Multi-Locational Evaluation of Forage-Suited Selected Sudan Pearl Millet [Pennisetum glaucum (L.) R. Br.] Accessions Identified High-Yielding and Stable Genotypes in Irrigated, Arid Environments

Abstract

Pearl millet [Pennisetum glaucum (L.) R. Br.] is a subtropical grain and forage crop. It is privileged with several desirable forage attributes. Nevertheless, research on pearl millet is limited, especially as a forage crop, in developing countries. Therefore, the objectives of this study were to investigate the field performance and stability of pearl millet genotypes for forage yield across seven environments. The study was conducted in seven environments (combination of locations and seasons) during the 2016/2017–2018/2019 seasons. Twenty-five pearl millet genotypes, selected based on forage yield from a core collection of 200 accessions, were arranged in an alpha lattice design with three replications. The parameters measured were fresh forage yield, days to flowering, plant height, number of culms m⁻², leaf-to-stem ratio, and stem girth. The combined analysis revealed that environments, genotypes, and their interaction had significant effects on all traits studied except the genotypic effect on stem girth. Across the seven environments, four genotypes (G14, G01, G12, and G22) outyielded the check genotype in fresh matter yield by 20.7, 16.5, 11.0 and 9.8%, respectively. The additive main effects and multiplicative interaction (AMMI) analysis showed that the genotype, environment, and their interaction were highly significant (p ≤ 0.001) for fresh matter yield. The results of AMMI stability values (ASVs) and the genotype selection index (GSI) combined with the AMMI estimate-based selection showed that genotypes G14, G22 and G01 were the most stable and adapted genotypes and were superior to the check genotype. These results indicate that forage pearl millet varieties could be developed directly through evaluating the wealth of available collections or indirectly through hybridization in crop breeding programs.

1. Introduction

In the era of climate change, pearl millet [Pennisetum glaucum (L.) R. Br.] could be one of the most climate-resilient crops due to its better adaptability to marginal environments and high nutritional values [1,2,3,4,5,6]. Pearl millet is a subtropical grain and forage crop and was probably domesticated 5000 years ago in Africa in the savannah south of the Sahara and west of the Nile [4,7]. It is a staple food for 90 million impoverished people and is cultivated on 30 million ha in the arid and semi-arid tropical regions of Asia and Africa. Pearl millet is particularly beneficial due to its high nutritional value, climate resilience, and potential for higher economic returns in marginal conditions compared to other cereal crops [2].

As forage, it is grown in several subtropical countries. Pearl millet is also advantageous as a dual-purpose crop because it is an excellent livestock feed, both as grain and fodder [8]. Pearl millet outperforms other cereals such as wheat, maize, rice, sorghum, and barley, especially under marginal conditions, due to its high photosynthetic efficiency, higher dry matter production capacity, better drought tolerance, early maturity with the capability to regrowth for multiple harvests, and ability to survive under adverse agro-climatic conditions with lower inputs and higher economic returns [1,2,4,9]. Due to its heat stress tolerance and water use efficiency, pearl millet is well adapted to harsh climates where other crops fail to produce economic yields [10,11].

Because of the suitability of millets for cultivation under adverse and changing climatic conditions, their resilience to pests and diseases, their better nutritional values, and their capability to ensure food security as an indigenous crop in arid regions, the year 2023 was declared by the United Nations General Assembly as the International Year of Millets [12]. Millets, in general, can grow in marginal areas such as arid lands with minimal inputs and are resilient to changes in climate. As such, they are an ideal solution for countries, especially in arid areas, to increase self-sufficiency and reduce dependence on imported cereal grains [12].

Compared to maize, Sudan grass, and sorghum, pearl millet is privileged with several desirable forage attributes, e.g., it is known to be more tolerant to abiotic stresses, such as drought, salinity, high temperature, and soil nutrient deficiency, compared to other cereal crops such as sorghum, wheat, maize and rice [13,14]; it also has high leafiness and tillering capacity, a high percentage of crude protein, and high forage yield [4,15], in addition to excellent regenerative ability, which permits multi-cut forage production and grazing [16]. Moreover, it is free from prussic acid at all growth stages [15].

Crop yield is a function of genotype, environment, and their interaction. The evaluation of genotype × environment interaction (GEI) is necessary when different genotypes respond differently to different environments [17]. A significant GEI can seriously compromise efforts to select superior genotypes for crop adaptation and variety development programs [17,18,19]. Several methods have been developed and used to evaluate the GEI in different crops, including pearl millet [20]. Amongst them, the additive main effects and multiplicative interaction (AMMI) analysis have been used to identify superior and well-adapted genotypes, while genotype and genotype × environment (GGE) biplot analysis has been used for the graphical discrimination of the genotypes and environments [17,18,20]. The AMMI model incorporates both the additive and multiplicative components of the two-way analysis of variance (ANOVA). The AMMI model first fits the additive effects for the main effects of genotype (G) and environment (E) using the usual additive ANOVA procedure and then fits the multiplicative effects for the genotype by environment interaction (GEI) by principal component analysis (PCA) [21]. In addition, the graphic illustration of the GGE biplot was found to be a valuable tool in displaying the which-won-where pattern, which could lead to the identification of different mega-environments; in identifying high-yielding and stable cultivars; and in assessing the discriminating ability and representativeness of the test environments [22].

In Sudan, pearl millet is well adapted almost all over the country [4,7]. However, it had never been considered a forage crop to any degree. Instead, it is grown mainly for grain production in the warm and drier regions of western Sudan (Darfur and Kordofan), with little in the central clay plains, e.g., Gezira, Gadarif, Sennar and Blue Nile [23]. The 5-year average in 2015/16–2019/20 of the harvested area of pearl millet in Sudan was 3.4 M ha, with a total grain production slightly above 1.3 M tons [23]. It has desirable forage attributes as mentioned above [16]; nevertheless, it has not been grown for forage production in Sudan.

Sudan is endowed with a huge number of livestock as well as a very vast range of land. Paradoxically, however, the livestock products are neither cheap in the local markets nor contribute as much as their potential to exports. This paradox could largely be attributed to the fact that the animal production system is more extensive than intensive. Hence, livestock is mainly reared on natural rangelands, crop residues, and little-cultivated forage crops. A practical approach to improve the situation of animal production is through establishing intensive production farms for both dairy and meat production. This is highly plausible due to the increasing adoption of modern irrigation systems, such as central pivot irrigation, by several farmers and investors. Currently, the most popular forages under these systems are long-standing perennial crops, mainly Rhodes grass and alfalfa. Adopting a two-course legume/grass rotation by introducing crops with short life cycles could be a more feasible and productive system and improve forage yield and nutritive value [15,24]. Exploring the potential of pearl millet as an irrigated forage crop could provide a short duration of multiple-cut forage grass. As an endogenous crop in Sudan, pearl millet showed a huge number of accessions with wide variations.

As an exception to a few introduced grain varieties, research on pearl millet is limited in Sudan. In fact, more than 3200 accessions were collected from different parts of Sudan and available in the Agricultural Plant Genetic Resources for Research and Conservation Centre (APGRC) of the Agricultural Research Corporation in Wad Medani, Sudan (https://www.genesys-pgr.org/a/overview/v2GajzMOG5e, accessed on 6 November 2023). Such a huge number of accessions could be utilized to improve the crop for both grain and forage production.

Differential fresh matter yield [25], forage quality [26], and genetic diversity [27,28] among the pearl millet genotypes were evident. However, not even a single forage pearl millet variety has been developed in Sudan. Therefore, the objectives of this study were to evaluate the field performance, stability, and the GEI for forage yield and other related traits of selected pearl millet [Pennisetum glaucum (L) R. Br] genotypes under different environments.

2. Materials and Methods

2.1. Experimental Sites and Plant Materials

The experiment was conducted for three consecutive seasons at the Rahad site (2016/2017, R1; 2017/2018, R2 and 2018/2019, R3) and two seasons at the Gezira and Sennar sites (2016/2017, G1, S1 and 2018/2019, G2, S2, respectively).

Table 1. Coordinates (latitude and longitude), altitude, soil characteristics and precipitation for the experimental sites where 25 pearl millet accessions were grown.

Characteristics	Experimental Sites
	Gezira (Gezira State)	Rahad (Gedarif State)	Sennar (Sennar State)
Latitude	14°24′ N	14°08′ N	13°33′ N
Longitude	33°29′ E	34°06′ E	33°34′ E
Altitude (masl)	406.9	426	421
Soil type	Typic Haplocambids	Typic Haplocambids	Typic Haplocambids
Clay content (%)	58	67	60
pH	8.3	7.8	7.8–8.5
Organic carbon (%)	0.20	0.74	0.40–0.50
Total nitrogen (%)	0.05	0.04	0.05
Rainfall (average 2016–2018, mm)	308	451	456
Mean maximum temperature (July–October, °C)	36.3	-	36.9
Mean minimum temperature (July–October, °C)	22.9	-	20.5

summarizes the basic information about the three experimental sites, including coordinates (latitude and longitude), altitude, soil characteristics, and precipitation. Minimum, mean and maximum temperatures at Gezira (Wad Medani) and Sennar were always similar, with Sennar sometimes being slightly cooler (Table S1). The highest amount of rainfall was always recorded at Rahad, followed by Sennar, and most of the rainfall occurred during July and August at the three experimental sites. The pearl millet genotypes used in this study were provided by the Agricultural Plant Genetic Resources for Research and Conservation Centre (APGRC) of the Agricultural Research Corporation (ARC), Sudan. Two hundred pearl millet accessions, mostly collected from the Darfur and Kordofan regions of Sudan, were evaluated, and one hundred of them were selected for having glabrous (non-hairy) leaves [25,26,29]. Twenty-five genotypes were identified as top-ranking for their yield. The pearl millet accessions were purified by pure line selection, so the numerical suffix −1 was added to the original accession number to distinguish it from the original accession (Table S2).

2.2. Experimentation and Data Collection

The selected twenty-five genotypes were arranged in an alpha lattice design with three replications. To increase the experimental precision, reduce the coefficient of variation, and increase the proportion of total variation due to blocking, each replication was subdivided into five blocks, each with five accessions. In each season, the land was disc plowed and harrowed, leveled, and ridged to 80 cm. The plot size was four ridges 5.25 m in length each with a spacing of 0.8 m between ridges. The planting rate was five seeds/hole, which was later decreased to three plants/hole. The sowing in each season and location was performed in the first week of July. Irrigation was performed every 12–14 days or as needed according to weather conditions, particularly considering rainfall and temperature. Fertilizer was applied at the second irrigation (187 kg urea/ha). The experiments were kept free from weeds by hand weeding, and the ridges were kept intact by earthing up after every weeding. Days to 50% flowering (DF) were recorded as the number of days from sowing to a stage when 50% of plants in the plot reached flowering. When each genotype reached the 85% flowering stage, data were collected on plant height (cm), number of culms m⁻², leaf-to-stem ratio, stem girth (cm), and fresh forage yield (t/ha), as previously described [25]. Briefly, the number of culms m⁻² (CLM) was recorded as the average number of culms of three counts per square meter in each plot. Plant height (PH) was the average of five measurements of the main shoot, and leaf-to-stem ratio (LSR) was measured as the average of three plants on a dry matter basis. The accessions were harvested manually from the ground level in an area of 3.2 m⁻² and the fresh matter yield (FY) was weighed immediately in the field.

2.3. Statistical Analysis

The statistical analysis was conducted using Genstat Twenty-second Edition (VSN International Ltd., Hemel Hempstead, UK) [30]. The statistical analysis was carried out on the combined data, with each location–season combination considered as an environment. The best linear unbiased estimates (BLUEs) of the traits studied were generated using the residual maximum likelihood (REML) method of mixed model theory. The genotype (G), environment (E), and their interaction (GEI) were allotted to the fixed model, whereas blocks nested within replications were assigned to the random model. The model of additive main effect and multiplicative interaction (AMMI) was performed to partition the GEI. The AMMI stability value (ASV) was used for the stability analysis of genotypes, in addition to the non-parametric index and genotype selection index (GSI), calculated by the following formula [31]:

GSI = RASV + RY

where RASV is the rank of ASV and RY is the rank of mean fresh matter yield of genotypes across environments.

Furthermore, genotype and genotype × environment (GGE) biplot analysis was performed to identify mega-environment, superior genotypes (which won where), and representative and discriminating environments [22]. The violin plots were constructed using GraphPad Prism version 10.1.2 for Windows, GraphPad Software, Boston, MA, USA. Hierarchical clustering and heatmap analysis of the phenotypic traits were performed using the R package heatmap version 4.0.2 [32]. The distance measure used in clustering rows and columns based on the possible values is “correlation” for Pearson correlation, and all the distances are supported by “Euclidean”. Scale parameters were based on the normalized values.

3. Results

3.1. Genotypic and Environmental Effects

Table 2. Minimum, maximum and mean values of fresh matter yield (t/ha), days to 50% flowering, plant height (cm), leaf-to-stem ratio, number of culms m⁻² and stem girth (cm) of the 25 pearl millet genotypes grown across different environments, together with Chi probabilities and standard errors (SE) for environment, genotype and their interaction.

	Fresh Matter Yield (t ha−1)	Days to Flowering	Number of Culms (m−1)	Plant Height (cm)	Leaf to Stem Ratio	Stem Girth (cm)
No. of environment	7	7	5	7	5	3
Min	6.5 (R1)	46 (R1)	14.8 (S2)	133 (S2)	0.131 (R2)	2.90 (G1)
Max	26.5 (R3)	70 (R1)	49.6 (G1)	179 (G1)	0.411 (G1)	4.40 (G2)
Mean	15.6	56	31.7	156	0.241	3.57
Chi probability
Environment (E)	<0.001	<0.001	<0.001	<0.001	<0.001	<0.001
Genotype (G)	<0.001	<0.001	<0.001	<0.001	<0.001	0.728
G × E	<0.001	<0.001	<0.001	<0.001	<0.001	0.082
SE ±
E	0.66	0.7	1.53	1.7	0.0094	0.065
G	0.62	0.6	1.05	1.4	0.0104	0.113
G × E	1.63	1.5	2.57	3.8	0.0255	0.196

G1, Gezira 2016/2017; G2, Gezira 2018/2019; R1, Rahad 2016/2017; R2, Rahad 2017/2018; R3, Rahad 2018/2019; S2, Sennar 2018/2019.

summarizes the minimum, maximum and mean values of fresh matter yield (t/ha), days to 50% flowering, plant height (cm), leaf-to-stem ratio, number of culms m⁻² and stem girth (cm) of the 25 pearl millet genotypes grown across different environments (combination of location and season). The combined analysis revealed that environments and genotypes had highly significant effects on all traits studied except the genotypic effect on stem girth (Table 2). The G × E interaction effects were also highly significant for all traits studied except for stem girth, which was significant at p = 0.025.

3.2. Fresh Matter Yield (FY)

The FY of the twenty-five genotypes in the seven environments is shown in Table 2. Across all genotypes, the mean FY was highest at Rahad 2018/19 (R3), followed by Gezira 2018/19 (G2), whereas the lowest FY was recorded at Rahad 2016/17 (R1). Across all environments, genotypes G14, G01, G12 and G22 outyielded the check (G25) by 20.7, 16.5, 11 and 9.8%, respectively. Compared to the check, the FYs of genotype G14 were higher in the seven environments, whereas that of G01, G12 and G22 were higher in five, four and three environments, respectively (

Table 3. Fresh matter yield (t/ha) of twenty-five genotypes of pearl millet (Pennisetum glacum L.) grown across seven environments at Gezira, Rahad and Sennar sites, Sudan.

Genotype (G)	Environment (E)							Mean (G)	Relative Performance (%)
	Gezira 2016/17 (G1)	Gezira 2018/19 (G2)	Rahad 2016/17 (R1)	Rahad 2017/18 (R2)	Rahad 2018/19 (R3)	Sennar 2016/17 (S1)	Sennar 2018/19 (S2)
G01	20.29	22.60	10.21	18.68	26.54	12.81	12.73	17.70	116.5
G02	22.80	19.70	9.90	14.39	19.23	8.89	9.94	14.98	98.6
G03	21.72	17.74	10.96	13.73	19.47	12.37	13.37	15.62	102.8
G04	15.64	22.26	7.07	13.83	26.33	8.87	15.18	15.60	102.7
G05	24.86	20.97	15.59	11.74	21.72	7.79	11.59	16.32	107.4
G06	17.99	25.69	6.94	12.14	18.82	16.28	14.20	16.01	105.4
G07	16.22	22.98	12.67	7.11	19.72	11.91	12.08	14.67	96.6
G08	19.14	19.60	9.84	15.21	18.22	10.25	11.53	14.83	97.6
G09	19.36	26.52	7.62	12.57	18.82	10.42	10.93	15.18	99.9
G10	22.31	11.03	9.70	13.92	18.95	8.77	12.94	13.95	91.8
G11	16.11	17.02	9.27	11.88	25.05	10.32	12.72	14.62	96.3
G12	27.18	21.42	6.90	16.23	21.98	8.60	15.69	16.86	111.0
G13	20.15	19.16	8.06	14.35	20.73	12.60	12.70	15.39	101.3
G14	23.80	27.36	10.67	19.23	24.08	11.51	11.70	18.34	120.7
G15	17.70	16.70	8.01	17.27	19.20	10.63	14.62	14.88	97.9
G16	13.98	15.46	7.01	12.29	23.74	7.27	14.04	13.40	88.2
G17	13.22	20.38	10.04	25.70	19.79	13.73	11.25	16.30	107.3
G18	17.09	17.44	10.04	21.36	20.29	10.30	12.02	15.51	102.1
G19	23.15	21.76	8.41	16.65	21.64	7.53	11.37	15.79	103.9
G20	19.27	12.40	8.63	12.73	19.34	11.17	17.25	14.40	94.8
G21	19.27	23.11	10.48	12.90	23.91	13.83	8.34	15.98	105.2
G22	16.54	22.91	6.03	23.40	25.43	9.50	12.92	16.68	109.8
G23	21.02	15.91	8.51	17.26	22.57	7.15	13.44	15.12	99.6
G24	19.65	22.80	7.02	12.15	25.42	10.55	12.32	15.70	103.4
G25	17.23	23.56	10.40	15.43	18.67	9.86	11.19	15.19	100.0
Mean (E)	19.43	20.26	9.20	15.29	21.59	10.52	12.64	15.56
SE±	for environment = 0.663, for genotype = 0.616, for G × E = 1.629

).

3.3. Growth Attributes

3.3.1. Days to Flowering

Across genotypes, the mean days to flowering were latest at R1, which showed the widest range of flowering together with R2. Narrow flowering ranges were observed at S2, G1 and G2 (Figure 1a). In R1 and R2, G12 (70 and 65 days, respectively) and G22 (69 and 57 days, respectively) were significantly late in attaining flowering compared to the check (60 and 53 days, respectively) (Table S2). Genotypes G01 and G14 were comparable to the check in R1 but were significantly earlier (49 and 48 days, respectively) than the check (54 days) in R2. At Sinnar, G01, G12, G14 and G22 were significantly different from the check in S1 but not in S2 (Table S2).

3.3.2. Number of Culms

The highest numbers of culms m⁻² were recorded at G1 and G2, whereas the lowest number was recorded at S2. Genotypes showed wide ranges at both Gezira sites. On the contrary, S2 showed a narrow range among the genotypes in the number of culms m⁻² (Figure 1b). At G1 and G2, genotypes G24, G05 and G12 showed a consistently higher number of culms than the check (G25), whereas at R2, G18, G02 and G19 were the top-ranking genotypes. At R3, G17, G09 and G16 ranked top, whereas at S2, G13, G08 and G20 were the top-ranking genotypes. Across all environments, G24 and G09 showed a 16.2 and 11.4% higher number of culms over the check, respectively (Table S3).

3.3.3. Plant Height

The mean plant height at G1 and G2 was 168 and 160 cm, respectively. Plants were shorter at the Rahad sites, with mean plant heights of 148, 152 and 155 cm at R1, R2 and R3, respectively. Narrow ranges in plant height among the genotypes were observed at R2 and R3, whereas the widest range was observed at S1, followed by S1 and G1 (Figure 1c). Plant height at S1 was taller (160 cm) than that at S2 (147 cm). Except for R3, the five tallest genotypes were taller than the check variety in all environments (Table S4).

3.3.4. Leaf-to-Stem Ratio

The Gezira environments showed higher leaf-to-stem ratio (LSR) values compared to the Rahad environments (Figure 1d). Within the same location at Gezira, the LSR values ranged from 0.283 to 0.323, whereas at Rahad, the range was from 0.189 to 0.217. Across all environments, G16 showed the highest LSR values, whereas G01 showed the lowest value. In all environments, the five top-ranking genotypes had higher LSR values than the check genotype. Across the five environments, G16, G22, G21 and G24 were the top-ranking genotypes (Table S5).

3.3.5. Stem Girth

The stem girth was measured in only three environments (G1, G2 and S2). As shown earlier, genotypes had no effect on stem girth; however, the effect of the environment was highly significant. The stem girth at G2 was significantly higher than that at both G1 and S2 (Figure 1e). The mean stem girth at G2 was 4.05 cm, compared to 3.43 and 3.23 cm at G1 and S2, respectively (Table S6).

3.4. Heatmap Clustering Analysis

The cluster heatmap analysis illustrated the responses of different forage-related traits of the 25 pearl millet genotypes grown in 7 environments. The 40 trait combinations clustered the 25 pearl millet genotypes into 6 groups (Figure 2a). The largest group consisted of nine genotypes, followed by a group consisting of six genotypes. The check variety was grouped with G01, G04 and G17. Genotypes G14 and G19 were placed in a separate group. Genotypes G11, G19 and G22 were grouped together, while G12 was clustered alone (Figure 2a).

When the means of the six traits were used, the twenty-five pearl millet genotypes were also clustered into six groups (Figure 2b). However, there were differences in the distribution of the genotypes among these clusters. The largest group consisted of eight genotypes, including genotypes with a forage yield (FY) around or below the mean, followed by a group consisting of seven genotypes, including the check variety, which had an FY around the mean and with low SG values. The third group consisted of five genotypes, most of which had below-average CLM. Genotypes G14 and G01 were placed in a separate group as they were the top fresh-matter yielders. Similarly, genotypes G12 and G22 were grouped together because of their high FY and tall height. The genotype with the lowest FY and highest LSR (G16) was clustered alone (Figure 2b).

3.5. Genotype × Environment Interaction and Stability of Fresh Matter Yield

The analysis of variance of the additive main effects and multiplicative interaction (AMMI) showed that the genotype (G), environment (E) and their interaction (GEI) significantly (p ≤ 0.001) affected the fresh matter yield of the pearl millet genotypes (

Table 4. The ANOVA of AMMI for fresh matter yield of 25 pearl millet genotypes in 7 environments.

Source	DF	SS	MS	VR	F pr	% Explained
Total	524	20,054	38.3
Treatments	174	15,786	90.7	8.01	<0.001
Genotypes	24	681	28.4	2.51	<0.001	4.3
Environments	6	11,087	1847.8	56	<0.001	70.2
Block	14	462	33	2.91	<0.001
Interactions	144	4018	27.9	2.46	<0.001	25.5
IPCA 1	29	1279	44.1	3.89	<0.001	31.8
IPCA 2	27	1173	43.4	3.84	<0.001	29.2
IPCA 3	25	539	21.6	1.9	0.0064	13.4
Residuals	63	1027	16.3	1.44	0.0231
Error	336	3805	11.3

DF = degree of freedom, SS = sum of squares, MS = mean squares, VR = variance ratio, F pr = F probability.

). Based on the total treatment sum squares, E, G, and GEI accounted for 70.4%, 4.3%, and 25.5% of the variance, respectively. The GEI was partitioned into first, second, and third interaction principal component axes (IPCA1, IPCA2, and IPCA3). IPCA1, IPCA2, and IPCA3 were highly significant and contributed 31.8, 29.2, and 13.4%, respectively, to the total GEI sum of squares (Table 4).

3.5.1. AMMI Stability Value and Genotype Selection Index

Based on the ASV of fresh forage yield, genotypes G11, G13, G19, G25, and G14 showed the lowest ASVs of 0.129, 0.184, 0.198, 0.397, and 0.450, respectively (

Table 5. AMMI estimates of fresh yield (t/ha), AMMI stability value (ASV), and genotype selection index (GSI) of 25 pearl millet genotypes grown in 7 environments in Sudan.

Genotype Code	Mean FY	Rank (FY)	Stability (ASV)	Rank (ASV)	GSI Value	Rank (GSI)
G01	17.21	3	1.079	13	16	5
G02	15.06	16	0.971	11	27	11
G03	16.07	8	1.237	15	23	9
G04	15.67	11	1.793	22	33	19
G05	15.83	10	2.058	23	33	18
G06	16.29	6	0.487	6	12	4
G07	14.91	18	1.675	21	39	24
G08	14.25	23	0.885	9	32	16
G09	14.67	19	0.957	10	29	13
G10	14.5	20	1.629	19	39	25
G11	15.27	15	0.129	1	16	7
G12	16.72	4	0.757	8	12	3
G13	15.32	14	0.184	2	16	6
G14	18.39	1	0.45	5	6	1
G15	14.33	22	1.278	16	38	23
G16	13.49	25	0.55	7	32	17
G17	15.94	9	2.782	25	34	20
G18	16.35	5	1.049	12	17	8
G19	16.17	7	0.198	3	10	2
G20	14.49	21	1.157	14	35	21
G21	15.63	12	1.352	18	30	14
G22	17.75	2	2.506	24	26	10
G23	15.03	17	1.665	20	37	22
G24	15.47	13	1.347	17	30	15
G25	14.18	24	0.397	4	28	12

). Genotypes G01 and G12, with their high forage yield, showed moderate ASVs, whereas G22 was one of the most unstable genotypes according to ASV. However, based on the GSI, which combines the rank of ASV with the rank of the genotypes according to their fresh yield, G14, G19, G12, G06, and G01 showed low values and ranked first to fifth, respectively. Genotypes G11, G13, and G25, which had low ASVs, were ranked seventh, sixth, and twelfth, respectively, by their GSI values (Table 5).

3.5.2. AMMI Estimate

The four best pearl millet genotypes were arranged based on AMMI estimates in the seven environments. Genotype G14 was among the four best genotypes in four environments, whereas G03, G17 and G22 were among the top four genotypes in three environments. G01, G04, G05, G06 and G18 appeared in two environments among the best four genotypes. G25 (check) was not among the four best-ranking genotypes in any environment (

Table 6. The four best genotypes in each of the seven environments as per AMMI selections.

Environment	Mean	Score	1st	2nd	3rd	4th
Gezira 2016/17 (G1)	19.43	3.144	G05	G12	G14	G10
Gezira 2018/19 (G2)	20.26	−1.662	G22	G24	G01	G04
Rahad 2016/17 (R1)	9.20	1.362	G03	G05	G14	G06
Rahad 2017/18 (R2)	15.29	−2.228	G17	G22	G18	G23
Rahad 2018/19 (R3)	21.59	−1.031	G22	G14	G01	G04
Sennar 2016/17 (S1)	10.52	0.088	G06	G07	G17	G03
Sennar 2018/19 (S2)	12.64	0.327	G17	G03	G14	G18

).

3.5.3. GGE Scattered Biplot

The scattered biplot of the GGE analysis showed that 56.16% of the variance was due to PC1 (30.87%) and PC2 (25.29%) (Figure 3a). The seven environments were grouped into three mega-environments. The first mega-environment comprised S1 (Sennar 2016/17), G2 (Gezira 2018/19), and R3 (Rahad 2018/19), with G22 being the winning genotype. The second mega-environment included G1 (Gezira 2016/17) and R1 (Rahad 2016/17), where G05 performed well, whereas the third mega-environment included R2 (Rahad 2017/18) and S2 (Sennar 2018/19), with G17 being the winning genotype. Environments G2, R2 and R3 were the most discriminating environments, while S1, R1 and S2 were the least discriminating. Both S1 and R3 were more representative, but R3 was a representative and discriminating environment (Figure 3b). On the other hand, R2 and G2 were less representative but highly discriminative and thus good for identifying specific adapted genotypes. Relative to the ideal genotype (with a high mean yield across all environments and high stability), G22, G01, G04, and G14 were the most desirable genotypes (Figure 3c). On the other hand, G12, G20, G15 and G23 were the least desirable genotypes due to their poor performance across all environments. Although G25 (the check) was more stable, its fresh matter yield was similar to the grand mean of the fresh matter yield.

4. Discussion

The importance of the current study stems from the fact that it is the first multi-location evaluation of high-ranking, forage-based selected pearl millet genotypes and the first exploratory study of forage-related variations among accessions of a Sudanese collection of pearl millet [25,29]. In this study, it was evident that some pearl millet genotypes produced fresh matter yield significantly more than the check cultivar at the three locations. This, in turn, encourages further exploration of forage-suited accessions among Sudanese pearl millet collections [15,24].

The estimated annual forage gap in Sudan is more than 28 million tons [26]. However, about 23 million tons (82%) are in the form of production rations (i.e., concentrates). The expansion of concentrate production in Sudan is highly unlikely under the current crop production system. The situation is further exacerbated by the growing trend to export alfalfa hay. An alternative way to minimize the concentrate gap in Sudan should be based on the horizontal and vertical expansion in the production of high-quality forages. From this point of view, the importance of pearl millet as a forage crop that plays such a role could be appreciated.

Pearl millet is privileged over the common cereal forages in Sudan, such as sorghum, barley, and maize, for its high yields and quality, in addition to its high regenerative ability, suitability to more than one season, and freedom of prussic acid [15,16]. Compared to forage sorghum in Sudan, the fresh forage yields of the higher-yielding pearl millet genotypes of this study are either similar or slightly lower [33,34].

The analysis of variance of growth attributes revealed highly significant differences among the pearl millet genotypes. This indicated that the genotypes used in this study, although screened and selected in previous studies [25,26], had sufficient variability. The results also showed that growth attributes (days to 50% flowering, plant height, number of culms m⁻², leaf/stem ratio) were significantly affected by genotype, environment, and their interaction. The significant difference among the tested pearl millet genotypes in the growth attributes may reflect their differential responses to the environment.

On average, the Gezira location showed higher fresh matter yield compared to Sennar and Rahad in 2016/17 and 2017/18. In fact, a wide range of seasonal variation was observed in Rahad. Although the mean days to flowering were longer in R1, the LSR was lower, and the plant height was shorter compared to G1 and G2. This might be related to the seasonal variation in the climatic data, mainly rainfall, as the temperatures during the growing seasons were mostly similar, especially in Gezira and Sennar. In consideration of the heavy clay soils at the three locations, which may result in waterlogging with heavy and unevenly distributed rains, and the fact that the crop received supplementary irrigation as needed, it can be postulated that the low rainfall at Gezira may have favored the vegetative growth. Further investigation is warranted to elucidate the impact of climatic conditions (temperature, rainfall, day length) on the performance of pearl millet as an irrigated forage crop.

The FYs at G2 and S2 were associated closely with the plant height, while those at R2, and to some extent at G1, were associated with CLM (Figure 2a). With the means of the six traits, the FY was associated with CLM but not with LSR and DF (Figure 2b). In terms of plant height, G12 and G22 were among the tallest, whereas G01 and G14 were among the shortest. The mean LSR of G01 and G14 was below average, while that of G22 was above average. Leafy semidwarf types of forage hybrids with high leaf-to-stem ratios have been released and cultivated in the United States [35]. Among the four top-ranking genotypes for mean FY, two (G12 and G22) showed above-average days to flowering, while two (G01 and G14) exhibited below-average days to flowering. Although days to flowering were reported to be significantly correlated with rainfall in the Sahel region [36] and were the main factor determining grain yield under mid-season and terminal drought stress [37], such an association was absent or weak under non-stress irrigated conditions [37] similar to those in the current study.

The existence of genotype × environment interaction is challenging for crop improvement and could hamper genetic gain, especially under stress conditions such as those usually facing pearl millet production [10,11,18,38]. Understanding the environmental causes of the GEI could facilitate the identification of adaptive plant traits and may lead to a more rational choice of test environments for crop breeding programs [18,19,38].

The cluster heatmap analysis conducted on twenty-five pearl millet genotypes grown in seven environments using various forage-related traits illustrated the grouping of genotypes based on similarities or dissimilarities in these forage-related traits. The twenty-five pearl millet genotypes were grouped into six distinct clusters based on these trait combinations. Despite the fact that the pearl millet genotypes used in this study were selected from a larger population on the basis of forage-attributed traits [25,26], the cluster analysis provided insights into the diversity of trait responses among the 25 pearl millet genotypes. A better understanding of the pattern of the trait variation could provide useful information for forage breeding programs aiming at developing cultivars with desired forage-attributed traits under different environmental conditions.

Information on GEI is important for plant breeders to develop improved, stable and better-adapted varieties. In this study, the statistical analysis showed that the genotype and environment main effects and genotype × environment interaction were very highly significant (p ≤ 0.001) for forage yield and all other traits except stem girth, indicating the differential response of genotypes in cross-testing environments [18].

The AMMI analysis also revealed significant effects of genotype, environment, and their interaction (GEI) on fresh matter yield. The substantial proportion of the variance attributed to the environmental effect indicated the great impact of the environment on the fresh matter yield of pearl millet. The interaction between the environment and the genotype was also an important source of variation, which was further partitioned into three IPCAs, all of which were found to be significant.

The scattered GGE biplot showed that the seasonal variation within the same location was more prominent than the locational variation. The mega-environments identified are composed of different environments from different locations. It is therefore important to consider seasonal variations and include more locations in future studies for evaluating pearl millet for forage production more effectively. Likewise, the genotypic ranking within the same location varied from season to season. For instance, G14 was not among the best four genotypes in R2, while it ranked second and third in R3 and R1, respectively. This highlights the importance of seasonal variations, which need to be considered in the future through in-depth analysis of the crop performance against closely monitored and detailed climatic conditions. The combined use of the AMMI stability value (ASV) and the genotype selection index (GSI) to evaluate genotype stability and adaptability across the environments provided better insights into genotype performance. Genotype G14, with lower values of both ASV and GSI, displayed desirable stability and performance across environments compared to genotypes, demonstrating a consistently high yield across different conditions. Other genotypes, such as G12 and G01, also showed varying levels of stability and performance.

The ranking of the genotypes according to the AMMI estimate, based on their performance across different environments, reinforced the superiority of G14, which was among the best four genotypes in four environments. Genotypes G03, G17 and G22 were among the top four genotypes in three environments, whereas G01, G04, G05, G06 and G18 were among the four top-ranking genotypes in two environments. On the other hand, the check genotype, G25, was not among the four top-ranking genotypes in any of the seven environments.

Pearl millet was identified, a long time ago, to provide good-quality, high-yielding summer grazing forage [39]. The wide genetic variation reported here and elsewhere with regard to growth habits, light requirements, drought tolerance, response to temperatures, forage yield, and forage nutritional quality is deemed promising for further forage yield improvement in pearl millet [25,26,35]. Future efforts to develop pearl millet for forage production should consider the development of hybrids that capitalize on the existing genetic diversity.

Our findings clearly demonstrate the usefulness of the pearl millet accessions collected from different locations in Sudan. It was possible to identify high-yielding, stable accessions for forage yield from among a relatively small number of 200 accessions that were systematically evaluated across seasons. This indicates that exploring the full potential of all available accessions may result in identifying more useful genetic materials for direct use or use in breeding programs. Currently, most of the crop breeding programs in Sudan depend mainly on introducing genotypes from abroad. Henceforth, we believe that the significant variations found in this study could greatly impact future breeding efforts. This is particularly true for indigenous crops such as pearl millet and sorghum, due to the large number of accessions available in the Sudan National Genebank.

5. Conclusions

The wide range of performance observed among the tested pearl millet genotypes in this study indicated the possibility of developing pearl millet varieties for forage production directly through evaluating the wealth of available collections or indirectly through a crop breeding program. However, the existence of highly significant genotype x environment interaction needs to be carefully addressed. Across all environments, genotypes G14, G01, G12 and G22 outyielded the check cultivar by 20.7, 16.5, 11.0 and 9.8%, respectively. The results of ASV and the GSI combined with the AMMI estimate-based selection showed that genotypes G14, G22 and G01 were the most stable and adapted genotypes, demonstrating their superiority over the check cultivar, which exhibited average values for the above attributes. The above-mentioned high-yielding and stable genotypes could be recommended as forage-suited varieties in the irrigated subtropical environment of Sudan. The results of this study indicate that forage pearl millet varieties could be developed directly through the evaluation of the available wealth collections or indirectly through hybridization in crop breeding programs.