Introduction

The main limitation of livestock production in Africa is the instability in yield and low quality of native forage species ^{(Pamo et al., 2007)}. Buffelgrass [Pennisetum ciliare (L.) Link] is a C₄, apomictic species, common in the Africa subtropics. It has been introduced into arid and semiarid areas around the world, as Argentina ^{(Carloni-Jarrys et al., 2018)} because of its high forage production potential, soil erosion control, and drought tolerance ^{(Quiroga et al., 2013)}. Genetic diversity for forage traits in buffelgrass offers an opportunity to make selection and to develop new varieties with good production to face the challenges of climate change. There is currently a high demand for forage to feed livestock in areas with low precipitation, thus justifying plant breeding programs ^{(Carloni-Jarrys et al., 2018)} to improve forage production, forage quality and seed yield. There are morpho-agronomic characteristics associated with forage production and seed yield, which can be measured to get a better understanding of the potential within the species ^{(Ribeiro et al., 2019}; ^{Wassie et al., 2018)}. Exploring the natural variability through a morphological characterization could be used to select the best accessions with some characteristics of interest ^{(Alves et al., 2014)}. The International Livestock Research Institute (ILRI) holds a collection of 157 accessions of buffelgrass in its field genebank in Ethiopia; thus, it is necessary to perform a phenotypic characterization of this germoplasm to identify novel sources of variation, as well as outstanding plant material. The objectives of this study were to determine phenotypic diversity of 147 accessions and 10 cultivars of buffelgrass in Zwai, Ethiopia, and to identify the best accessions with characteristics for forage and seed production.

Materials and methods

The study was carried out at the Zwai Experiment Station of ILRI in Ethiopia, located at 7° 54' N, 38° 44' E and altitude of 1640 masl. The average annual precipitation is 600 mm and the peak concentration of rainfall is recorded between July and September. The average annual minimum and maximum temperatures are 13 and 27 ^°C, respectively. Soil is loamy sand with 0.5 m depth and pH from 8.1 to 8.4.

Accessions were planted at least ten years before this study and each accession was established into a 2 × 5 m unreplicated plot, 36 plants per plot with 50 cm separation. The accessions and cultivars maintained in the field genbank came from 10 African countries (Table 1). Some accessions and cultivars were donated by other programs and their origin was unknown. Plots were irrigated five times during the dry season only for maintenance, and 200 kg of nitrogen and 60 kg of phosphorus per hectare were applied.

On June 16th, 2014, all plants were cut at a height of 150 mm above ground, characterization began on July 16th and finished on December 5th. Six descriptors were recorded from eight plants located at the center of the plot ^{(Van de Wouw et al., 1999)}; plant height (PH) was measured in cm from the ground level to the tip of all inflorescence, forage height (FH) was also measured in cm from the ground level to the tip of leaves, panicles per plant (PP) was the number of inflorescences per plant. Eight panicles were randomly taken to count the number of spikelets per panicle (SP) and caryopses per panicle (CP). For days to 50 % of flowering (DF), the whole group of plants in the plot was observed and the number of days from cutting until half of the plot reached flowering was registered.

The Pearson correlation coefficient was calculated with the CORR procedure ^{(SAS Institute, 2011)} to observe the relationship among variables. Furthermore, a hierarchical cluster analysis using the CLUSTER procedure ^{(SAS Institute, 2011)} by the Ward clustering method was performed. Groups were chosen based on PST2 (pseudo-T ² statistic). To determine the variables with higher discriminating power the STEPDISC procedure ^{(SAS Institute, 2011)} was used. The discriminant function for the probability that an accession belonged to a particular group was analyzed with the DISCRIM procedure ^{(SAS Institute, 2011)}. Also, if a difference between groups was recorded (P ≤ 0.05), a multivariate analysis of variance (MANOVA) with the Wilks’ Lambda statistic was calculated ^{(SAS Institute, 2011)}.

Results and discussion

The 157 accessions of buffelgrass showed a high degree of diversity in all agro-morphological characteristics measured. Plant and forage height varied from 34 to 138 and from 24 to 118 cm, respectively. Panicles per plant ranged from 6 to 322, days to 50 % flowering were from 48 to 71, spikelets per panicle ranged from 13 to 285 and caryopses per panicle from 3 to 113. The results obtained for plant height were similar to those of previous studies ^{(M’Seddi et al., 2002)}. In general, our collection showed higher variability in the rest of the traits compared to results reported by ^{Mseddi et al. (2004)}. This discrepancy could partially be attributed to the higher number of accessions in this collection than the 51 accessions in the aforementioned study, perhaps due to the selection of a broader range of eco-geographical locations.

The results obtained reveal a significant potential of this collection to be used in plant breeding programs. High correlation coefficients (P ≤ 0.0001) were observed between plant height and forage height (r = 0.98), and between panicles per plant and days to 50 % flowering (r = -0.626). The rest of the correlation coefficients were weak. Most advanced accessions showed a higher number of panicles per plant, which agrees with the result in the previously mentioned study of 52 accessions of buffelgrass ^{(Mseddi et al., 2004)}. ^{Morales et al. (2012)} found that in Arizona cottontop [Digitaria californica (Benth.) Henr.] forage height is more closely related to yield than plant height. It might be possible to extrapolate from these results and propose that buffelgrass accessions with large plant height could be used for forage production.

Hierarchical cluster analysis assigned the 157 buffelgrass accessions into four groups (Table 1, Figure 1). Also, MANOVA with Wilks’ Lambda statistic showed differences among these groups (P ≤ 0.0001). Group I included 57 accessions along with Karasberg, Nunbank and Viva cultivars. This group was characterized by early flowering, large plant height, and forage height traits. Group II included 50 accessions and the cultivars Biloela, Boorara, Gayndah, Kongwa and Towoomba. This group was characterized by late flowering, low number of panicles per plant and caryopses per panicle. Group III consisted of 44 accessions and the cultivars American and Molopo, with a larger number of panicles per plant, early flowering and small size. Group IV included six accessions; this group was the tallest one and contained the largest number of spikelets per panicle and caryopses per panicle.

Figure 1 Dendrogram showing the integration of four groups using the Ward method based on agro-morphological characteristics of 157 buffelgrass accessions. 

Discriminant analysis demonstrated that most traits, except FH, contributed to separate clusters (P ≤ 0.05), (Table 2) and the probability that an accession belonged to a particular group was higher than 95 %. The results on days to 50 % flowering and plant height agree with those of previous studies ^{(Griffa et al., 2012}; ^{Jorge et al., 2008)}; however, some groupings among cultivars were different which could also be related to the location and the period when experiments were performed, as plant characteristics are influenced by environmental as well as by genetic factors. Plant height has been shown to positively correlate with yield forage in buffelgrass (Mansoor et al., 2012; ^{Mseddi et al., 2004}); hence, Groups I and IV can potentially be used for silage or hay production.

In some species, characteristics such as number of panicles per plant and total number of seeds per panicle have been shown to be related to seed yield ^{(Youssef and Hansson, 2019)}. Since Groups III and IV have the largest number of panicles per plant and caryopses per panicle, respectively, accessions in these groups could be used to select for improved seed yield in a plant breeding program. The accessions in Group IV could be selected as new cultivars as they have good productivity and seed yield; nevertheless, it should be noted that it is necessary to perform research focused on seed yield and production ^{(Morales-Nieto et al., 2017)}, as well as on forage quality. It is concluded that the studied collection of 157 accessions of buffelgrass exhibited high variability in all characteristics measured, highlighting the opportunity to improve this species through a breeding program. Accessions 1442, 19448, 19457, 19459, 19462 and 19464, native to South Africa and Zimbabwe, are the tallest ones and have the largest number of spikelets per spike and caryopses per spike; therefore, they may have potential to be included directly in future plant breeding programs for enhancing seed and forage production of buffelgrass.