Selection of High Yield and Stable Maize Hybrids in Mega-Environments of Java Island, Indonesia

: Determination of grain yields of stable and high-yielding maize hybrids in a wide environment requires high accuracy. Many stability measurement methods have been used in multi-environment experiments. However, the relationships among the different methods are still difﬁcult to understand. The objectives of this study were to 1. Identify the effect of growing season and location (Environments = E), hybrids (Genotypes = G), and their interactions (GEIs) on grain yields; 2. Select high-yielding and stable maize hybrids in a wide range of environments; 3. Determine the relationship between each stability estimation; and 4. Determine the mega-environment of maize hybrid and identify the best locations for testing. Field experiments were conducted at ten locations in Java Island, Indonesia, for two growing seasons using a randomized completed block design with three replications. The experimental results showed that the main effects of the growing season, location, hybrid, and GEIs, signiﬁcantly affected maize hybrid yields. Stability estimations of TOP, S (3) , S (6) , NP (2) , NP (3) , KR, NP (4) , CV i , and b i , belong to the concept of dynamic stability that can be used to select maize hybrids in favorable environments, while other estimations were classiﬁed as in the static stability. Three maize hybrids were successfully selected, with high and stable yields based on numerical and visual stability estimations, namely SC2, SC7, and SC9. The three hybrids can be used as candidates for sustainable maize development programs. The dry season, the rainy season, and the combination of two growing seasons produced three mega-environments. GJRS and KARS were the most discriminative environments. Both environments can be used as favorable environments for selecting the ideal maize hybrid.


Introduction
Maize is an important carbohydrate source. It also contains other important nutrients for the human body, such as protein and minerals [1,2]. This commodity, is among the most important staple foods after rice and wheat. Nowadays, the demand for maize is high, but the existing production is still unable to fulfill these demands. Several factors limit maize production, i.e., varieties; environmental conditions including cropping system, location, season, and infection due to insects and pathogens; and interaction between genotype and environment [3,4]. Due to the importance of maize and the limitations in their production, the development of maize hybrids possessing high-yield and adaptability to environmental changes is required.
Evaluation of maize hybrids in various locations and growing seasons are required to select high-yield and stable varieties. Thus, it can determine the mega-environment, which can effectively select a representative location for an efficient selection. However, selection under these various environments, is very complex due to genotype and environmental interactions (GEIs) [5][6][7]. The emergence of GEIs in multi-environmental experiments is due to unpredictable macro-and micro-environmental influences such as temperature, rainfall,

Field Experiments and Data Collection
Field experiments were conducted at ten locations in Java Island, Indonesia, for two growing seasons (Table 2, Figure 1). The experiment used a randomized completed block design which was repeated three times. Each hybrid was planted on a plot measuring 3 × 5 m, with a spacing of 0.75 × 0.2 m. The data was gathered at harvest, 93 days after planting following the standard Descriptor for Maize [38]. Sowing and harvesting were done manually. The number of plants harvested in each plot was 100. The yield of each hybrid in each experimental plot was converted in ton.ha −1 .

Data Analysis
Combined analysis of variance (ANOVA) was used to estimate the GEIs. The statistical equation was as follows: where Yopqr is the value of line o in plot r, and the value in location p of each replication q;

Data Analysis
Combined analysis of variance (ANOVA) was used to estimate the GEIs. The statistical equation was as follows: where Y opqr is the value of line o in plot r, and the value in location p of each replication q; µ is the grand mean; G o is the effect of line o; E p is the effect of the environment p; GE op is the effect of genotype by environment interactions on line o and environment p; R q(p) is the effect of replicate q on location p; B r(q) is the effect of replication q on plot r; and ε opqr is the error effects from line o in plot r and repeat q of location p, respectively. The combined ANOVA was calculated using the R program. Yield stability was estimated based on parametric and non-parametric measurements. Details of the stability measurements are presented in Table 3. The stability of grain yields based on parametric and non-parametric measurements was analyzed using the online software STABILITYSOFT [39].
The graphical stability of grain yields and determination of discriminative environment and mega-environment was analyzed by GGE biplot with the following equation [25]: where Y mn ; µ m ; β n ; k; λ o ; α mo and γ no ; ε mn are the performance in location 'n' from line 'm'; overall average grain yield; the influence of location 'n'; the number of primer components; the singular value from primer component 'o'; the value of line 'm' and location 'n' for primer component 'o'; and the error of the line 'm' in location 'n,' respectively. Table 3. Formula of parametric and non-parametric stability measurements.

Formula Source
Non-parametric stability measurements Huehn; Nassar and Huehn [19,20] S Kang [21] The computation of the TOP-rank is based on scoring of genotypes as 'Top', 'Mid' or 'Low' within each environment. The genotypes that are frequently occurred in the 'Top' third are considered to be stable.
Fox [22] where, X ij : grain yield total in of the i th hybrid in j th environment, X i. : average of the grain yield total from i th hybrid at all (sixteen) environments, X .j : mean of the grain yield in the j th environment, X ... : average of the grain yield total, p and q: numbers of environments and hybrids; SD l : standard deviation of GEIs. r ij : stability rank of the i th hybrid in the j th environment; r i. : average rank if i th hybrid in all environments; and N: number of environment. r * ij : stability rank of the i th hybrid in the j th environment (adjusted data); M * di : adjusted data (median rank); M di : unajusted data (median rank's of the same parameters). RGY: rank of grain yield; Rσ 2 i = Rank of Shukla stability measurement.
The R program was used to visualize the distribution pattern of the hybrids and environments tested in the dry season, rainy season, and the average of both. Spearman rank correlation and Principal Component Analysis (PCA) were used to estimate the relationship between stability measurements and classify them into clear groups. The SPSS 19th software was used to analyze correlation and PCA [40].
The combined stability analysis includes parametric and non-parametric stability measurements. The stability of maize hybrids based on ranks of parametric and nonparametric stability estimation was combined by using Hierarchical Clustering Analysis (HCA). The SPSS 19th software was used to estimate HCA [40,41].
The sustainability index (SI) was estimated by the following formula used by [42]: where Y is the mean performance of a maize hybrid, σn is the standard deviation, and YM is the best performance of a maize hybrid in any environment. The SI values were classified arbitrarily into five groups, i.e., very low (up to 20%), low (21% to 40%), moderate (41% to 60%), high (61% to 80%), and very high (above 80%) [43]. SI was calculated using Microsoft Excel 2013.

Genotype by Environment Interactions of Maize Hybrids Yield
The combined analysis of variance (ANOVA) for maize hybrids evaluated in ten locations during two growing seasons in Java island was presented in Table 4. There was a very significant variation (p < 0.01) in yields among hybrids (genotypes), environment (season, location, season × location), and their interactions (Genotype × season, Genotype × Location, and Genotype × season × Location) ( Table 4). The highest difference was shown by the interaction effect of growing season and location (L × S) of 40.13%, whereas the main effect of genotype (hybrid) accounted for 14.59%, the location was 12.55%, and their interaction (GEIs) was 12.69%. This result confirmed that the hybrid was a significant factor in environmental interactions in this experiment. The significant variation of the main effects (genotypes and environment) and their interactions indicated differences in the hybrid performance under a broad environment for maize production on Java Island. It was presented in Table 2 that the range of average yield in 20 environments was from 5.46 t.ha −1 (LBDS) to 11.75 t.ha −1 (NNRS) ( Table 2). The average data for two growing seasons recorded the highest average yield at Jatinangor, Sumedang, West Java (9.68 t.ha −1) and the lowest at Arjasari, Bandung, West Java (6.40 t.ha −1) ( Table 2).
In some locations, the average yield was higher in the dry than in the rainy season (Gumukmas, Ngadiluwih, Jogonalan) ( Table 2). This is because the land used during the dry season in the three locations is ex-paddy land; therefore, the condition of the land is still wet in the dry season. Alibu et al. [44] reported that wet and humid land conditions in the dry season showed excellent maize yields. Therefore, maize planting on wetlands or ex-paddy during the dry season is preferable.
The effect of GEIs frequently occurred for maize yields in multi-environmental experiments [1]. The difference in yield performance of maize hybrids was probably due to differences in genetic background and various environmental conditions. The hybrids used resulted from directed crosses between two parental lines with a far genetic background [37]. Thus, the environmental factors, i.e., locations, seasons, and cultivation systems, affected yield performances [10,45]. The percentage of environmental influence, which is high on maize yields, indicated that the environment for maize production in Java Island is very broad. Variances in environmental conditions for maize production can lead to differences in yield and yield quality of maize hybrid [10,33,46,47]. The response of maize hybrids to the various environments indicates the importance of GEIs.
The GEIs effect has a great impact on the plant selection process. The emergence of the effect of GEIs in multi-environment experiments makes the selection process complicated and less efficient [7,27,48,49]. Breeders need to allocate more resources, time, and money to evaluate a set of potential superior genotypes. In addition, more locations need to be surveyed to establish the multi-locations field of evaluation. Determination of the megaenvironment, establishing a representative location, and applying stability methods for testing and analysis are required to effectively select high-yielding and stable genotypes in a wide range of environments.

Selection of High-Yielding and Stable Maize Hybrids in a Wide Environment Using Combined Stability Analysis
Stability analysis of maize yields using parametric measurements for every maize hybrid in twenty environments (ten locations and two planting seasons) were presented in Table 5. According to Eberhart and Russell [11], the stability of each maize hybrid was determined by its regression coefficient (bi) and deviation of variance (S 2 di). An estimation of b i = 1 and a low estimate of S 2 di indicates a very stable hybrid. The SC2, SC6, SC7, and SC9 hybrids had b i values that were not significantly different from one (1), where the SC6 hybrid produced yields lower than the overall average, while SC2, SC7, and SC9 hybrids were higher than the overall average. Estimation of S 2 d i indicated SC5, SC8, SC6, and SC9 as maize hybrids possessing the lowest values. Based on the linear regression estimation, hybrids with values of b i = 1 and S 2 d i = 0 were the most stable, so SC6 and SC9 were the most stable maize hybrids based on this measurement. Average grain yields for the hybrids tested in 20 environments ranged from 5.46 t.ha −1 to 11.75 t.ha −1 , with SC2 and SC9 maize hybrids having the highest average yields and SC5 and SC6 the lowest ( Table 2). Based on the stability ranking of W i 2 , σ 2 i , CVi, and θ (i) estimation, SC8 was determined as the most stable maize hybrid, followed by SC5 and SC6. Of the three selected hybrids, only SC8 had above-average yields. The stability estimation of S 2 di selected SC5 as the most stable, followed by SC8 and SC6. Stability estimation of Di also revealed SC5 as the most stable maize hybrid, followed by SC8 and SC9. The maize hybrid of SC3 was the highest yield performance hybrid in all test environments. Stability measurement of bi and θ i selected SC2 as the most stable maize hybrid, followed by SC6 and SC7 for bi and SC1 and SC3 for θ i , where the SC2 maize hybrid had an above-average overall performance.
Non-parametric stability estimation is presented in Table 6. It was shown that each hybrid had different potential in terms of stability. In non-parametric measurements, SC2 hybrids were designated as the most stable hybrid by stability measurements of S (3) , S (6) , NP (2) , NP (4) , and TOP. The SC5 hybrid was indicated as the most stable by S (1) and S (2) measurements. The SC8 hybrid was indicated as the most stable by NP (1) and KR measurements. The SC9 hybrid was revealed as the most stable maize hybrid by stability measurements of NP (3) and KR. This maize hybrid also has above-average yield performance.  According to Ahmadi et al. (2015) [26], the selection of stable genotypes with one stability measurement was considered less effective and accurate. Similar results were also revealed by several researchers who used various stability measurements to select stable and high-yielding genotypes, including barley [28], maize [33], sweet potatoes [7,9], turmeric [50], and soybeans [35,51]. Every stability estimation generally selected a different hybrid as a stable genotype. However, several stability estimations selected similar hybrids as stable hybrids. These stability estimation included CV i , θ i , S (1) , S (3) , S (6) , NP (3) , and NP (4) which determined SC3 as a stable hybrid. In addition, stability measurements had the same output in terms of ranking the stability of all maize hybrids, namely Wi 2 , σ 2 i , and θ (i) . In this case, measurements with the same stability ranks can select stable genotypes [28].
Applying combined stability by the use of parametric and non-parametric stability estimation can increase the accuracy of the maize hybrid selections. The application could help select high-yield performance and stability of potential genotypes in a wide environment based on a single measurement [7,26]. Some researchers use the average sum rank (AR) to determine the stability of the tested genotypes, wherein the genotype with the smallest AR value was determined as the most stable genotype [9,[26][27][28]. In this study, SC9 was identified as a hybrid with the smallest AR, followed by maize hybrids of SC9, SC7, and SC2. These three hybrids also had high average yields.
The stability of maize hybrids based on ranks of parametric and non-parametric stability estimation was combined using the Hierarchical Clustering Analysis (HCA) (Figure 2). Based on this HCA, maize hybrids were divided into three main clusters, namely: 1. stable low yield cluster consisting of SC5 and SC6 maize hybrids; 2. unstable medium yield cluster consisting of SC1, SC3, and SC4 maize hybrids; 3. stable high yield cluster consisted of SC2, SC7, SC8, and SC9 maize hybrids. The first group was not recommended since they have a low yield. The second group can be used as hybrids with medium yield performance in specific environments. Ruswandi et al. (2020) [1] mentioned that maize hybrids with highyield performance in a specific environment could be superior hybrids in this particular area. A way to increase the income/economics of maize farmers in certain environments is by utilizing the potential of these genotypes. According to Maulana et al. (2020) [7], high agricultural product yields will impact the community's economy. The third group of maize hybrids was considered the most ideal group because of its yield performance and wide adaptability [26,51]. Thus, the third group was identified as more stable with high yield performance in a wide environment based on parametric and non-parametric stability measurements. Similar studies revealed that combined stability by measuring parametric and non-parametric stability could successfully determine stable and high-yielding genotypes in various commercial crops and a wide environment [27,28,30,31,34,35]. This current study found that the combined stability analysis can be effectively used to select high-yielding and stable maize hybrids in various crop environments and seasons on Java island. environments is by utilizing the potential of these genotypes. According to Maulana et al. (2020) [7], high agricultural product yields will impact the community's economy. The third group of maize hybrids was considered the most ideal group because of its yield performance and wide adaptability [26,51]. Thus, the third group was identified as more stable with high yield performance in a wide environment based on parametric and nonparametric stability measurements. Similar studies revealed that combined stability by measuring parametric and non-parametric stability could successfully determine stable and high-yielding genotypes in various commercial crops and a wide environment [27,28,30,31,34,35]. This current study found that the combined stability analysis can be effectively used to select high-yielding and stable maize hybrids in various crop environments and seasons on Java island.

Figure 2.
Maize hybrids were grouped based on parametric and non-parametric stability ranks at ten locations for two growing seasons.

The Relationship between Parametric and Non-Parametric Stability Measurements
To determine the relationship between the different stability measures and combine them into clear groups, Principal Component Analysis (PCA) was used. The first four PCs with eigenvalues >1 resulted in a cumulative value of 96.37% of the total variation between parametric and non-parametric measurements ( Table 7). The first two components were used to visualize the PCA biplot because they had the highest variability values (PC1 = 45.51% and PC2 = 36.02%) and eigenvalues of 8.65 and 6.84, respectively, as shown in Figure 3. Parametric and non-parametric measurements were classified into four main groups, namely: the first group consisted of NP (1) , S (2) , Di, Wi 2 , S 2 di, θ₍ᵢ₎, and σ 2 ᵢ; the second group consisted of S (1) ; the third group consisted of yields (Y) with TOP, S (3) , S (6) , NP (2) , NP (3) , KR, NP (4) , CVi, and bi measurements; and the fourth group consisted of θᵢ measurement.  Maize hybrids were grouped based on parametric and non-parametric stability ranks at ten locations for two growing seasons.
Graphical visualization based on PCA biplots was used to understand the relationship between the measurement and the stability concept ( Figure 3). PCA biplots were taken from the highest values of the first two PCs (Table 7). Based on PCA analysis, all stability measures were classified into four groups. The first group consisted of NP (1) , S (2) , D i , W i 2 , S 2 d i , θ (i) , and σ 2 i ; the second group consisted of S (1) ; the third group consisted of yields (Y) with TOP, S (3) , S (6) , NP (2) , NP (3) , KR, NP (4) , CV i , and b i measurements; and the fourth group consisted of θ i measurement. The first two and the fourth groups represent the concept of static stability, so they can be used to select hybrids in less favorable environments [52]. The third group showed that the measures were positively and significantly correlated based on Spearman's rank correlation to maize hybrid yields, providing a measure of dynamic stability. They can be used to recommend ideal maize hybrids under favorable environmental conditions [27,28].

Mega-Environment of Maize Hybrid and Identification of the Best Locations
GGE biplot analysis for dry, rainy, and combined seasons are presented in Figures 4-6, respectively. Based on the GGE biplot of the dry season (Figure 4), 60.92% of the total variation for grain yield was explained by PC1 (40.19%) and PC2 (20.73%). The 'which won where/what' pattern showed five sectors for ten locations with different winning (vertex) hybrids. The vertex hybrids were SC2 in KADS, LBDS, BBDS, NKDS, PPDS, and GJDS; SC3 hybrid in NNDS and ASDS; SC8 hybrid in JKDS and JTDS, while other vertex hybrids, namely SC1 and SC5 did not have locations that fall in the sector. For the rainy season, with two PCs accounting for 63.30% of the total variation for grain yield (PC1 = 43.18% and PC2 = 20.12%), the GGE biplot revealed three mega-environments ( Figure 5). The first mega-environment consisted of NNRS, PPRS, GJRS, KARS, ASRS, JKRS, and BBRS, with the winning hybrid SC2. The second mega-environment includes LBRS and NKRS, with SC7 as the winning hybrid. The third mega-environment includes JTRS, with the SC4 and SC8 as the winning hybrids. SC6 is the vertex hybrid in the sector without an environment.
For the combined data in dry and rainy seasons, GGE biplots of ten locations in two growing seasons revealed that the first two PCs accounted for 54.14% of the total variation (PC1 = 37.61% and PC = 16.53%) ( Figure 6). Based on the biplot, there were five sectors with different winning hybrids (vertex). The vertex hybrids were SC1, SC7, SC2, and SC4. Figure 6 represents the three mega-environments. The first mega-environment consisted of JTDS and JTRS with the winning hybrid SC4. The second mega environment included BBRS, JKRS, GJRS, KARS, PPRS, NNRS, ASDS, NNDS, GJDS, NKDS, JKDS, KADS, LBDS, PPDS, and BBDS, with SC2 as the winning hybrid. The third mega-environment included ASRS, LBRS, and NKRS, with the SC1 and SC7 as the winning hybrids. SC5 and SC6 were vertex hybrids in the sector without environment, indicating that their yield performance was poor in all test environments in this study.
According to the 'ranking environment' pattern of the GGE biplot presented in Figure 7, the GJRS and KARS environments were ideal for testing because they were at the ideal point (small arrow). These two locations were ideal for selecting superior hybrids because they had high discriminating power and representation. The JTRS environment was farthest from the ideal point and close to the center of the biplot axis. This location provides little information about the maize hybrids tested, so they are unsuitable for testing. Other environments were close to the ideal point but were outside the first circle, so it was useful for selecting hybrids in specific environments.
The GGE biplot can provide an overview of the differences between hybrids and environmental characteristics tested in multi-environment experiments. One advantage of the biplot that showed the distribution pattern of hybrids and environments was "which won where/what biplot" [53]. One of the characteristics of this biplot was the presence of polygons that indicate the location of the hybrid being tested. Hybrids at the top of the polygon (vertex) have the highest yields in the environment in that sector. Another important feature of this pattern was the grouping of environments, which suggested the possibility of different mega-environments [1,29,54]. In this current study, it was shown that within each growing season, the sites fall into different groups, and the pattern of site grouping varied throughout the seasons. The first two PCs explained 54.14-63.30% of the total variability due to the effects of hybrid (G), environment (location and growing season), and their interactions (Figures 4-6). The GGE biplot depicted the distribution of hybrids and environments in each season, and the average of the two seasons showed two and three mega-environments. a measure of dynamic stability. They can be used to recommend ideal maize hybrids under favorable environmental conditions [27,28].

Mega-Environment of Maize Hybrid and Identification of the Best Locations
GGE biplot analysis for dry, rainy, and combined seasons are presented in Figures  4-6, respectively. Based on the GGE biplot of the dry season (Figure 4), 60.92% of the total variation for grain yield was explained by PC1 (40.19%) and PC2 (20.73%). The 'which won where/what' pattern showed five sectors for ten locations with different winning (vertex) hybrids. The vertex hybrids were SC2 in KADS, LBDS, BBDS, NKDS, PPDS, and GJDS; SC3 hybrid in NNDS and ASDS; SC8 hybrid in JKDS and JTDS, while other vertex hybrids, namely SC1 and SC5 did not have locations that fall in the sector.  a measure of dynamic stability. They can be used to recommend ideal maize hybrids under favorable environmental conditions [27,28].

Mega-Environment of Maize Hybrid and Identification of the Best Locations
GGE biplot analysis for dry, rainy, and combined seasons are presented in Figures  4-6, respectively. Based on the GGE biplot of the dry season (Figure 4), 60.92% of the total variation for grain yield was explained by PC1 (40.19%) and PC2 (20.73%). The 'which won where/what' pattern showed five sectors for ten locations with different winning (vertex) hybrids. The vertex hybrids were SC2 in KADS, LBDS, BBDS, NKDS, PPDS, and GJDS; SC3 hybrid in NNDS and ASDS; SC8 hybrid in JKDS and JTDS, while other vertex hybrids, namely SC1 and SC5 did not have locations that fall in the sector.   Tables 1 and 2 for legends.
In the main mega-environment, SC2 was at the peak of the vertex in both dry, rainy, and combined seasons. Meanwhile, mega-environment (single) showed the difference between vertex hybrids. SC9 hybrids were always close to the center of the axis in each growing season and the combined one. This showed that SC9 tends to be stable in various environmental conditions; in other words, it has a small GEIs response. The two strategies for evaluating mega-environmental data (analysis of each growing season and its combination) showed that there was more than one mega-environment for maize breeding programs in various regions of Java island (Indonesia) and divided them into certain sub-regions. However, based on the combined data during two growing seasons, it was found that three mega-environments with different winning hybrids indicated the presence of maize hybrids specific to the mega-environment and the presence of substantial GEIs. The ideal hybrid is the hybrid with high yield and stability in multi-environment testing [27,28,33].
The hybrid SC2, followed by SC1, SC4, SC7, and SC9, were identified as ideal hybrids compared to others. This was confirmed by numerical measurements (parametric and non-parametric), where HCA separated SC2, SC7, and SC9 in the stable and high-yield groups (Figure 1). Based on these results, both measurement steps (numerical and graphical) produced the same pattern in selecting stable and high-yielding maize hybrids. This finding is similar to previous studies, which reported that stability measurements based on parametric, non-parametric, and GGE biplots resulted in the similar result for selection of stable and high-yielding genotypes, including sweet potato [27,54] and safflower [55].  For the rainy season, with two PCs accounting for 63.30% of the total variation for grain yield (PC1 = 43.18% and PC2 = 20.12%), the GGE biplot revealed three mega-environments ( Figure 5). The first mega-environment consisted of NNRS, PPRS, GJRS, KARS, ASRS, JKRS, and BBRS, with the winning hybrid SC2. The second mega-environment includes LBRS and NKRS, with SC7 as the winning hybrid. The third mega-environment includes JTRS, with the SC4 and SC8 as the winning hybrids. SC6 is the vertex hybrid in the sector without an environment.
For the combined data in dry and rainy seasons, GGE biplots of ten locations in two growing seasons revealed that the first two PCs accounted for 54.14% of the total variation (PC1 = 37.61% and PC = 16.53%) ( Figure 6). Based on the biplot, there were five sectors with different winning hybrids (vertex). The vertex hybrids were SC1, SC7, SC2, and SC4. Figure 6 represents the three mega-environments. The first mega-environment consisted of JTDS and JTRS with the winning hybrid SC4. The second mega environment included BBRS, JKRS, GJRS, KARS, PPRS, NNRS, ASDS, NNDS, GJDS, NKDS, JKDS, KADS, LBDS, PPDS, and BBDS, with SC2 as the winning hybrid. The third mega-environment included ASRS, LBRS, and NKRS, with the SC1 and SC7 as the winning hybrids. SC5 and SC6 were vertex hybrids in the sector without environment, indicating that their yield performance was poor in all test environments in this study.
According to the 'ranking environment' pattern of the GGE biplot presented in Figure 7, the GJRS and KARS environments were ideal for testing because they were at the ideal point (small arrow). These two locations were ideal for selecting superior hybrids because they had high discriminating power and representation. The JTRS environment was farthest from the ideal point and close to the center of the biplot axis. This location provides little information about the maize hybrids tested, so they are unsuitable for testing. Other environments were close to the ideal point but were outside the first circle, so it was useful for selecting hybrids in specific environments.  Table 2 for legends.
The GGE biplot can provide an overview of the differences between hybrids and environmental characteristics tested in multi-environment experiments. One advantage of the biplot that showed the distribution pattern of hybrids and environments was "which won where/what biplot" [53]. One of the characteristics of this biplot was the presence of polygons that indicate the location of the hybrid being tested. Hybrids at the top of the polygon (vertex) have the highest yields in the environment in that sector. Another important feature of this pattern was the grouping of environments, which suggested the possibility of different mega-environments [1,29,54]. In this current study, it was shown that within each growing season, the sites fall into different groups, and the pattern of site grouping varied throughout the seasons. The first two PCs explained 54.14-63.30% of the total variability due to the effects of hybrid (G), environment (location and growing sea-  Table 2 for legends. In multi-environment evaluation, discriminative and representative locations were very important. The ideal environment (location) should differentiate the hybrids being evaluated and the representation of all environments [53,56]. In this current study, the GGE biplot revealed that the environments of GJRS and KARS were the discriminative location(s) and were at the ideal point (small arrow) of testing location(s) for multi-environment evaluation in Java Island (Figure 7). Contrary to that result, the JTRS was the farthest location from the ideal point and close to the center point of the biplot axis. This location provided small information about the maize hybrids tested; therefore, it was not suitable for multi-environment evaluation on Java island. Overall, the maize hybrids trial in megaenvironments selected the five best genotypes (stable and high-yielding); were SC1, SC2, SC4, SC7, and SC9. In addition, both the dry season, the rainy season, and the combination of the two seasons produce three mega-environments. The GGE biplot has also succeeded in determining two representative environments for testing that can be used for large-scale development in the future, namely GJRS and KARS.

Stability Maize Hybrids Based on Sustainability Index (SI)
The Sustainability Index (SI) evaluation is presented in Table 9. Some researchers revealed that estimates of high SI indicate stability levels in certain genotype(s) [33,43,57]. In this current study, SI was divided into five groups: very low, low, medium, high, and very high [33,43]. Estimating SI for grain yield in maize ranged from 49.51% (moderate) to 60.77% (high). The small range of SI was due to the genetic background of the planting materials originating from the selected hybrid. The moderate SI was shown by hybrid SC1 (49.51%), SC2 (58.54%), SC4 (53.01%), SC5 (49.86%), SC6 (56.57%), SC7 (58.14%), SC8 (58.78%), and SC9 (57.43%). Only the maize hybrid of SC3 showed high SI (60.77%). The maize hybrid of SC3 showed a high average yield at 8.43 ton.ha −1 with high SI of 60.77%, indicating the highest performance and stability of this hybrid (Table 8). On the contrary to this result, maize hybrids of SC5 and SC6 showed medium SI at 49.86% and 56.57%, respectively, with a low yield at 7.86 tons per ha and 7.90 tons per ha. This result of SI for the two low-yield maize hybrids indicated the stability of grain yield. This result is similar to the previous combined analysis, as presented in Figure 2, wherein these two maize hybrids were grouped into a stable low-yield hybrid. The other maize hybrid showing high yield and SI nearly high were maize hybrids of SC2, SC7, SC8, and SC9. Generally, maize hybrids with moderate to high SI and performed yield above average can be categorized as ideal genotypes.   [33] reported a similar strategy to select high-yield maize hybrids by using SI.
Information of selected maize hybrids based on different stability analyses was summarized in Table 10. Based on this Table, three maize hybrids were selected, namely maize hybrids of SC2, SC7, and SC9. These three maize hybrids have high yields and are stable in different Java island environments, so that they can be recommended for maize development programs in Indonesia.

Conclusions
The results of the analysis showed that the main effects of the growing season, location, hybrid (G), and their interactions had a significant influence (p < 0.01) on the variation of maize hybrid yields in Java Island, Indonesia. Stability measurements NP (1) , Wi 2 , S 2 di, θ (i) , σ 2 i , S (1) , S (2) , and Di were included in the concept of static stability, while TOP, S (3) , S (6) , NP (2) , NP (3) , KR, NP (4) , CV i , and b i measurements were included in the concept of dynamic stability. SC2, SC7, and SC9 were identified as the most stable and high-yielding yields, so they can be recommended for maize development programs in Indonesia. The dry season, the rainy season, and the combination of the two seasons produce three megaenvironments. GJRS and KARS were the most representative environments with high discriminatory power, so they can be used as favorable environments for selecting the ideal maize hybrid.

Data Availability Statement:
The data used to support the findings of this study are included within the article.