1. Introduction
The global population is expected to grow approximately 1% in the coming years, and the global use of major agricultural commodities, including wheat, is expected to grow by 1.1% due to both population growth and per capita growth in demand [
1]. Global wheat production in 2022/23 is estimated at 770 million tonnes and is expected to rise to 840 million tonnes [
1], of which durum wheat (
Triticum durum Desf.) has an average production of approximately 40 million tonnes [
2]. The increase in global cereal production results from increased yield, considering the limited potential for the availability of more arable land. However, environmental concerns and policies (e.g., the EU Green Deal) could negatively affect the yield in some places [
1]. This increases the need to sustain and even increase the yield of field crops more sustainably, exploiting available techniques and agricultural practices that contribute to an efficient increase in productivity.
Originating from the Levant, where it was first domesticated, durum wheat was an essential cultivar for Greeks, Phoenicians, Romans, and all Mediterranean countries, which until 1955 planted local landraces [
2]. Durum wheat holds substantial economic significance due to its distinctive attributes and the food products, notably pasta. It serves as a crucial energy source, offering diverse vitamins, minerals, and other essential nutritional compounds vital for human dietary requirements [
3]. Over the last few decades, researchers have conducted many breeding programs. New genotypes produced by these breeding programs have to be evaluated in multi-environmental trials; AMMI analysis then needs to be conducted to group the genotypes being assessed into the most suitable environments [
4]. Although innovations in the field of genetics have had the ultimate effect of increasing the yield [
5], classical breeding programs are still important and, along with the use of new techniques and technologies, could continue to provide new cultivars of durum wheat [
6]. Multi-location trials are also necessary to identify the most suitable genotypes for the target locations.
G × E interaction involving genotype performance changes across different environments and can complicate the identification of superior genotypes. This phenomenon is known as the crossover concept [
7]. The ultimate target remains to identify the genotype that produces high and stable yields in various environments; this can then be put forth as a superior selection with high adaptability [
4,
8]. Genotypes with confirmed high grain yields and stability under certain environmental conditions could be used as parental lines in breeding programs in locations with similar conditions [
9]. Various statistical tools have been developed to address G × E interpretation and crop stability and are classified into two categories based on their statistical properties: parametric and non-parametric measures. Commonly used parametric models are the variance of deviations from regression (
) [
10], the stability variance (
) [
11], the coefficient of variability (
) [
12], and the
stability measure [
13] from AMMI analysis [
14]. Concerning the non-parametric statistics, many models have been suggested, including Si by Nassaar and Huehn [
15] and NPi by Thennarasu [
16], which are based on similar genotype ranks across environments (stable genotypes), with genotypes that display similar ranks across environments assumed to be stable. Parametric and non-parametric statistical methods may approach stability differently. These approaches introduce two distinct facets of stability: the “static” or “biological” and the “agronomic” or “dynamic” [
17]. In terms of the static aspect, a genotype is considered stable when it maintains a consistent yield across various environments. However, none of these methods comprehensively explain genotype performance across diverse environments [
18].
AMMI and GGE biplot analyses were used to visualize the results of the interactions between genotypes and environments. These analyses aim to identify varieties that perform above the average and are stable in multiple environments and, second, suggest using varieties with low stability in particular environments by exploiting their adaptability under specific environmental conditions [
19]. The GGE biplot model [
7,
20] offers a comprehensive visual assessment of the entire genotype–environment interaction, presenting a biplot that encapsulates both average yield performance and stability. GGE effectively filters out the noise originating from the environmental main effect (E) and highlights the two crucial components: genotype effects (G) and G × E interactions. The genotype’s value is visually represented using scores obtained from principal component analysis. Genotypes positioned closer to the performance line are deemed more stable compared with those located farther away. Moreover, the model quantifies a genotype’s distance from the “ideal genotype”, which represents the one with the highest mean performance and absolute stability, effectively occupying the central position within a series of concentric cycles. Although such an ideal genotype is rarely found in practice, it is a reference point for evaluating genotypes. The concentric circles, centered on the ideal genotype, facilitate the visualization of the relative distances between all and ideal genotypes. The plot distance between any given genotype and the ideal genotype can be used to measure its desirability [
7,
20].
Abiotic components that describe an environment are the non-living factors that affect the growth of living organisms, like plants, which grow in the ecosystem. Climatic parameters like rain, precipitation, and temperature are significant factors that characterize an environment [
21]. Additionally, wind conditions, moisture, and light are some characteristics of the environment that affect grain yield [
22]. Furthermore, soil parameters, like pH [
21] and salinity [
22], affect the adaptability of specific cereal genotypes. In summary, an agricultural system incorporates all the parameters mentioned above, which can affect crop growth and biomass accumulation and turn a percentage of biomass into seed production. Similarly, the agronomical practices adopted affect productivity and the efficient use of resources, biomass accumulation, and eventually higher grain yield. As a result, agricultural systems where different farming practices are applied are considered different environments.
In the last years, the Genotype (G) × Environment (E) interaction has been escalated to a triple interaction that includes the Management (M) factor, which is considered a third factor [
5], or in some cases, is incorporated into the Environment factor. Identifying the ideal sowing window for durum wheat is now an essential management practice with a principal role in high yield, which is highly associated with the grain-filling period [
23]. Similarly, the agronomical practices adopted affect productivity and the efficient use of resources, biomass accumulation, and, eventually, grain yield. As a result, agricultural systems where different farming practices are applied can represent different environments [
24]. Nitrogen fertilization is another management practice crucial for grain yield and quality. The interaction between genotype and N availability should be exploited to produce durum wheat, a crop with quality requirements [
25]. Unlike other cereal crops, durum wheat cultivation is not currently investigated for plant density, so this question is solved. Regarding irrigation, more and more countries in the Mediterranean region and in other areas where durum wheat is cultivated with supplemental irrigation focus their research and the strategy of their breeding programs on drought-tolerant genotypes in a period where intense climatic events are escalating [
26,
27].
This study aimed to identify the most productive and stable genotypes across Mediterranean farming systems, including biological and low/high productivity farming systems, through a comparative examination of the yield and analyses of the GGE biplot alongside fifteen parametric and non-parametric stability models.
3. Results
3.1. Analysis of Variance and Mean Performance
Based on the ANOVA (
Table 3) for all the environments, the grain yield of durum wheat was significantly affected by environment, genotype, and their interaction (
p < 0.001 for all effects). Even the partial ANOVA of high-yielding environments (E4–E6) and low-yielding environments (E1–E3) indicated that both G, E, and their interaction significantly affected the yield. Another interesting finding from this analysis is that for all environments, the highest percentage of variation was explained by G × E (43%), followed by E (41.7%) and G (15.3%). However, in low-yielding environments, E, G × E, and G were more similar, explaining 39%, 37.2%, and 23.3% of the variation, respectively. In the case of high-yielding environments, the G × E interaction explained 51.9% of the variation, G explained 46.6%, and E explained 1.5%.
The studied durum wheat genotypes across the environments showed significant imprints of productivity (
Table 4). Based on the results from all the examined environments, G5 had the highest grain yield (3.77 ton.ha
−1), followed by G6 (3.61 ton.ha
−1), G8 (3.47 ton.ha
−1), G12 (3.37 ton.ha
−1), and G2 (3.33 ton.ha
−1); however, there were no statistically significant differences between these best-performing genotypes. The genotypes with the lowest overall productivity were G1 (2.31 ton.ha
−1), G9 (2.76 ton.ha
−1), G16 (2.76 ton.ha
−1), and G11 (2.78 ton.ha
−1).
The results of the productivity of the genotypes in each environment are particularly interesting. In the case of E1 (low-yielding environment), G8 (4.31 ton.ha−1), G13 (4.13 ton.ha−1), and G3 (4.13 ton.ha−1) had the highest yields, but with no significant differences between them, while genotypes G1 (1.55 ton.ha−1) and G10 (2.15 ton.ha−1) recorded the lowest yields. In E2, an organic low-yielding environment, G13 (2.75 ton.ha−1), G1 (2.64 ton.ha−1), G5 (2.62 ton.ha−1), G2 (2.46 ton.ha−1), and G4 (4.40 ton.ha−1) gave the highest yields, with no statistically significant differences between them. The lowest grain yields were recorded for G6 (1.73 ton.ha−1) and G10 (1.84 ton.ha−1). In E3, a late-sowing low-yielding environment, G6 (3.08 ton.ha−1) had the highest grain production, with significant differences compared with all the other genotypes, followed by G3 (2.71 ton.ha−1) and G5 (2.67 ton.ha−1). Genotypes G1 (1.98 ton.ha) and G16 (1.98 ton.ha−1) recorded the lowest grain yields. In E4, a high-yielding environment, G14 had the highest grain production (5.27 ton.ha−1), with statistically significant differences compared with all the other genotypes, i.e., G2 (4.56 ton.ha−1), G6 (4.46 ton.ha−1), G11 (4.41 ton.ha−1), and G16 (4.31 ton.ha−1), followed by a very good performance and no statistically significant differences between them. In this environment, the lowest grain yields were recorded for G1 (2.27 ton.ha−1) and G9 (2.58 ton.ha−1). In E5, a high-yielding environment, G5 had the highest grain production (5.73 ton.ha−1), with statistically significant differences compared with all the other genotypes, i.e., G4 (4.62 ton.ha−1), G3 (4.43 ton.ha−1), G12 (4.30 ton.ha−1), G6 (4.10 ton.ha−1), and G2 (4.06 ton.ha−1), which had no statistically significant differences between them, while genotype G15 (2.43 ton.ha−1) had the lowest grain yield. Finally, in E6, a high-yielding environment, genotypes G5 (5.73 ton.ha−1), G6 (5.24 ton.ha−1), G12 (4.30 ton.ha−1), G10 (4.01 ton.ha−1), and G4 (3.96 ton.ha−1) recorded high grain yields, while genotypes G11 (2.12 ton.ha−1) and G1 (2.19 ton.ha−1) had low grain yields.
3.2. AMMI and GGE Biplot Analyses
Based on AMMI data analysis, two mega-environments, including all six environments (three low-productivity and three high-productivity fields), were identified, as depicted in
Figure 1. E1, E2, and E3 form the first mega-environment, representing the low-productivity environments, while E4, E5, and E6 form the second mega-environment, representing the high-productivity environments. Based on the position of the two environment subgroups, genotypes such as G1 and G9 are better suited to low-productivity environments, while genotypes such as G14, G6, G5, and G12 are better adapted to high-productivity environments.
Combined and partial GGE biplot analyses explained 73.06% (
Figure 2), 92.42% (
Figure 3), and 92.48% (
Figure 4) of the total variability in all, low-, and high-productivity environments, respectively. Genotypes G6 and G12 were very close to the stability line in the combined analysis of all the environments.
The GGE biplot identified G8 as the best-performing genotype for the low-productivity fields, which also showed high stability. G3 and G15 followed, with inferior productivity and stability. As for the high-productivity environments, GGE biplot analysis indicated that G6 was the genotype closest to the ideal, as it had the best performance and high stability. This was followed by G12 and G5, with the latter showing comparatively low stability.
3.3. Prominent Genotypes Based on Multiple Stability Parameters
Several parametric and non-parametric indices were calculated to evaluate the genotypes based on their adoption in different environments.
Table 5 displays the genotypes found in each index’s first five and last five positions based on the ranking for each respective statistic. In the total evaluation of all environments, G8, G2, and G7 dominated the top 5 ranking, with frequencies of 18/18, 16/18, and 13/18, respectively, identifying them as the most stable genotypes. For the bottom five rankings, G11 and G1 had the worst positions, with frequencies of 18/18 and 14/18, respectively. In the subgroup with the three high-productivity environments, G12, G8, and G6 were the genotypes with the highest presence in the top 5 ranking, with frequencies of 16/18, 16/18, and 12/16, respectively. In the bottom five ranking, G16 and G11 had frequencies of 16/18 for all indices, followed by G3 with 12/18. For the low-productivity subgroup, G2, G9, and G13 were the genotypes with the highest presence in the top 5 rankings, with frequencies of 17/18, 14/18, and 10/18, respectively. Considering the bottom five ranking, genotypes G11 and G10 each had a high frequency of 15/18, followed by G5 (14/18) and G3 (11/18).
Table 6 displays the rank correlations between the statistical measures assessed across the six environments. The mean grain yield demonstrates a strong positive correlation with measures of GGE,
,
KR,
S(3),
S(6),
NP(2),
NP(3), and
NP(4). Among the statistical measurements that were estimated, GGE biplot,
, and
consider yield and G × E interaction for cultivar evaluation, while the remaining statistical measures mainly consider stability. GGE analysis,
, and
KR were strongly and positively correlated, but had no correlation found with
. Moreover, strong positive correlations were recorded for: (1)
σ2i with all except
S(3) and
S(6), (2)
CVi with
s2di, (3)
Wi2 with all except
S(3),
S(6), and
NP(4), (4)
θ(i) with all except
S(3),
S(6), and
NP(4), (5)
S(1) with
S(2) and
NP(4), (6)
S(2) with
S(3) and
NP(1), (7)
S(3) with
S(6),
NP(1),
NP(2),
NP(3), and
NP(4), (8)
S(6) with
NP(2),
NP(3), and
NP(4), and (9) between
NP(2),
NP(3), and
NP(4).
4. Discussion
Durum wheat stands as the predominant cereal crop in the Mediterranean basin. Environmental and climatic elements, with notable sensitivity to water availability and elevated temperatures during the grain-filling period, can significantly affect the productivity of durum wheat [
6,
37,
38]. Limited water availability and adverse conditions underline the significance of identifying the most productive and stable genotypes across different Mediterranean farming systems. Selection of the most appropriate crop variety stands out as the most cost-effective and efficient method for enhancing the productivity of a Mediterranean agricultural system. This strategic choice enables increased and more consistent production outcomes without a parallel rise in production costs, thereby maximizing resource utilization. In this multi-environment trial, durum wheat varieties such as Meridiano (G5), Mexicali-81 (G6), Simeto (G8), Elpida (G12), and Canavaro (G2) had mean grain yield productivities ranging from 3.33 to 3.77 ton.ha
−1 based on the total evaluation of all environments, an outcome that was more or less expected. In the case of low-productivity environments, Mexicali-81 (G6) and Simeto (G8) were high-performing genotypes, although the relatively new varieties Maestrale (G3) and Zoi (G13), released in 2004 and 2011, respectively, also had high productivity. With tonnage above 5.00 tonnes per hectare, Meridiano (G5), Mexicali-81 (G6), and Thraki (G14) had high grain yield production in the high-productivity environments.
In parallel, the adoption of organic farming is on the rise globally. The European Commission (EC) has set an ambitious goal of converting a minimum of 25% of the European Union’s agricultural land to organic farming by 2030 [
39]. However, the challenge is that modern commercial varieties are better suited for high-input farming, leaving a gap in cultivars adaptable to organic or low-input agriculture. One potential solution to this issue in wheat farming could be found in landraces, as they have demonstrated remarkable adaptability and productivity in low-input farming systems or under organic conditions. In our research, the landrace Limnos G1 exhibited high productivity (2.64 ton.ha
−1) in the organic environment (E2) and ranked among the top three most productive cultivars; thus, proposing the cultivation of the Limnos landrace in organic farming systems could be a viable option. Furthermore, integrating landraces into agricultural systems has the added benefit of contributing to biodiversity expansion. This is particularly significant in wheat breeding, as landraces play a crucial role in reducing the genetic erosion of cultivated wheat. By broadening the genetic base in breeding programs, landraces provide essential traits that help crops cope with the challenges of changing environments, such as those influenced by climate change [
40,
41]. In the context of an autogamous crop like wheat, landraces represent a mix of pure lines. This characteristic can lead to the development of genetically stable genotypes or commercial cultivars relatively quickly, making them suitable for organic or low-input environments. Numerous successful examples exist where intensive breeding programs incorporating effective plant selection under ultra-low plant densities have harnessed the variations in landraces/cultivars in wheat [
42,
43] and other autogamous species [
44,
45].
The timing of sowing stands out as a critical factor influencing cereal production and quality, as emphasized by McLeod et al. [
46]. The ideal sowing period typically falls around November in the Mediterranean Basin countries. Despite this, farmers frequently face challenges, leading to delayed sowing attributed to excessively wet or dry soil conditions. Notably, delaying sowing beyond the optimal agronomically recommended period often decreases yield [
46] since it exposes durum wheat crops to potentially adverse conditions such as high air temperature and deficient soil moisture during the critical period of anthesis and grain filling. This situation becomes particularly relevant when considering climate change projections, as Porter and Gawith hypothesized [
47]; therefore, evaluating cultivars in late-sowing environments becomes crucial. In Environment 3 (E3), characterized by late-sowing and low yields, Mexicali-81 (G6) exhibited the highest grain production, showing significant differences from all the other genotypes. Following closely were Meridiano (G5) and Maestrale (G3), with Meridiano (G5) also demonstrating notable performance.
Based on Simmonds [
48], genotypes contributed 25–40%, the environment contributed 30–60%, and G × E interaction contributed the remaining 15–25%; thus, multi-environment field evaluation is a necessary practice that has been followed for the selection of high-yield best-adapted genotypes for many types of cereals cultivated for their grains, including durum wheat [
49] and other species such as barley [
50] where kernel yield is significant. Moreover, considering the effect of G × E on wheat yield per ha increase [
48], the study of this parameter is of high importance since genotype rank changes from one environment to another. However, the concept of crossover interaction hampers the discovery of superior genotypes [
51]. Several statistical tools have been proposed for G × E interpretation and the study of crop stability. However, a complete visual evaluation of all G × E interaction aspects is possible with the GGE biplot model [
7,
52].
This study aimed to compare the utilization of the GGE biplot alongside fifteen parametric and non-parametric stability models in assessing durum wheat genotypes. Another target of the study was the identification of the most productive and stable genotypes across different Mediterranean farming systems, including biological and low/high productivity farming systems. The statistical metrics mentioned above can be categorized into two main groups. The first set considers the inherent genetic characteristics (G) and the interactions between genotypes and their respective environments (GE). Consequently, they offer a comprehensive assessment considering yield performance and genotype stability under varying conditions. Conversely, the second set of statistics emphasizes genotype–environment interactions (GE) while primarily focusing on stability as a critical factor in genotype evaluation. A supplementary but essential parameter for evaluating a genotype is its stability under different environmental conditions. The stability statistics indicated that, in general, in the overall evaluation, Simeto (G8), Canavaro (G2), Monastir (G7), and M. Aurelio (G4) were the most stable genotypes across different environmental conditions, while Anna (G11) and Limnos (G1) had significant fluctuations in their performance across the different environments. In the low-performing environments, Canavaro (G2) and Svevo (G9) were the most stable genotypes, while Anna (G11) and Vendeta (G10) had the lowest stability. In the high-yielding environments, Elpida (G12), Simeto (G8), and Mexicali-81 (G6) showed the highest stability. On the contrary, Sifnos (G16) and Anna (G11) were exceptionally unstable. Adapting specific genotypes to high- and low-yielding environments is a valuable tool for evaluating varieties [
53]. Selection of the most suitable genotype and fertilizer application is now considered a precision farming component [
54].
All statistical tools that consider both G and G × E, including GGE biplot,
Pi, and
KR, were strongly correlated with each other and yield, apart from
. Similar results in other environmental evaluations of wheat were found by Smutna et al. [
53]. These findings align with prior research indicating that these tools contribute to the agronomic concept of stability [
18,
55].
was positively correlated with
σ2i,
CVi,
s2di,
Wi2, and
θ(i), parameters that do not correlate with yield, thus demoting the highest yielding genotypes. Similar results were found by other wheat researchers [
53]. Moreover, the
Si,
Npi, and
CVi statistics are influenced by both yield and stability. Consequently, they promote genotypes that display typical stability and may not excel under optimal conditions [
18,
56].
Extensive crossover genotype–environment (GE) interactions imply that the current genotype evaluation procedures may not accurately reflect the prevailing conditions. However, utilizing the GGE biplot has proven efficacious in facilitating the straightforward identification of superior genotypes. The GGE biplot model comprehensively considers both genotypic (G) and genotype–environment (GE) interactions, presenting a graphical representation that serves as a valuable methodology for assessing the adaptability and stability of genotypes across diverse environments. This approach offers a visual and integrated evaluation of performance and stability [
7,
20,
24,
52].
GGE biplot analysis of all environments, which evaluated both average yield performance and stability, indicated that Mexicali-81 (G6) and Elpida (G12) are the varieties closest to the ideal genotype. Simeto (G8) and Maestrale (G3) had good performance in the low-yielding environments, while Mexicali-81 (G6) and Elpida (G12) were the varieties closest to the ideal genotype in the high-yielding environments, validating the results of the evaluation of all the environments. GGE biplot analysis has been used to evaluate genotypes in wheat, with successful results [
57] and adoption to heat stress and irrigation environments [
58]. Considering both performance and stability through visual analysis constitutes a significant advantage, instilling confidence in the agronomists in the decision-making process of promoting superior genotypes. Consequently, this approach aids breeders in identifying genotypes that merit release for cultivation.
Evaluation of all environments for a combination of high yield and stable production showed that the best genotypes were G8 (Simeto), G2 (Canavaro), and G12 (Elpida). Similarly, G12 (Elpida), G8 (Simeto), and G6 (Mexicali-81) were the best genotypes in the subgroup with the three high-productivity environments, followed by G2 (Canavaro), while G2 (Canavaro), G13 (Zoi), and G8 (Simeto) were the best genotypes in the low-productivity environment subgroup.
5. Conclusions
Evaluation of sixteen durum wheat genotypes in six multi-environmental trials confirmed that a detailed study of the Genotype × Environment interaction could increase the yield of durum wheat farming systems. The stability models used in the study resulted in a diverse ranking of genotypes. A very interesting result was that many statistical tools that consider both G and G × E, such as GGE biplot, Pi, and KR, showed strong correlations with each other and with yield. This aided in the identification of the most suitable genotypes. This statistical set centered on the agronomic aspect of stability. On the other hand, statistical parameters focusing on the statical aspect of stability, such as , σ2i, Cvi, s2di, Wi2, KR, and θ(i), downplayed the significance of the highest yielding genotypes. Ultimately, the GGE biplot proved to be a highly effective statistical tool for assessing the worth of wheat genotypes in terms of both specific and general adaptation. It achieved this while preserving yield capacity and providing graphical representations. An interesting outcome emerged regarding the landrace Lemnos, suggesting its potential use in organic agriculture. Finally, our study revealed that some commercial cultivars exhibited better performance under late-sowing conditions, indicating their tolerance of more adverse conditions during the critical grain filling period, such as higher air temperature and deficient soil moisture, i.e., conditions correlated with climate change.