A simplified maturity index to quantify the development stage of perennial ryegrass (Lolium perenne L.) and its relationship with yield and nutritive value

The study was conducted as a 2-year field experiment at the Hohenschulen experimental station (54° 18′ N, 9° 58′ E, altitude: 24 m) of the Christian-Albrechts-University, Kiel, Northern Germany. The site is characterized by its sandy loam soils. Average annual temperatures were 10.0°C in 2006 and 9.9°C in 2007. The year 2006 was characterized by lower total annual precipitation, amounting to 707 mm. Higher total annual precipitation values were recorded in the second experimental year (2007), amounting to 926 mm for the experimental site.

Twenty diploid intermediate-heading perennial ryegrass (Lolium perenne L.) genotypes (Table 1), which provide the range of phenological variation found in the corresponding maturity group, were evaluated with respect to their yield and quality performance. Three replicated 3 m by 6.5 m plots per genotype were sown in a randomized complete block design.

Table 1

Genotypes included in the study

Tabelle 1. Liste der in der Studie verwendeten Sorten

The experimental plots were sown on September 6 the year before sampling and the sampling was performed in the following two years (2006/2007). All plots were treated similarly and received 300 kg N ha^-1 in the form of calcium-ammonium nitrate, split into four applications, namely 100, 80, 80 and 40 kg N ha^-1 applied before the first, second, third and fourth harvests, respectively. In addition, 80 kg P₂O₅ ha^-1 were applied at one dose in spring of each year. A potassium fertilizer (40% K₂O, 6% MgSO₄, 4% S, 3% Na) was given at the rate of 360 kg K₂O ha^-1 (200 and 160 kg K₂O ha^-1 before the first and third harvests, respectively). Folicur® (1-(4-chlorophenyl)-4,4-dimethyl-3-(1,2,4-triazol-1-ylmethyl)pentan-3-ol) was sprayed against crown rust (Puccinia coronata) at a rate of 0.7 l ha^-1 two weeks after each harvest beginning from the second one.

The harvest dates for the two experimental seasons are presented in Table 2. The plots were managed with four cuts, and for the first and second cut, three sampling dates were set, to account for the variation within genotypes and sampling dates within the first cut. The first and third sampling dates within the first and second cuts were performed at a smaller area in the plot, as the second sampling corresponds to the four-cut management. The first sampling for the first cut was taken when the first node was detected for about half of the cultivars. The second and third samplings for the first cut were taken at the early and late silage maturity stages (corresponding to the ear emergence), respectively. The second cut was also designed to have 3 samplings (first sampling occurred usually in one corner of the plot and third sampling next to it; the sampling number and cuts are depicted in Table 2). The samplings of the second cut were performed 35 up to 44 days after the samplings within the first cut.

Table 2

Harvesting schedule for the two experimental seasons

Tabelle 2. Erntetermine für die zwei Versuchsjahre

At the time of sampling, the grass was cut manually to 5 cm stubble height. The actual sampling area within each plot varied between 0.25 and 0.5 m² depending on the amount of above ground biomass. After the third sampling within the first and second cut, the remainder of the plot was harvested with a Haldrup plot-harvester to 5 cm stubble height. All forage from the sample area within each plot was weighed to quantify the fresh herbage weight. A representative sub-sample was then dried at 60°C reaching a constant weight to determine the dry matter (DM) content for laboratory analysis.

The phenological stage of the plants was quantitatively monitored at each sampling date by cutting a representative sample of around 50 tillers randomly from each plot to ground level. Tillers were classified according to the 17 stages of development described by Park (1980) as presented in Table 3. In both years, only 10 stages, namely those from 3 to 12, were detected in the experimental plots. Numerical indices for quantifying the phenological development included two different approaches. Firstly, the Mean Stage by Count (MSC), representing the reference method, was calculated as the average of the individual stage categories present in the herbage sample, weighted for the number of tillers at each stage (Moore et al., 1991): MSC=∑<!-- ? -->inSi⋅<!-- ? -->NiC$$\begin{array}{} \displaystyle MSC=\frac{\sum\limits^n_iS_i\cdot N_i}{C} \end{array} $$

where S_i = stage number, N_i = number of tillers in stage S_i, and C = total number of tillers in herbage sample.

Table 3

Classification of Lolium perenne L. developmental stages according to Park (1980)

Tabelle 3. Klassifizierung des Entwicklungsstadiums von Lolium perenne L. nach Park (1980)

In the second approach, the percentage of tillers above a given developmental stage i as defined according to Park (1980) was calculated, where i varied between the first and the last of the observed stages, resulting in a total of 10 different simple maturity indices (SMI_i), which were later on tested for their suitability: SMIi=∑<!-- ? -->inNiC$$\begin{array}{} \displaystyle SMI_i=\frac{\sum\limits^n_iN_i}{C} \end{array} $$

where SMI_i = Simple Maturity Index beginning from stage i, N_i = number of tillers in stage i, and C = total number of tillers in herbage sample.

The dried sub-samples were uniformly ground using a Cyclotec 1093 sample mill (Foss, Sweden) to a particle size of 1 mm. All available samples were scanned twice using NIR-Systems 5000 monochromator (Perstrop Analytical Inc., Silver Spring, MD 20904, USA), where the software (ISI version) for data collection and manipulation was supplied by Infrasoft International (ISI, Port Matilda, PA, USA). Calibration and validation subsets were relatively small in number, with 36 and 30 calibration samples, and 14 and 20 validation samples in year 1 and 2, respectively, since an already existing NIRS calibration was refined.

The concentrations of neutral detergent fiber (NDF), acid detergent fiber (ADF) and acid detergent lignin (ADL) were determined sequentially as described by Van Soest et al. (1991) using the semiautomatic ANKOM²²⁰ Fiber Analyzer (ANKOM Technology, Macedon, NY, USA). NDF and ADF were analyzed without a heat stable amylase and expressed inclusive of residual ash, while ADL content was corrected after the residual ash content. The in vitro cellulase technique developed by De Boever et al. (1988) was used to determine the digestibility of herbage samples. The percentage of digestible organic matter (DOM) was then calculated by applying the following equation of Weissbach et al. (1999): DOM(%)=100×<!-- ? -->(940−<!-- - -->CA−<!-- - -->0.62×<!-- ? -->EULOS−<!-- - -->0.000221×<!-- ? -->EULOS2)/(1000−<!-- - -->CA)$$\begin{array}{} \displaystyle \rm DOM(\text%)=100\times(940-CA-0.62\times EULOS-0.000221\\\rm\times EULOS^2)/(1000-CA) \end{array} $$

where CA = crude ash and EULOS = enzyme insoluble organic matter; CA and EULOS are expressed in g kg^-1 DM. Calibrations were later developed by regressing the laboratory-determined values of sample subsets against the NIR spectral data. Means and ranges as well as coefficients of correlation and standard errors of calibration and validation for the investigated quality parameters are presented in Table 4.

Table 4

Statistical data of the calibration of NIRS for NDF, ADF, ADL and DOM of the investigated genotypes

Tabelle 4. NIRS Kalibrationsstatistik für NDF, ADF, ADL und VOM der untersuchten Sorten

The data from both growing seasons were averaged, because the ‘year’ includes two aspects, namely climatic conditions and sward age, which cannot be separated from each other statistically.

To develop a simple maturity index (SMI_i) to quantify the developmental stage of a perennial grass sward, correlations between the tested maturity indices, that is, MSC and SMI_i (with i between three and twelve), on the one hand and yield and forage quality traits (NDF, ADF, ADL and DOM) on the other hand, were computed using Spearman’s rank correlation coefficient in SAS 9.1 PROC CORR (SAS Institute, 2000) as the data were not normally distributed (Table 5). Coefficients were calculated on data that were averaged over year and replicate.

Table 5

Spearman’s rank correlation coefficients between the tested maturity indices, i.e., Mean stage by count (MSC) and Simple maturity indices SMI_i (i = 3-12), and forage quality parameters

Tabelle 5. Spearman Korrelationskoeffizienten zwischen den getesteten Indices, d.h. “mean stage by count” (MSC) und “simply maturity indices” SMI_i (i = 3-12), und Futterqualitätsparameter

To explore and compare the behavior of treatments and their influence on the studied yield and quality parameters in the presence and absence of the maturity index as a covariable (i.e., the SMI_i), the statistical analysis was conducted twice with PROC MIXED for a randomized complete block design. The analysis without the covariable included cut, sampling date, genotype and block, as well as their interactions, as fixed effects. The second analysis was done in presence of the covariable expressed through its main effect. In both analyses, cut × sampling date was treated as a repeated measure and analyzed using the unstructured (UN) covariance structure and the REPEATED statement in PROC MIXED. Least square means were separated by the SAS PDIFF option in PROC MIXED. Significance was declared at P < 0.05 and adjusted using Bonferroni-Holm procedure (Holm, 1979).

The correlation between MSC and the tested quality parameters reached the level of significance for the three sampling dates within the first cut, in addition to the first and second sampling dates within the second cut (Table 5). In the third sampling date within the second cut, MSC was significantly correlated only to ADL and DOM contents. Correlation was non-significant between MSC and yield in both cuts. Correlation analyses between the percentage of tillers above a given development stage and the investigated parameters revealed that considering the percentage of tillers above the boot stage (stage 8 – SMI8, Table 3) showed comparable significances and magnitude of correlations to that produced from the MSC for both cuts and all sampling dates. Correlation coefficients clearly fluctuated for the quality parameters and reached mostly the highest values in case of the DOM. Generally, higher correlation coefficients were observed for the first rather than for the second cut.

As a result of the correlation analyses, it was decided to include the SMI8 as a covariable in the analysis of variance and run the statistical analysis twice, with and without the covariable, to explore its influence on the impact of genotype, cut, sampling date and their interaction on the studied parameters, as mentioned above. Only the effects including genotypic variations will be presented and discussed in detail.

The analysis of variance performed without using SMI8 as covariable is presented in Table 6. The genotype × cut interaction was observed with respect to the fiber fractions and yield. However, it turned non-significant in case of yield after correcting with Bonferroni-Holm test. Except for the ADL content, the cut × sampling date interaction influenced all the other parameters.

Table 6

F-values of genotype, sampling date, cut and their interactions on yield and feed quality parameters without using covariable (SMI8)

Tabelle 6. F-Werte für Sorte, Erntetermin, Aufwuchs und Wechselwirkungen für Ertrag und Futterqualität ohne Kovariable (SMI8)

Table 7 presents the combined analysis of variance considering the covariable SMI8. Concerning the two-way interactions, a similar trend to that illustrated in Table 6 was observed, where the genotype interacted with the cut for the fiber fractions, while the cut × sampling date influenced both the yield and fiber fractions. The results confirm the hypothesis that the covariable SMI8 exerted a significant influence on yield and quality measurements. The slopes reported in the table show that each unit’s increase in the covariable SMI8 caused an increase of 0.90, 0.30, 0.30 and 0.05 units in the yield (t ha^-1), NDF, ADF and ADL (g kg^-1), respectively, as well as a decrease of 0.43 units in the amount of DOM (g kg^-1).

Table 7

F-values of genotype, sampling date, cut and their interactions on yield and forage quality parameters with covariable (SMI8)

Tabelle 7. F-Werte für Sorte, Erntetermin, Aufwuchs und Wechselwirkungen für Ertrag und Futterqualität mit Kovariable (SMI8)

The comparison of means for the genotype × cut interaction of NDF and ADF contents are displayed in Table 8 for both analyses, that is, with or without covariable. The inclusion of the covariable resulted in several effects. While in the first cut, no significant genotype effect was left at all after considering the covariable, the larger effects were observed in the second cut. Firstly, the difference between the maximum and minimum NDF contents amounted to 57.2 and 55.2 g NDF kg^-1 for the second cut without and with the contribution of the covariable, respectively. The same applied to the ADF content, where the range amounted to 41.2 and 36.4 g ADF kg^-1 for the second cut without and with considering the covariable, respectively. The inclusion of the covariable reduced the variation accounted for the development stage on NDF and ADF content. Secondly, it was detected that the inclusion of the covariable affected the significances among the genotypes within the second cut, modifying the ranking of genotypes from the highest to the lowest content of NDF and ADF. Similar results could be observed for ADL and DOM (g kg^-1), as shown in Table 9. The difference between the maximum and minimum ADL content in the first cut amounted to 6.0 g kg^-1 without the covariable but only 4.7 g kg^-1 after including it. The inclusion of the covariable reduced the variation accounted for the development stage on ADL content and DOM. Likewise, the ranking of genotypes changed for ADL and DOM after introducing the covariable in cut 1, while in cut 2 this effect was not observed. Nonetheless, the covariable did not contribute substantially in changing the ranking of the genotypes in both ADL and DOM within a cut. Generally, we observed higher contents of fiber fractions and consequently a lower digestibility in the second compared to the first cut.

Table 8

Means of NDF and ADF contents (g kg^-1) of the 20 tested genotypes for the two cuts as analyzed with and without the covariable (SMI8)

Tabelle 8. Mittlere Gehalte für NDF und ADF (g kg^-1) der 20 untersuchten Sorten aus 2 Aufwüchsen mit und ohne Berücksichtigung der Kovariable (SMI8)

Table 9

Means of ADL and DOM contents (g kg^-1) of the 20 tested genotypes for the two cuts as analyzed with and without the covariable (SMI8)

Tabelle 9. Mittlere Gehalte für ADL und VOM (g kg^-1) der 20 untersuchten Sorten aus 2 Aufwüchsen mit und ohne Berücksichtigung der Kovariable (SMI8)

Analysis of correlations between growth stages and yield and constituents of nutritive value (Table 5) revealed variable correlation coefficients, with the percentage of tillers at or beyond the boot phase. Our simplified numerical maturity index SMI8 provided similar correlations to the studied parameters as the mean stage by count (MSC). This was in agreement with the findings of Ansquer et al. (2009), who identified three phenological stages, namely the start of stem elongation, flowering, and seed ripening, as key to manage dynamics of growth and demography of temperate grassland species. Similarly, Mika (1983) found that cultivar differences in timothy were more pronounced at ear emergence than at later sampling dates suggesting that a sampling date at or near the ear emergence is to be preferred for routine evaluation. Our results using SMI8 support this observation.

Maturity ratings in the current study were based on individual tillers, which proved to be an accurate method for quantifying the different developmental stages in herbage samples. However, Van Santen and Casler (1990b), in comparing the individual tiller versus the whole plant in maturity rating, suggested that the maturity can effectively be rated using the simpler whole plant visual rating. This conclusion, however, maybe misleading since their tiller samples comprised only the five most mature individual tillers, which may not be considered as representative. A larger random sample, as in our study, might have avoided this bias.

The covariable SMI8 affected all the studied parameters, where a main effect was always detected. This suggests that the maturity aspect was reflected in distinctive morphological development and nutritive value (Cop et al., 2009). Relationships between morphological stage and nutritive value of herbage for livestock are well documented (Sanderson and Wedin, 1989; Valente et al., 2000; Jeangross et al., 2001; Pontes et al., 2007).

The analysis of variance (Table 7) highlighted a relatively strong influence of the covariable on yield, but a relatively low impact on the quality parameters, as indicated by the slopes. We suggest that this low effect in case of the quality parameters was in part due to the design of the combined analysis which involved two cuts with three sampling dates each. This structure may have partially contributed to mask the impact of the covariable, where the magnitude of effect of the studied covariate was dependent on other components of the analysis of variance and their interactions. Therefore, an additional analysis was conducted using a simpler statistical model, exemplified for the DOM content, and the results are shown in Tables 10 and 11. Only the second sampling date in each cut was considered, and the two cuts were analyzed separately with and without the covariable. The second sampling date, done at an early silage maturity, was chosen because it represents the developmental stage at which grasses are commonly harvested in intensive dairy farming systems.

Table 10

Genotypic influence on DOM (g kg^-1) in absence and presence of the covariable (SMI8) for the 2^nd sampling date within 1^st and 2^nd cuts

Tabelle 10. Einfluss der Sorte auf die VOM (g kg^-1) mit oder ohne Berücksichtigung der Kovariable (SMI8) für den 2. Erntetermin des 1. und 2. Aufwuchses

Table 11

Means of DOM (g kg^-1) of the 20 tested genotypes for the 2^nd sampling date within 1^st and 2^nd cuts as analysed with and without the covariable (SMI8)

Tabelle 11. Mittlere VOM (g kg^-1) der 20 untersuchten Sorten für den 2. Erntetermin des 1. und 2. Aufwuchses mit oder ohne Berücksichtigung der Kovariable (SMI8)

It becomes evident that, in the first cut, the genotypes were variable in their DOM contents. After considering the covariable, the variation in DOM content was not significant among the 20 genotypes, suggesting that the development stage is largely responsible for the variation in DOM content (Table 11). Despite the different basis of the comparison that resulted from simplifying the structure of the new analysis of variance, comparing the slope in Table 7 (−0.43) with the newly calculated slope in Table 10 (−0.70) approves the hypothesis that the magnitude of the effect of covariable changed as the design of the analysis of variance changed. With the new slope, the correlation between the covariable and the DOM content is getting closer to that achieved from the correlation analysis in Table 5. First cut DOM means listed in Table 11 confirm the contribution of the covariable in minimizing the variations among the genotypes (from 56.1 to 33.6 g kg^-1 before and after considering the SMI8).

A different situation was observed in the case of the second cut, where the variations among the 20 tested genotypes remained after considering the covariable by 55.2 to 54.2 g kg^-1 (Table 11). Since the reproductive tillers are the main component of the covariable SMI8, its effect will become weaker and less distinguishable as the number of reproductive tillers decrease, which is clearly observed in the second cut.