Stability and Performance Assessment in 18 Short-Duration Rice Genotypes under Rain Fed Lowland Production Conditions of Ethiopia

In order to identify high yielding and stable short-duration lowland rice genotypes, field experiment was conducted with 18 rice genotypes for three consecutive years (2013-2015) at five locations in a randomized complete block design of three replications. Combined analysis of variance showed highly significant differences for the genotype and environments. The genotype by environment interaction (GEI) was also highly significant indicating differential response of genotypes to environments. The partitioning of total sum of squares exhibited that the GEI effect was a predominant source of variation (34.44%), followed by the environment and genotype effects of 24.3% and 19.04%, respectively. The GEI effect was nearly two times higher than that of the genotype effect, suggesting the presence of different environment groups. In the AMMI analysis, the first six interaction principal component axes (IPCA1 to IPCA6) were highly significant and together explained 92.18% of interaction sum of squares. AMMI stability value (ASV) discriminated genotypes G17, G16 and G8 as the stable genotypes. But, based on the yield stability index (YSI), the most stable genotypes with high grain yield were genotypes G16 and G11. AMMI1 and GGE biplots also recommended G16 and G11as stable and high yielding genotypes, whereas G2 as unstable but high yielding. Thus, genotypes G16 and G11could be released for wider adaptation while genotype G2 for specific adaptation.


Introduction
Rice is the world's second most important cereal crop next to wheat in terms of total production and after maize in terms of area coverage and productivity (FAOSTAT, 2020). The crop is grown worldwide over an area of 163.24 Mha with a total production of around 740.95 Mt and world average productivity of 4.66 t ha -1 (FAOSTAT, 2020). Globally, human consumption accounts for 85% of total production for rice, compared with 72% for wheat and 19% for maize and rice crop also provides 21% of global human per capita energy and 15% of per capita protein (Maclean et al., 2002). Rice has also become a commodity of strategic significance and the fastest growing food source in Africa. Its adoption as a principal staple food is increasing and is grown and consumed in more than 43 African countries (AfricaRice, 2017), and yet domestic production never meet local rice demand leading to huge annual import. In Ethiopia, rice is an increasingly important food, feed and cash crop and, it is also source of employment for the youth and women as well as other actors involved along the rice value chain. Despite its economic and food security importance, productivity of the crop is constrained by several factors such as terminal moisture stress, cold stress, low soil fertility, blast and sheath rot, among other things. Terminal moisture stress is predominantly a common problem in lowland rain fed rice cultivation. High yielding rice varieties with short growing duration are required to sustain rain fed lowland rice production in Ethiopia. To this end, introduction and screening of different rice genotype through multi-environment evaluation has been practiced under rice variety improvement program.
Multi-environment trials are conducted to evaluate yield stability performance of plant materials in diverse environmental conditions (Yan et al., 2000;Yan and Rajcan, 2002). Genotypes grown in different environments often show significant fluctuations of performance for yield and yield related traits. These changes are influenced by the different environmental conditions such as variations in moisture, soil nutrients, temperature and relative humidity, and this is referred to as genotype-by-environment (GE) interaction (Kang, 2002). The GE interaction reduces the association between phenotypic and genotypic values and leads to bias in the estimates of gene effects and combining ability for various characters that are sensitive to environmental fluctuations less amenable to selection (Farshadfar et al., 2000;Kang, 2002). Hence, GE interaction must be either exploited by selecting best genotype for each specific environment or avoided by selecting widely adapted and stable genotype across a wide range of environments (Eisemann et al., 1990;Kang, 2002).
Although different methods have been reported by scholars to understand pattern of GE interaction in multi-environment data, the two most often used statistical models are additive main effects and multiplicative interaction (AMMI) as reported by Gauch and Zobel (1988) and Zobel et al. (1988), and genotype plus GE interaction (GGE) based on Yan et al. (2000). The AMMI analysis combines analysis of variance for genotype and environment main effects with principal components analysis of the G x E interaction into a unified approach (Gauch, 1988;Zobel et al., 1988) while the GGE biplot as a data visualization tool is able to graphically demonstrate GE interaction patterns. GGE biplot is an effective tool to rank genotypes based on mean yield and stability and to identify mega-environments with corresponding winner genotypes as well as to evaluate test environments. Therefore, the objective of the current study was to identify high yielding shortduration lowland rice genotypes with wider or specific adaptation by applying AMMI and GGE statistical tools.

Plant materials, experimental design and trial management
In this study, including one check (Ediget), a total of 18 lowland rice genotypes were used and evaluated for grain yield and yield related traits (Table 1). The experiment was laid out using a randomized complete block design (RCBD) of three replications. Seeds of each genotype were hand drilled at the rate of 60 kgha -1 in a plot size of 6m 2 and with a spacing of 20cm between rows. Each experimental plot comprised six rows of the gross plot, with only four harvestable rows. Fertilizers (UREA and DAP) were applied as per to local recommendations. The DAP was applied all at planting while UREA was applied in three splits-at sowing, tillering, and panicle initiation. Other crop management practices were applied to the entire experimental area uniformly.

Experimental sites
This experiment was executed at Woreta, Pawe, Assosa, Mai-Tsebri, and Jimma research stations from 2013 to 2015 during the main cropping seasons under rain fed lowland conditions. As presented in Table 2

Data collection and statistical analysis
Data were collected for days to heading (DTH), days to maturity (DTM), panicle length (PL), plant height (PH), number of filled grains per panicle (FSP), fertility rate (FR), grain yield (Gy), thousand seed weight (TSW) and disease data such as leaf blast (LB) and brown spot (BS) were collected based on 0-9 scale following IRRI standard evaluation system (IRRI, 1996); where 0: immune, 1:highly resistant, 2: resistant, 3 and 4: moderately resistant, 5 and 6: moderately susceptible, 7: susceptible, and 8 and 9: highly susceptible. Grain yield harvested from each plot was converted into kgha -1 at 14% standard grain moisture content. Data were subjected to analysis of variance using the General Linear Model (PROC GLM) of the SAS Procedure version 9.0 of the SAS software (SAS, 2002) to determine significant variation among genotypes and environments and their interaction. Mean performance of different traits were separated using Least Significant Difference (LSD) method at 0.05 level of probability. Additive main effects and multiplicative interaction (AMMI) model was applied to assess the effect of genotype by environment interaction, and stability of rice genotypes  using GenStat (16 th edition) statistical package. Moreover, GGE analysis, according to Yan et al. (2000), was employed to visualize grain yield stability and performance, and identify specifically adapted genotypes among 18 rice genotypes at eleven environments. In this study, AMMI stability value (ASV) was estimated for each genotype according to the relative contributions of the principal component axis scores (IPCA1 and IPCA2) to the interaction sum of squares according to Purchase et al. (2000) as described below: ASV= Where, ASV= AMMI stability value; SS= sum of square; IPCA1 and IPCA2= the first and the second interaction principal component axes, respectively. The larger the IPCA score is, either negative or positive, the more adapted a genotype is to a certain environment. Smaller ASV scores indicate a more stable genotype across environments (Farshadfar et al., 2011). Yield stability index (YSI) was also estimated using the sum of the ranking based on yield and ranking based on the AMMI stability value i.e YSI= RASV+RY, where RASV is the rank of the genotypes based on the AMMI stability value; RY is the rank of the genotypes based on yield across environments. YSI incorporates both mean yield and stability in a single criterion (Tumuhimbise et al., 2014;Bose et al., 2014) and low values of YSI show desirable genotypes with high mean grain yield and stability.

Results and discussions 3.1 Variation in traits
The combined analysis of variance over locations and years of all traits, and the AMMI analysis of variance for grain yield are presented in Table 3 and Table 4, respectively. Means squares of genotype (G), location (L), and year (Y) showed highly significant variation for all traits considered except for filled grains per panicle in the case of year effect (Table 3). Two-way interactions of all combinations and the three-way interactions (G x L x Y) (except for fertile tillers) showed significant variation for all traits revealing the inconsistence performance of genotypes for different traits across locations and over the years. Similar results were reported by Hasan et al. (2014), Ogunbayo et al. (2014) and Bose et al. (2014) for rice genotypes performance across sites and over seasons. The result in AMMI analysis of variance for grain yield revealed that environment (E), genotype (G) and genotype by environment (GE) interaction were highly significant (P<0.001). In multi-environment trial data, the largest variation in grain yield is attributed to E, followed by GE interaction and then by G (Gauch, 2006;Yan and Kang, 2003). Table 3. Mean squares of grain yield and yield related traits in 18 lowland rice genotypes at five locations for three years  In this study, however, grain yield was largely influenced by GE interaction effect (34.4 %), followed by E (24.3%) and G (~19 %) effects which is in agreement with the findings Cantila et al. (2020) who reported that GE interaction, E, and G explained 52.3%, 26.8%, and 15.5% of the total variation. Treatment sum of square was also largely explained by GE interaction (44.3%), followed by E (~31.2%) and G (~24.5%) ( Table 4). The variation attributed to GE interaction was nearly twice that of the genotype effect.

Degree of freedom
As reported by Yan and Kang (2003), the large GE interaction effect relative to genotype implies that environments might be divided into mega-environments to which genotypes responded differently. In this study it was also observed that the first six highly significant IPCAs (IPCA1 to IPCA6) together explained 92.2% of the total GE interaction effect, with each accounting for 28, 22.5, 16.7, 9.2, 8.4 and 7.4% of GE interaction, respectively (Table 4). However, Cantila et al. (2020) reported that the first four highly significant IPCAs explained 35.8%, 26.9%, 16.9% and 13.4% of GE interaction sum of squares while Taddesse et al. (2017) reported that the first three significant IPCAs explained 35.6, 27.1 and 18.8% of the total GE interaction sum of squares, respectively.

Mean performance of rice genotypes
The mean values of growth and yield traits of16 lowland rice genotypes (days to 50% heading, days to 85% maturity, panicle length, plant height, filled grains per panicle, fertility rate, thousand seed weight, and grain yield) combined across eleven environments are presented in Table 5. In days to 50% heading and days to 85% maturity, nearly 44% of the genotypes had days to heading and days to maturity higher than the grand mean. The least days to heading was observed in three genotypes; G8 (72 days), G14 (71 days and G16 (72 days) which was slightly lower than the standard check, G18 (73 days), while only one genotype (G14) was earlier than the standard check in terms of days to maturity. In the case of panicle length, genotypes G1, G3, G4, G5, G11, G14, G15 and G16 exhibited the longest panicle length and slightly longer than the standard check and about 50% of genotypes showed the tallest plant height, measuring 90 to 114 cm which was higher than the grand mean, but G9, G7, G3, and G11 were significantly shorter than the standard check, measuring 71 to 74.6 cm. The total number of filled grains per panicle was the highest in G10, followed by G12, G17, G4, G16, G2 and G18 which was higher than the grand mean and they also had high grain fertility rate. Thousand seed weight was the highest in G16 (32.8 g) and G18 (32.8g), followed by G13 (31.4g), G15 (G31.3g), G14 (30.4g) and G17 (30.2g) with overall mean of 28.02g. Mean grain yield of genotypes also ranged from the lowest of 3439.1kgha -1 for G13 to the highest of 5812.3kgha -1 for G16 with grand mean of 4561 kgha -1 . Only three genotypes (G2, G11 and G16) significantly outperformed the standard check (G18) with mean grain yield of 5409.8, 5423.1 and 5812.3kgha -1 , respectively (Table 5). These high yielding genotypes also showed better resistance to major rice diseases (panicle blast and brown spot) compared to the other genotypes.

AMMI stability value (ASV) and yield stability index (YSI)
Ranking of 18 lowland rice genotypes based on mean grain yield, IPCA 1 score, ASV, and YSI is presented in Table 7. In terms of mean grain yield, genotype G16 ranked first followed by G11, G2, and G3 with 5812, 5423, 5410 and 4941 kg ha -1 , respectively. The IPCA1 scores also demonstrated that G16 was the most stable genotype, followed by G10, G8, and G6, whereas the other high yielding genotypes (G11, G2, and G3) were unstable as they had high IPCA1 scores. ASV stability measure as proposed by Purchase et al. (2000) also stated that genotypes with the least ASV or have the smallest distance from the origin in the biplot are considered as the most stable genotypes, whereas those which have the highest ASV are considered as unstable. Accordingly, G17 was the most stable genotype for grain yield, followed by G16, G8, G13 and G15, as they had the least ASV while G2 was the most unstable genotype, followed by G4, G3, and G5 (Table 7). Table 7. Ranking of 18 short-duration lowland rice genotypes based on, IPCA1scores, AMMI stability value (ASV), and yield stability index (YSI) mean grain yield (Gm, kgha -1 ) at eleven environments. The YSI estimate which combined mean yield and ASV rankings elucidated that G16 was the most stable genotype followed by G17, G11, and G18, because they had the least YSI. On the other hand, G1, G6 and G9 were the most unstable genotypes as they showed the highest YSI. Genotype G16 was the best genotype as it ranked first in mean yield, IPCA1 score YSI and, second in ASV. However, genotypes G11 and G2 which ranked second and third in mean yield showed inconsistency in stability ranking and, thus both were unstable. Inconsistency in the ranking of genotypes based on different approaches demonstrated the importance of considering both mean yield and stability performance to guide selection of genotypes in the breeding program (Farshadfar et al., 2011).

AMMI and GGE biplots
AMMI biplots, AMMI1 (IPCA1 vs mean yield) and AMMI2 (IPCA1 vs IPCA2) were applied to further illustrate the effect of each genotype, environment and the interaction in the multi-environment data as presented in Figure 1 and Figure 2, respectively. In AMMI1 biplot, genotypes or environments laid on the same vertical line had similar mean yields and those laid on the same horizontal lines had similar interaction patterns (Crossa et al., 1990). In addition, a genotype or an environment plotted on the right side of the central vertical axis had higher yield than those of left hand side and, if a genotype or an environment has IPCA1 score of nearly zero, it has less interaction effect . Accordingly, genotypes G2, G3, G4, G5, G10, G11, G12, G16, G17, and G18 exhibited above average in mean yield. Of the three best yielding genotypes (G16, G2, and G11), G16 had the lowest IPCA1 score suggesting its wider adaptation and can be cultivated across tested environments while G2 and G11 had relatively large IPCA1 scores and thus unstable; that is, they had specific adaptations. In contrast, genotypes G1, G6, G8, G9, and G13 performed bellow average in mean yield with lower IPCA1 scores except for G1 which had larger IPCA1 score and then highly interactive (Figure 1).
With regard to environments, the highest yielding environment was E4, followed by E11 and E2, all with large IPCA1 scores indicating their strong contribution to the interaction effect. In contrast, E1, E3, E6 and E7 were low yielding with large IPCA1 scores except for E6 and E7 that had relatively smaller IPCA1 scores. On the other hand, E5, E8 and E9 were average yielding environments and closer to the biplot origin suggesting their smaller contribution to the interaction (Figure 1).  Table 1.
In AMMI2 biplot, environments positioned far from the biplot origin had large contribution to the GE interaction and if they are closer to the origin, they contributed for the stability of genotypes. Similarly, genotypes close to the biplot origin are stable while those distant are unstable and genotypes and environments positioned close to each other in the biplot have positive associations (Silivera et al., 2012). Accordingly, significant GE interaction was attributed to E1, E2 and E3 as they were away from the biplot origin. In contrast, E6, E8 and E11 contributed the lowest to the GE interaction and thus most stable while the other environments (E4, E5, E7, E9 and E10) were intermediate (Figure 2). Genotypes G15, G16, and G18 were close to the biplot origin which suggested that they were relatively stable, G16 being the most stable and this was also in accordance with Figure 1. Located far away the biplot origin, genotypes G1, G2, G3, G5, and G12 were the most unstable and associated to different environments (Figure 2). GGE biplot analysis is also a data visualizing tool used for, among other things, evaluating cultivars based on average yield and stability performance, identifying best cultivar in each environment and grouping environments based on cultivars performance (Yan, 2001;Yan et al., 2007). In this study, the GGE biplots in Figures 3 and 4 each explained 59.11% of the total variation in grain yield of 18 rice genotypes, with the first and second principal component (PC1 and PC2) contributing 41.78% and 17.33% of the variations, respectively. Figure 3 demonstrates the ranking of 18 rice genotypes based on both mean grain yield and stability performance. Genotypes in the direction of the arrow or on the positive side of the vertical solid line are high yielding while those on negative side are with low mean yield (Yan, 2001). Moreover, genotypes with short vectors, regardless of their directions, are more stable whereas with longer vectors are unstable (Yan and Tinker, 2006). In the present study, the best performing genotypes in terms of mean yield were G16, G11, G2 and G4 while poor performing genotypes were G13, G6, G9, G8, G7 and G1 as illustrated in Figure 3. With regard to stability of genotypes as dictated by the length of genotype vectors in either direction, G16, G13, G9 and G8 could be considered as the most stable genotypes. However, the latter three genotypes were poor in terms of mean yield performance. Genotypes G11 and G2 were also the highest in mean yield, G11 being relatively stable while G2 was unstable. Thus, genotype G16 followed by G11 was the highest yielding genotypes and consequently, G16 is the most ideal genotype for rain fed lowland rice cultivation in all environments due to high mean yield and high yield stability while G11 can be recommended for specific environments.  Table 1.
Another most attractive feature of a GGE biplot is its ability to show the 'which-won-where' pattern in a genotype-by-environment dataset, as it graphically demonstrates relationships of genotypes to different environments (Yan and Tinker, 2006). Genotypes located on the vertices of the polygon performed either the best or the poorest in one or more environments. Accordingly, genotypes G16, G11, G4, and G10 were better in the environments E9, E8, E7,and E5, where as the genotypes G2, G12, G14, and G18 were better in the environments E1, E3, E4, and E11 (Figure 4).  Table 1.
Genotypes G3, and G5 also performed better in environments E2 and E10 while G1, G6, G7, G8, G9, G13 and G15 did not perform well in any of the environments. This biplot suggested the presence of three megaenvironments where genotypes G2, G3 and G16 as winner genotypes at each group of environments, whereas genotypes G1, G6, G13 though identified as vertex genotypes but they were not associated to any environments indicating as they were not best at least in one environment (Figure 4).  Table 1.

Conclusion and recommendation
The study of genotype x environment (GE) interaction is critical for appropriate genotype evaluation in multienvironment trials. Genotypes that showed both high mean yield performance and stability across a wide range of environmental conditions are desirable for rice production. However, the presence of GE interaction makes difficult which genotypes to select. In the current study, results indicated that the yield performance of rice genotypes was significantly influenced by GE interaction effects which contributed nearly two times higher than genotype effects. The GGE biplots and AMMI were also used to compare the performance of different genotypes across test environments. Genotypes G16, G11 and G2 were identified as the top three high yielding genotypes. Genotype, G16 was the highest yielding and most stable genotype, followed by G11 as shown by AMMI stability value, yield stability index, and GGE ranking biplot, while the remaining tested genotypes showed inconsistent performances across environments. Thus, considering the grain yield performance and stability of genotypes, G16 and G11 could be recommended for wider adaptation while G2 for specific adaptation.