Grasslands provide critical ecosystem services for humanity (Sollenberger et al., 2019). In Brazil, pastures occupy 156 million hectares across six previously defined biomes (Projeto MapBiomas, 2021). These pastures support a national herd of ~224 million cattle (IBGE, 2019) and places Brazil as the world's second-largest beef producer, after the United States (Jank et al., 2014). Pastoral farming systems in Brazil are heavily reliant on native forage species for productive stability and conservation of natural resources. Their use can reduce costs and risks associated with livestock production, which increases long-term system sustainability (Gasparetto et al., 2021).

The genus Paspalum comprises several native species with important forage characteristics for animal production. They exhibit adaptability to the range of different ecosystems where they are present as a pasture component (Novo et al., 2016), which means there is high potential for genetic improvement (Motta et al., 2017). The center of genetic diversity is located in the tropical region of South America (Chase, 1929; Valls, 1987). The genus is native to the southern grasslands, is most abundant in Brazil (Rio Grande do Sul), Uruguay, and Argentina, and has previously been recognized for its high yield and forage quality (Steiner et al., 2022). In Argentina, the region with the greatest diversity of species is Mesopotamia (Morrone et al., 2012).

A major objective of forage plant breeding programs in southern Brazil is to obtain hybrids adapted to diverse edaphoclimatic conditions, which have superior biomass production compared with cultivars already on the market (Saraiva et al., 2021). There is considerable commercial and academic interest in improvement of different Paspalum species to increase the productivity of native pastures and extend their use as improved cultivated pastures (Steiner et al., 2022).

The significant time and resource requirements involved in obtaining consistent data within breeding programs, considering the extensive number of genotypes and characters studied (Jank et al., 2014), requires the determination of the minimum number of measurements for the selection of superior genotypes (Rodrigues et al., 2020). This is a major issue faced by forage breeding programs, which need to determine the number of measurements necessary to accurately estimate the differences between genotypes (Toebe et al., 2020). This is generally determined through repeatability analysis, which aims to predict the genotypic value of a genotype over time with predefined determination coefficients (Chaves et al., 2018). The repeatability analysis reduces time, costs, and labor within the experimental period to optimize the process of launching new cultivars into the market (Torres et al., 2015; Rodrigues et al., 2020).

The objective was to estimate the repeatability coefficients of forage characteristics in intraspecific hybrids of P. notatum Flügge using different methods.

The analysis of variance of the forage variables was statistically significant for all effects tested (genotype, harvest, and their interaction), except PH which showed significant differences for main effects (genotype and harvest) only (Table 2). Experimental (CV^e; 38.3 (GH) to 128.4% (TPD)) and genetic (CV^g; 18.1 (GH) to 77.9% (SDM)) coefficients of variation were high for all characters evaluated (Table 2). Heritability estimates ranged from 0.33 (LSR) to 0.95 (PH). Characters of interest within the forage plant genetic improvement program, associated with forage quality and production, such as LDM (0.85) and TDM (0.81), also had high heritability values.

The presence of significant genotype × harvest interactions for all forage characters, except PH, reinforced the need to estimate the repeatability coefficients for the characters because responses vary over time (Table 3). For TDM, ρ^ ranged from 0.19 (ANOVA II) to 0.88 (PCACov). With the exception of the ANOVA II method (54%), the coefficient of determination (R²) exceeded 80%, which indicated reliability in identifying P. notatum genotypes with superior TDM (Table 3). The pattern of the PCACov and ANOVA II, which produced the highest and lowest ρ^, respectively, was repeated for the other characteristics evaluated in the study. The ρ^ values obtained for TDM and LDM traits were closely correlated, which indicated that the genotypes evaluated in this experiment performed well for important agronomic traits required for future genetic improvement. The LSR ρ^ were poor and ranged from 0.03 (ANOVA II) to 0.49 (PCACov), with associated R² from 12 to 83%, respectively. Analysis showed that PCACov, PCACor, and SACov methods almost consistently gave the highest ρ^ for all characteristics evaluated, while ANOVA I, ANOVA II, and SACor had the lowest ρ^ values.

The estimated minimum number of harvests needed to select the best hybrids identified by the five analysis methods for TDM, with a precision of 0.85, ranged from 1 (PCACov) to 24 (ANOVA II; Table 4). When the precision level was increased to 0.95, minimum harvest numbers for TDM increased for all methods and ranged from 3 (PCACov) to 79 (ANOVA II). For LDM, an agronomically important trait that provides an indirect measure of forage quality, the estimated number of harvests required for a precision of 0.85 ranged from 1 (PCACov) to 16 (ANOVA II). Increasing the required precision level to 0.95 meant the minimum number of harvests increased to 3 and 54, respectively. Across all traits evaluated, the estimated minimum number of harvests identified by all five analysis methods was the highest for LSR and varied substantially across methods within a specific precision level. For example, the minimum number of harvests ranged from 6 (PCACov) to 198 (ANOVA II) with precision of 0.85 and from 20 (PCACov) to 663 (ANOVA II) with a precision of 0.95.

Estimated minimum harvest numbers were higher for ANOVA II and I methods when compared with PCACov, PCACor, and SA. The principal components method generated the lowest estimates, especially when based on the variance and covariance matrix (PCACov). Structural analysis (SA) produced values very close to PCACor for most of the characteristics studied.

In general, forage characteristics evaluated by PCACor had higher genotypic stability values than those produced by the ANOVA II model (Table 5). For LDM, SDM, LSR, IDM, and TDM variables, the highest ρ^ were obtained when harvests 1 and 2 were correlated, regardless of the method used. The highest ρ^ for TPD (ρ^ = 0.84; R² = 91.5%) and GH (ρ^ = 0.67; R² = 80.2%) traits were observed for the correlation of harvests 3 and 4, and for PH using harvests 4 and 5 (ρ^ = 0.94; R² = 96.7%). High ρ^ were observed for most of the characters studied. The exception was the LSR trait, which indicated that it was not stable among the genotypes evaluated.

The basic premise for selection is the presence and knowledge of genetic variability within the population (Nielsen et al., 2014; Figueiredo et al., 2019; Sant'Anna et al., 2021; Steiner et al., 2022). The findings of the present study indicated that genetic variability existed among the hybrids evaluated, which allows genetic gain via direct selection in all traits quantified (Table 2).

The experimental (CV_e) and genetic (CV_g) variation coefficients were higher than those found by other authors who have previously evaluated forage characteristics (Machado et al., 2021; Silveira et al., 2022) or seed production traits (Lopes et al., 2017; Lopes et al., 2019) of P. notatum hybrids. The current experiment found that CV_e was always greater than CV_g (Table 2), which showed that the environmental effect dominated results rather than genetic effects. This result is contradictory to previous studies because with greater experimental precision the greatest expressions of genotypic variability were not expressed (Shimoya et al., 2002). The tendency of a greater CV_e compared with CV_g was not observed by Lopes et al. (2019). Precision (CV_e) and repeatability (ρ^) are influenced by management decisions, experimental design, and/or biotic and abiotic factors (Johnson and Frey, 1967; Vela-Cardenas and Frey, 1972; Neyhart et al., 2022). Furthermore, early selection in perennial plants can be affected by numerous factors, such as slow establishment, which would lead to high coefficients of experimental variation. Additionally, the seedlings were very young when transplanted, which may have led to an increase in the coefficients of variation. Conversely, high CV_e values can mean greater phenotypic variation (Wang et al., 2022). At the same time, CV_g provides information about the magnitude of the variability present within the population. This allows comparison of the levels of genetic variability present among different genotypes, environments, and traits (Ferrão et al., 2008). The current experiment confirmed that genetic variability studies within breeding programs must quantify both CV_e and CV_g during analysis (Cortes et al., 2019). Populations that exhibit high CV_g indicate potential for improvement via genetic gains in the breeding program (Zanata et al., 2010). The main advantage of utilizing the coefficient of variation in trait analysis is to enable the quantification and weighting of the proportion of variation within a population, which exists due to genetic and environmental factors (Kampa et al., 2020). This parameter, along with the h², provides the basis for decision-making to increase forage production within the current P. notatum Flügge breeding program.

Heritability values quantify how much of the total observed variation was caused by genotypic factors (Ferreira et al., 2020). Resende (2015) classified h² into three classes; low (h² < 0.15), moderate (0.15 < h² < 0.50), and high (h² > 0.50). Accordingly, the h² parameter generated in the analysis of this experiment (Table 2) was high for most forage traits evaluated. The exception was LSR, which is greatly influenced by environmental conditions over time. Higher h² values allow the identification of superior hybrids within the population studied. Selection of these hybrids in the next stage of the breeding program could improve forage gains obtained in subsequently selected progeny (Majidi et al., 2009).

The high number of genotypes and characters studied within breeding programs (Jank et al., 2014), requires significant time and resources to obtain consistent data and needs an estimate of the minimum number of measurements for selection of superior genotypes (Rodrigues et al., 2020). Therefore, several studies have been conducted to estimate the repeatability (ρ^) of characteristics of interest in forage plants including Urochloa spp. (Basso et al., 2009; Souza Sobrinho et al., 2010; Teixeira et al., 2011; Martuscello et al., 2013; Matias et al., 2016; Coêlho et al., 2018; Figueiredo et al., 2019), Megathyrsus maximus (Martuscello et al., 2007; Braz et al., 2015; Martuscello et al., 2015; Coêlho et al., 2018; Ferreira et al., 2019), Medicago sativa L. (Botrel et al., 2000; Ferreira et al., 2010), Pennisetum purpureum Schum. (Shimoya et al., 2002; Rodrigues et al., 2020), and Lolium multiflorum Lamarck (Rios et al., 2019). Until now this information has not been available for the Paspalum species. The ρ^ value indicates a consistent ranking of genotypic performance in a specific location over time (Neyhart et al., 2022). In our study, ρ^ values were medium-to-high for most characteristics (Table 3), based on the criteria of Resende (2002). In most cases, the highest estimates were obtained through the PCACov analysis method, followed by PCACor. These results are supported by previously published work, which also showed that ρ^ estimates obtained by ANOVA are generally lower than the estimates obtained by other methods (Cargnelutti Filho et al., 2004; Martuscello et al., 2007; Martuscello et al., 2015). Our work showed that the PCACov method estimates of ρ^ tended to be higher than those obtained by the other methods, which is supported by the results of Martuscello et al. (2015). The difference among ρ^ estimates from the analysis methods indicated that evaluation by more than one analysis method is required to obtain a reliable parameter. Evaluation of the analysis procedure as a whole means the real value can probably be found within the range of estimates calculated (Martuscello et al., 2007). In this context, identification of the most appropriate analysis method should be used as a strategy to improve parameter estimates for future decision making (Martuscello et al., 2015).

The study of genotypic stabilization contributes to increasing the reliability of the selection process for a trait by identifying groups of repeated measurements with a higher level of association. This is based on the assumption that gene expression can be influenced not only by the stage of development but also by various climatic conditions and management changes that plants experience throughout the year (Braz et al., 2015). For most traits studied here, genotypic stabilization between two harvests showed the highest correlation (Table 5). The evaluation between successive harvests may not represent genotypic stabilization, but the occurrence of two very similar harvests (Ferreira et al., 2019). This probably reflects they have been exposed to similar environmental conditions, rather than non-adjacent harvests, which likely experience less similar environmental conditions due to seasonal changes. It is necessary to include longer evaluation periods with more harvests to determine genotypic stabilization. Thus, the evaluation of stabilized genotypes to obtain ρ^ estimates is extremely important for selecting material within breeding programs (Martuscello et al., 2015).

Finally, our study reports, for the first time, estimates of the repeatability coefficient for forage production traits in hybrids of P. notatum Flügge. Our results showed values differed among the methodologies applied (Table 3). In general, the repeatability coefficient estimates were of medium to high magnitude for most traits evaluated and indicated reliability in identifying superior hybrids of P. notatum Flügge. For all methods evaluated, increasing the accuracy from 0.85 to 0.95 would require a large increase in the number of harvests, but this adds little in terms of precision, especially when there are many hybrids to be evaluated within the improvement program. In addition, the h² and CV_g genetic parameters indicated a favorable situation for genetic gains with selection within the studied population.

For reliable early selection based on LDM, LSR, and TDM traits of P. notatum Flügge, two to four harvests are recommended. Broad heritability for most of the characters studied suggests this would provide a favorable situation for direct selection for increased forage production. The principal component analysis, based on the covariance matrix, has the highest repeatability estimates compared with the other methods. For early selection, this method is recommended for identification of superior hybrids of P. notatum Flügge.