularly important in highly heterogeneous pastures, where herbage bulk density is more variable depending on the combination of species found at a particular point.
The RSE from CYM was lower than 10% (RSE = 9.2%), with minimal improvement in RSE and prediction precision after botanical composition calibration.A trend was detected only when U. brizantha (p < 0.10; Table 2).These findings suggest that double sampling techniques for herbage mass estimation in tropical pastures are important for considering potential botanical composition effects.This may be associated to the fact that CYM allows visual estimates of different structural characteristics of the forages within each quadrat, which are selected and ranked according to direct measurements that cover a wide range of herbage mass found within each paddock (George et al. 2006).Thus, CYM and CYM calibrated to the U. brizantha proportion (CYMub) technique provided lower MSPE and higher CCC (Table 3).
Heterogeneous pastures tend to be more complex in vertical structure than pure grass swards due to differences in height, bulk density, leaf and stem arrangement, and growth patterns within the canopy (Sanderson et al. 2006;Murphy et al. 2021).The undisturbed sward height technique could have limitations in predicting herbage mass in heterogeneous pastures, since it only considers one aspect (height) of the vertical structure components of pasture.For example, when using the undisturbed sward height technique for herbage mass estimation in pure grass swards, Costa et al. ( 2009) reported a higher R 2 (0.88) than that found in this investigation (R 2 = 0.68).This could be the result of the higher heterogeneity of the pastures in the current research.The findings of this study are consistent with those of George et al. (2006) in annual rangelands, indicating that CYM is a slightly better predictor of herbage mass than undisturbed sward height.Murphy et al. (1995) and Martin et al. (2005) investigated several techniques for herbage mass estimation in naturalised swards and pastures and concluded that indirect techniques, such as sward rulers or rising plate meters, may perform poorly when the botanical composition is not uniform, ground surface is irregular, or there are significant amounts of senescent or dry material.
Although it is recommended that the R 2 values for herbage mass prediction be ≥ 0.80 (Cóser et al. 2003), the R 2 values of the present study for SHQ and SHP were similar to those reported by O' Sullivan et al. (1987), who used a stick in perennial ryegrass swards.Both the SHQ and SHP models also indicated better relationships than the models described by Sanderson et al. (2001) and Harmoney et al. (1997), who used a pasture ruler and plate meter as indirect techniques for estimating herbage mass in grass-legume pastures in dairy farms and in heterogeneous pastures, respectively.In both cited studies, low R 2 values were attributed mainly to sward heterogeneity and bias of the measurements between the observers.In this study, the four observers were trained for one month to reduce bias and, considering the higher heterogeneity of the growth habits of the species found in the evaluated heterogeneous pastures, the SHQ and SHP models could be considered acceptable.In addition, the CCC of both SHQ and SHP  Table 2: Characteristics of models tested to calibrate the comparative yield method (CYM), sward height quadrat (SHQ), and sward height paddock (SHP) techniques to predict herbage mass per unit area (g DM m −2 ) and their effects when adjusted to the dominant grasses proportion were higher than 0.80 (Table 3), indicating good precision and accuracy (Liao and Lewis 2000).The precision of the predicted herbage mass values using SHQ and SHP allowed for lower standard deviation (SD), coefficient of variation (CV%) and mean bias than CYM (Table 3).Previous investigations have reported the importance of botanical composition calibration when developing models for herbage mass prediction from grassland swards (George et al. 2006;Fricke et al. 2011).Therefore, the most representative grasses were tested and included in the models.When the U. brizantha proportion was included, both undisturbed sward height models (SHQ and SHP) were improved, achieving a higher R 2 , lower residual means and RSE, as well as a lower RMSPE and CCC when compared to the simple models (Table 3).The CYM, CYMub and SHQ calibrated to the U. brizantha proportion (SHQub) techniques allowed RMSPE values to be lower than half of the SD (<74.73) of HHM.In this regard, the differences in growth habits and other structural characteristics between predominant species in tropical heterogeneous pastures and their inclusion could be the primary method for improving the accuracy of herbage mass prediction using indirect techniques such as undisturbed sward height.Urochloa brizantha cv.Marandú have a tufted growth habit, with geniculately ascending culms (Cook et al. 2020).A similar observation was made in the current study, where it was noted that in quadrats where U. brizantha was the dominant species, a lower herbage bulk density was measured when compared to U. decumbens and native grass during both pre-grazing (1.44 vs 1.56 and 1.73 mg DM cm −3 ) and post-grazing (1.28 vs 1.56 and 1.74 mg DM cm −3 ) cuts (Table S1).This could be explained by the fact that U. brizantha has a less uniform leaf arrangement along a vertical distribution gradient, which could affect the relationship between mass and height of the grass.These findings are in line with those of previous studies (Gonzalez et al. 1990;Duru and Bossuet 1992), which also evaluated the accuracy of herbage mass prediction in heterogeneous pastures systems using indirect techniques.Gonzalez et al. (1990) and Duru and Bossuet (1992) concluded that the variation between the herbage mass and undisturbed sward height relationship was mainly due to differences in the structural characteristics of the pastures, such as herbage bulk density and botanical composition.
Calibration using the U. brizantha proportion in both undisturbed sward height models also helped to increase the accuracy of herbage mass prediction, with a reduction in RSE (as percentage of the mean herbage mass) from 21.0% to 18.6% and from 22.0% to 20.9% for SHQ and SHP, respectively.Although the models calibrated with U. brizantha still had an RSE above 10%, a value considered acceptable (Rayburn and Rayburn 1998), these values were still lower than those reported by Harmoney et al. (1997) for herbage mass prediction models in which the structural characteristics were not considered.It is important to consider that an RSE ≤ 10% for techniques such as sward height may be difficult to obtain under farm conditions because of the inherent spatiotemporal variability of pastures, especially when they are mixed (Sanderson et al. 2001).As the undisturbed sward height technique is less laborious and easier to apply under farm conditions, future research is needed in heterogeneous swards to test the calibration of sward height models with other forage characteristics associated with botanical composition, especially in livestock systems where tropical heterogeneous pastures are predominant.
In conclusion, botanical composition affected the models developed for the undisturbed sward height technique, improving the precision of herbage mass estimation in tropical heterogeneous pastures with differences in growth habits between representative dominant species.Thus, calibration with the proportion of U. brizantha in undisturbed sward height models is useful for increasing the accuracy of herbage mass prediction under these conditions.
Abbreviations: CI = confidence interval, RSE = residual standard error, DW = Durbin-Watson statistic, JB = Jarque-Bera test for normality Note: U. decumbens was adjusted for Urochloa decumbens cv.Basilisk proportion and U. brizantha was adjusted to Urochloa brizantha var.Marandú proportion.The native grass was adjusted to the sum of Paspalum conjugatum, Paspalum notatum and Cynodon dactylon.

Table 1 :
Herbage botanical composition during pre-grazing and post-grazing of evaluated paddocks during the evaluation period a Includes Paspalum conjugatum, Paspalum notatum and Cynodon dactylon as dominant native grasses.b Legumes of Desmodium and Arachis pintoi genera.c Consists of weeds composed mainly of Paspalum virgatum, Andropogon bicornis and Cyperaceae sp.

Table 3 :
Prediction precision of different techniques to estimate herbage mass per unit area (g DM m −2 ) and the effect of adjustment for Urochloa brizantha cv.Marandú proportion(n = 176)