HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS		QUICK SEARCH:   [advanced] Author: Keyword(s): Year:  Vol:  Page:

This Article Abstract Figures Only Full Text (PDF) Free Alert me when this article is cited Alert me if a correction is posted Services Similar articles in this journal Alert me to new issues of the journal Download to citation manager Citing Articles Citing Articles via Google Scholar Google Scholar Articles by Ceretta, S. Articles by van Eeuwijk, F. Search for Related Content PubMed Articles by Ceretta, S. Articles by van Eeuwijk, F. Agricola Articles by Ceretta, S. Articles by van Eeuwijk, F. Related Collections Biometrics Other Crops Data Management

Grain Yield Variation in Malting Barley Cultivars in Uruguay and Its Consequences for the Design of a Trials Network

^a Programa Cultivos de Secano, Instituto Nacional de Investigación Agropecuaria, CC 39173, CP 70000, Colonia, Uruguay
^b Lab. of Plant Breeding, Dep. of Plant Sciences, Wageningen Univ., P.O. Box 386, 6700 AJ Wageningen, The Netherlands

^* Corresponding author (ceretta{at}inia.org.uy).

	ABSTRACT

TOP NOTES ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS DISCUSSION REFERENCES

 
The efficiency of cultivar trial networks is an important subject in official cultivar testing. We investigated this efficiency for malting barley (Hordeum vulgare L.) in Uruguay, using data on 213 cultivars tested across an eight-year period at six locations. The variance-components approach was used to quantify the effects of years, locations, sowing dates and replicates on the precision of cultivar mean comparisons. The relationships among testing environments and genotypic adaptation patterns were explored via biplots. Factorial regression was used to model genotype x environment interaction (GEI) directly in relation to measured environmental variables. Variance components indicated that both the number of locations and sowing dates could be reduced. Biplot analysis identified some repeatable GEI patterns. Factorial regression showed that mean daily temperature during the emergence-heading period and daily minimum temperature at heading explained 20% of GEI. Still, the majority of the GEI appeared to be highly nonrepeatable. A future network should focus on wide adaptation while enhancing the chances to exploit specific adaptation to the prevalent temperature conditions by sampling contrasting sowing dates at different locations.

Abbreviations: AMMI, additive main effect and multiplicative interaction effect • CPD, critical percentage difference • CS, Colonia Suiza • GEI, genotype x environment interaction • GGE, genotypic main effect plus GE • GL, genotype x location interaction • GLY • genotype x location x year interaction • GY, genotype x year interaction • LE, La Estanzuela • MET, multiple environment trial • OL • Ombúes de Lavalle • PCA, principal components analysis • PRE_EH, the precipitation accumulated during the emergence-heading period • PRE_HH, precipitation accumulated during the heading-harvesting period • PY, Paysandú • RAD_HH, the solar radiation accumulated during the heading-harvesting period • RAD_{AVG_}EH, the daily solar radiation averaged across the emergence-heading period • RAD_{AVG_}HH, the daily solar radiation averaged across the heading-harvesting period • TMAX_H, the daily maximum temperature averaged across an interval of 7 d before and after heading • TMED_EH, daily medium temperature accumulated during the emergence-heading period • TMED_{AVG_}EH, daily medium temperature averaged across the emergence-heading period • TMIN_H, the daily minimum temperature averaged across an interval of 7 d before and after heading • TPE, target population of environments • TR, Tarariras • YG, Young

	NOTES

TOP NOTES ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS DISCUSSION REFERENCES

Received for publication June 27, 2006.

Grain Yield Variation in Malting Barley Cultivars in Uruguay and Its Consequences for the Design of a Trials Network

^a Programa Cultivos de Secano, Instituto Nacional de Investigación Agropecuaria, CC 39173, CP 70000, Colonia, Uruguay
^b Lab. of Plant Breeding, Dep. of Plant Sciences, Wageningen Univ., P.O. Box 386, 6700 AJ Wageningen, The Netherlands

^* Corresponding author (ceretta{at}inia.org.uy).

The efficiency of cultivar trial networks is an important subject in official cultivar testing. We investigated this efficiency for malting barley (Hordeum vulgare L.) in Uruguay, using data on 213 cultivars tested across an eight-year period at six locations. The variance-components approach was used to quantify the effects of years, locations, sowing dates and replicates on the precision of cultivar mean comparisons. The relationships among testing environments and genotypic adaptation patterns were explored via biplots. Factorial regression was used to model genotype x environment interaction (GEI) directly in relation to measured environmental variables. Variance components indicated that both the number of locations and sowing dates could be reduced. Biplot analysis identified some repeatable GEI patterns. Factorial regression showed that mean daily temperature during the emergence-heading period and daily minimum temperature at heading explained 20% of GEI. Still, the majority of the GEI appeared to be highly nonrepeatable. A future network should focus on wide adaptation while enhancing the chances to exploit specific adaptation to the prevalent temperature conditions by sampling contrasting sowing dates at different locations.

Abbreviations: AMMI, additive main effect and multiplicative interaction effect • CPD, critical percentage difference • CS, Colonia Suiza • GEI, genotype x environment interaction • GGE, genotypic main effect plus GE • GL, genotype x location interaction • GLY • genotype x location x year interaction • GY, genotype x year interaction • LE, La Estanzuela • MET, multiple environment trial • OL • Ombúes de Lavalle • PCA, principal components analysis • PRE_EH, the precipitation accumulated during the emergence-heading period • PRE_HH, precipitation accumulated during the heading-harvesting period • PY, Paysandú • RAD_HH, the solar radiation accumulated during the heading-harvesting period • RAD_{AVG_}EH, the daily solar radiation averaged across the emergence-heading period • RAD_{AVG_}HH, the daily solar radiation averaged across the heading-harvesting period • TMAX_H, the daily maximum temperature averaged across an interval of 7 d before and after heading • TMED_EH, daily medium temperature accumulated during the emergence-heading period • TMED_{AVG_}EH, daily medium temperature averaged across the emergence-heading period • TMIN_H, the daily minimum temperature averaged across an interval of 7 d before and after heading • TPE, target population of environments • TR, Tarariras • YG, Young

	INTRODUCTION

TOP NOTES ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS DISCUSSION REFERENCES

 
MALTING BARLEY (Hordeum vulgare L.) is an important crop in Uruguay, where it is cultivated under rainfed conditions in the southwestern part of the country from approximately –34° to –32° latitude. Around 90% of the barley grain produced in Uruguay is locally processed and then exported as barley malt. This helps the malting industry promote the use of cultivars with good malting quality. However, good malting quality is often linked with poor agronomic adaptation, resulting in a reduced benefit to growers, thereby imposing a severe constraint on the development of the malting barley crop in Uruguay. Since 1981, the release of new cultivars has required statutory testing for production and use in multiple-environment trials (METs). The main goal of the statutory METs is to assess and predict the agronomic value and malting quality of candidate genotypes when they are grown in future years in the production area. This requires the test environments to be representative of the target population of environments (TPE); the distribution of environments characterizing a crop-growing area (Comstock, 1977). The environmental conditions included in METs are intended to form random samples of the environmental conditions characterizing the TPE. To this purpose, trials are conducted in locations representative of the range of latitude and soil types of the crop-growing area, and across years to sample the seasonal variation in weather conditions. Although a considerable number of locations and years may be necessary for adequate environmental sampling, resource limitations and the need to make improved cultivars rapidly available to farmers generate pressure to keep numbers of locations and years relatively small. Consequently, it is important to continuously try to optimize the network.

The efficiency of METs can be approached from a global and a local perspective. The global perspective is directed at precise predictions of cultivar performance across the TPE, whereas the local perspective implies precise predictions for defined regions or locations that require specifically adapted cultivars. We will use both perspectives in our evaluation of the Uruguayan malting barley network.

Patterson et al. (1977) proposed an approach within the global perspective that uses variance components for GEI to calculate the precision of trial networks as a function of the number of years, locations, and replicates per trial. Other researchers have since applied this approach (Talbot, 1984; Robinson, 1984). The approach is based on the assumption of complete exchangeability of individual trials. This assumption implies, for example, that additional locations and replicates can compensate for fewer years. Of course, the assumption of exchangeability of trials is often somewhat unrealistic. In practice, not only the number of trials but also the specific choice of locations and sowing seasons matters (Lin and Morrison, 1992).

Recently, Atlin et al. (2000) and Piepho and Möhring (2005) have discussed the evaluation of network efficiency for local predictions using a variance-components approach. The approach emphasizes the size of the repeatable component of GEI variance, namely, the genotype x region interaction and genotype x location (within region) interaction (GL) variance, in relation to the difficult-to-predict types of GEI, like genotype x year interaction (GY) and genotype x location x year interaction (GLY). Unfortunately, complex interactions, such as GLY, have often been reported to contribute most to the GEI variance (Campbell and Lafever, 1977; Patterson et al., 1977; Talbot, 1984; Annicchiarico, 1997; Chapman et al., 2000).

By adequate characterization of the test environments for stress types and patterns, repeatable GEI can be modeled (Chapman et al., 2000) and local predictions can be made. When the occurrence of relevant environmental stresses can be predicted, it would be possible to select for specific adaptation. Additionally, when relevant environmental stresses can be identified, the implementation of trials under managed environments, could be used in METs as an alternative to random sampling of locations and years (Cooper et al., 1995). This should eventually allow for a reduction in the size of the trials network. Our task with respect to the evaluation of the trials network for local prediction is thus first to assess the amount of repeatable GEI and to find the environmental factors that create this type of GEI. Next, the set of environments should be examined for the extent of overlap for critical environmental factors. For the identification of environmental variables, we will use two classes of statistical models: bilinear models for the exploration of the relationships between environments (Crossa and Cornelius, 2002) and factorial regression models for testing whether particular environmental variables could be responsible for GEI (van Eeuwijk, 1995; van Eeuwijk et al., 1996).

In this paper, to evaluate the efficiency of the trial network for local prediction, we use an approach that also focuses on repeatable GEI, principally GL; instead of a variance-components approach, we aim at identifying environmental factors that are possible causes of GEI. We wanted to avoid the assumption of exchangeability of test environments that underlies the variance-components approach.

The major objective of this paper is to evaluate the efficiency of the official Uruguayan malting-barley trials system, both from a global and a local perspective. A secondary objective is to illustrate the integrated use of different statistical techniques to evaluate trial systems in general. Possible sources of environmental variation related to the presence of large and complex interactions are discussed. Finally, recommendations are given for a more efficient design of the official Uruguayan malting-barley trials network.

	MATERIALS AND METHODS

TOP NOTES ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS DISCUSSION REFERENCES

 
Data
To evaluate the efficiency of the official Uruguayan cultivar trials for malting barley for grain yield, data were available for 213 cultivars that were tested across an eight-year period (1991–1998). (Note, we will use the terms cultivar and genotype interchangeably). Each year, six testing locations were used: La Estanzuela (LE), Tarariras (TR), Colonia Suiza (CS) and Ombúes de Lavalle (OL) situated in the southern region of Uruguay, Young (YG) situated in the north-central region, and Paysandú (PY) situated in the northern region (Fig. 1 ). The maximum number of trials grown each year was nine. For LE and YG locations, three and two sowing dates, respectively, were used and trials were coded as early (er), medium (md), and late (lt) for the sowing dates of June, July, and August/September, respectively. Each year, new cultivars entered the trials network, while others were dropped. As a result, the data set was unbalanced across years, but balanced within years. Five cultivars (Ana, Bonita, Clipper, Defra, and E. Quebracho) were present in all trials. The total number of cultivars tested each year and the number of common cultivars among years are presented in Table 1 . Only cultivars tested for at least two years were included in this study.

View larger version (14K): [in this window] [in a new window]   Figure 1. Map of Uruguay indicating the position of the test locations: Colonia Suiza (CS), La Estanzuela (LE), Ombúes de Lavalle (OL), Paysandú (PY), Tarariras (TR) and Young (YG).

Assessing the Precision of the Trials Network using Variance-Components Analysis
Variance components for grain yield were estimated by fitting the following random model to the cultivar means:

[1]

where µ is the general mean, c_i is the cultivar main effect, e_j is the environmental main effect (combination of location, year, and sowing date), (cy)_ik is the cultivar x year interaction effect, (cl)_il is the cultivar x location interaction effect, (cyl)_ikl is the cultivar x year x location interaction effect. Because we used cultivar means, the residual term, f_ijkl, contains a contribution of the effect of the interaction between cultivars and sowing dates within trials (location x year combinations) and plot error. All random terms were assumed normally distributed, with zero mean and variances _c², _e², _cy², _cl², _cyl², and _f², referring to the random terms defined above. All variance components were estimated by restricted maximum likelihood (REML), as implemented in Genstat 5 (Genstat 5 Committee, 1993). An estimate of the plot error, ², was obtained from the median of the error mean squares reported for individual trials. The plot errors complied with a Chi-square distribution, so we could rely on the median of the individual plot errors as being an appropriate estimate to work with subsequently. An estimate for cultivar x sowing date within year x location interaction, _cyls², then follows from _cyls² = _f² – ²/n_r, where n_r = 3, the number of replicates per trial.

The critical percentage difference (CPD) has been proposed as a measure of the precision of a trial system (Patterson et al., 1977; Talbot, 1984). The CPD is the minimum difference between a candidate cultivar and a check cultivar (expressed as a percentage of the grand mean) that can be detected according to the particular design of the trials network. Consequently, the lower the CPD value, the higher the precision of a trials network. We calculated the CPD for different hypothetical networks resulting from different combinations of numbers of replicates, years, locations, and sowing dates. The CPD was calculated as:

[2]

where Z₍₎ is the value that a standard normal variate will exceed with probability , µ is the grand mean, and V is one-half of the variance of difference between cultivar means and computed as follows:

[3]

where n_y, n_l, n_s, and n_r denote the number of years, locations, sowing dates, and replicates, respectively.

Correlations between Environments
Correlations between Trials within Years on the Basis of Yield
Correlations between trials within years were studied via biplots (Gabriel, 1971; Kempton, 1984; van Eeuwijk, 1995; Yan and Kang, 2003; Graffelman and van Eeuwijk, 2005) following a principal components analysis of the cultivar by trial table of means, where the data were standardized per trial (column). This kind of analysis has been referred to as GGE biplot analysis (Yan et al., 2000). Effectively, the following multiplicative model was fitted to the previously standardized data, and we programmed the analysis in Genstat 5 (Genstat 5 Committee, 1993):

[4]

where Y_ij is the adjusted (and standardized) mean yield for the ith cultivar in the jth environment, µ is the grand mean, β_j is the environmental main effect, _m is a proportionality constant (singular value), _mi and _mj are genotypic and environmental scores, and _ij is a residual term. For our purposes, a graphical inspection of the most salient features of the correlations between trials, M was assigned a value of 2. (It should be emphasized that we see biplots merely as a quick way of inspecting the main features of data tables and matrices, and we are not interested in building multiplicative models for prediction. This is in contrast to Gauch [2006]). Genotypic and environmental scores can be interpreted as coordinates for plotting genotypes and environments together in a planar display, the biplot, with the coordinates for the genotypes being (₁^c_1i,₂^c_2i), whereas those for the environments are given by (₁^1–c_1j,₂^1–c_2j), with c being between 0 and 1. For investigating relations between environments, we used c = 0. The point representations of genotypes and environments in a biplot are referred to as genotype and environment markers, respectively.

A biplot helps visualize similarities between environments as well as identify genotypes and environments mostly responsible for the presence of interaction. For details on interpretation of biplots see van Eeuwijk (1995), van Eeuwijk et al. (1995), Yan et al. (2000), Yan and Kang (2003), Graffelman and van Eeuwijk (2005) and Yan and Tinker (2006).

Correlations between Trials across Years on the Basis of Yield
For this analysis, we focused on the two most contrasting locations regarding latitude, PY (north) and LE (south). Furthermore, as daily meteorological data for PY were only available for the period 1991 to 1995, we restricted the analysis to that period. A balanced cultivar by environment two-way table of means was constructed by filling the missing cells with the corresponding best linear unbiased estimators obtained from the fit of an additive model including the main effects for cultivar and trial/environment (location by year by sowing date combinations). Correlations between trials were studied in biplots as described above.

Correlations between Trials across Years based on Meteorological Information
In addition to yield, daily meteorological data were available for the trials at LE and PY during the five-year period 1991 to 1995 for temperature (°C), solar radiation (MJ m^–2), and precipitation (mm). From the meteorological variables, nine environmental indices were constructed as follows: PRE_EH, the precipitation, accumulated during the emergence-heading period; PRE_HH, the precipitation, accumulated during the heading-harvesting period; RAD_HH, the solar radiation, accumulated during the heading-harvesting period; RAD_{AVG_}EH, the daily solar radiation averaged across the emergence-heading period; RAD_{AVG_}HH, the daily solar radiation averaged across the heading-harvesting period; TMAX_H, the daily maximum temperature averaged across an interval of 7 d before and after heading; TMED_EH, daily medium temperature, accumulated during the emergence-heading period; TMED_{AVG_}EH, daily medium temperature averaged across the emergence-heading period; TMIN_H, the daily minimum temperature averaged across an interval of 7 d before and after heading. Principal component analysis (PCA) was performed on the set of nine standardized meteorological indices. The graphical representation (biplot) of the first two PCA axes was used to assess the main differences between testing environments and to roughly identify the meteorological features driving these differences.

Interpreting Genotype x Environment Interaction
To study patterns of GEI, the additive main effects and multiplicative interaction effects (AMMI) model (Gollob, 1968; Mandel, 1969; Gabriel, 1978; Gauch, 1988) was fitted to the already described balanced two-way table of cultivar x trial means for the trials at PY and LE in the period 1991 to 1995. The fitted model was

[5]

where Y_ij is the expression of the ith cultivar in the jth environment, µ is the grand mean, _i is the genotypic main effect, β_j is the environmental main effect, _m is a proportionality constant (singular value), _mi and _mj are the genotypic and environmental scores, and _ij is a residual. The biplot representation of the first two AMMI axes was used to identify highly interactive test environments and cultivars.

Factorial regression (Denis, 1988; 1991; van Eeuwijk et al., 1996; Voltas et al., 1999a,b) was used to describe the GEI in terms of differential sensitivity to measured environmental variables. After fitting the main effects of genotype and environment, concomitant variables on the levels of the environmental factor were introduced in the model. The general expression for the factorial regression model is

[6]

where Y_ij is the yield expression of the ith cultivar in the jth environment, µ, _i β_j, and _ij are as described above, _hi is the genotypic sensitivities, and z_hj is the measured concomitant variables. In addition, we defined four environmental indices derived from measured phenotypic traits that were averaged across cultivars for individual trials: grain yield (t ha^–1); season length expressed in number of days from emergence to heading; severity of leaf rust (Pucinia hordei), and severity of leaf blotch, a complex of leaf diseases that in Uruguay includes net blotch (Pyrenophora teres), spot blotch (Cochliobolus sativus), and scald (Rynchosporium secalis).

	RESULTS

TOP NOTES ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS DISCUSSION REFERENCES

 
Assessing the Precision of the Trials Network Using Variance Components Analysis
Estimates for variance components are presented in Table 2 . The largest proportion of the phenotypic variation (67%) was accounted for by the main effect of the environment (location x year by sowing date combinations). Although this variation causes large fluctuations in mean yields across trials, it does not affect the relative performance of cultivars and is, therefore, of little interest to plant breeders and for cultivar recommendation purposes. The agronomic value of cultivars depends on their mean performance across environments, whose variation is reflected in the variance component for the cultivar main effect, and the interaction of the cultivars with the environments, that is, variance components related to GEI. In our study, the cultivar x environment interaction effects, as expressed by the sum of the interaction variance components was twofold (14%) that of the variance component for the cultivar main effect (7%, Table 2). Only a minor part of the interaction was due to the two-way interactions of cultivar x location and cultivar x year (5%, Table 2), and the repeatable cultivar x location interaction amounted to only 2% of the total variation. The largest part of the interaction was due to higher order interactions, especially cultivar x trial interaction, that is, cultivar times sowing date within year x location combinations. The error variance was the largest source of variation after exclusion of the environmental main effect (12%, Table 2). In summary, only a small part of the total variation was due to mean differences between cultivars (wide adaptation) and predictable interaction (cultivar by location), leaving a large part of the cultivar-related variation as unpredictable. This unpredictable variation can only be managed within a trials network by choosing enough replications for the implied factors, that is, locations, years, sowing dates, and replicates (blocks within trials). We investigated the influence of changing the replication of the latter factors by considering the CPD of the trials network. When CPD increases, precision decreases. Figure 2 illustrates that increasing the number of replicates beyond two hardly affected CPD. Nevertheless, at advanced testing stages, breeders may prefer to use three or more replicates, to minimize the risk of having little or no information available on individual genotypes when plots fail to produce yield for whichever unforeseen reason. Therefore, for comparing the effect of varying number of years, locations, and sowing dates on CPD, we assumed a trials network with three replicates per trial. A second sowing date increased precision by 1.5%, whereas the effect of adding a third sowing date was negligible (<0.5%), (Fig. 3 ). This indicates that the number of sowing dates could be reduced from three (actual number of sowing dates) to two without sacrificing precision. The number of years in which the cultivars were tested had by far the largest effect on CPD. Assuming a trials network with two sowing dates, three replicates, and three locations, adding a second year of testing would result in around 5% decrease in CPD, whereas testing for a third year would decrease CPD by an additional 2% (Fig. 4 ). Beyond three years of testing, it would still be possible to improve precision, but the magnitude of this gain might not justify the extra resources invested. The effect of increasing the number of locations was a bit smaller than that of increasing the number of years (not shown). Little increase in precision was observed beyond four locations, regardless of the number of years (Fig. 4). Based on this study, it can be concluded that both the number of locations and sowing dates could actually be reduced. A trials network consisting of three locations and two sowing dates tested across three years could detect differences between cultivar means of about 11% (380 kg.ha^–1 in this study).

View larger version (10K): [in this window] [in a new window]   Figure 2. Effect of increasing the number of replicates on the critical percentage difference, CPD (%), for different number of years, considering a trials network with six locations and two sowing dates.

View larger version (10K): [in this window] [in a new window]   Figure 3. Effect of increasing the number of sowing dates on the critical percentage difference, CPD (%), for different number of locations considering a trials network with 3 yr of evaluation and three replicates per trial.

View larger version (13K): [in this window] [in a new window]   Figure 4. Effect of increasing the number of years on the critical percentage difference, CPD (%), for different number of locations considering a trials network with three replicates per trial and two sowing dates.

View larger version (29K): [in this window] [in a new window]   Figure 5. Genotypic main effect plus genotype x environment (GGE) biplots of grain yield for individual years: (a) 1993, (b) 1996, (c) 1997, (d) 1998. The vectors represent the trials grown at Colonia Suiza (CS), La Estanzuela (LEer, LEmd, and LElt = early, medium, and late sowing date, respectively), Ombúes de Lavalle (OL), Paysandú (PY), Tarariras (TR) and YGer and YGmd (early and medium sowing date, respectively). Cultivars are represented by circles. Standard cultivars appear as filled circles. The amount of variation accounted for by the PCA axes is given as a percentage of the total sum of squares for the genotypic main effect and genotype x environment interaction (GEI).

View larger version (17K): [in this window] [in a new window]   Figure 6. Grain yield analysis over years. Genotypic main effect plus genotype x environment (GGE) biplot representation for the two most extreme locations regarding latitude: La Estanzuela (LEer, LEmd, LElt = early, medium, and late sowing date) and Paysandú (PY) during 1991 to 1995. Trials are represented by squares. The last two digits of the trial labels correspond to the year of evaluation. Cultivars are represented by circles. Cultivar names are given for five standard cultivars checks (filled, dark gray) and four widely grown or released cultivars in the period 1991 to 1995 (filled, light gray). Biplot axes are shown for the two most contrasting trials, LE1_95 and PY_95. The amount of variation accounted for by the PCA axes is given as a percentage of the total sum of squares for the genotypic main effect and GE interaction (GEI)

Cultivars CLE 169, long-season with high photoperiod sensitivity, and Dayman, medium-short season with low photoperiod sensitivity, performed above average in all of the test environments and were thus broadly adapted. The graph suggests that CLE 169 performed relatively better in the early sowing date trials at LE, whereas Dayman performed slightly better in late sowing date trials at LE, as well as at PY. Cultivars Bonita and Stirling consistently showed poor yielding ability in all of the test environments. Both are old cultivars that had been in use long before the trials network started.

Correlations between Trials across Years on the Basis of Meteorological Information
The result of the PCA on the environmental indices derived from measured meteorological variables is presented in a biplot for the first two principal components (Fig. 7 ). Early trials at LE were characterized by medium to low mean temperature and had the lowest values for mean solar radiation. In contrast, late trials at the same location were of medium to high mean temperature and high mean solar radiation. In the same figure, it can be seen that one of the late trials at LE, LE_{lt_}95, had considerably lower mean temperature than the rest of the late trials. In this sense, LE_{lt_}95 was comparable to earlier trials at the same location. Particularly, the minimum temperature at heading was very low for LE_{lt_}95 (7.5°C). This temperature was equal to that registered during the same year for the early trial (LE_{er_}95) and much lower than that for the medium trial, LE_{md_}95 (9.6°C). The similarities in temperature between LE_{lt_}95 and the early trials at the same location could provide an explanation for the similarities in cultivar responses as observed in the GGE biplot analysis across years (Fig. 6). Trials at PY were in general very similar to late trials at LE regarding mean medium temperature and had intermediate to low values for mean solar radiation (Fig. 7). These environments showed the highest maximum and minimum temperature at heading time. Medium sowing date trials at LE were intermediate in mean temperature between early and late trials at the same location, and they showed a large variation in mean solar radiation. Environment LE_{md_}93 scored very high in mean solar radiation. Based on the length of the vector, the information on precipitation appeared not very relevant for the description of the test environments.

View larger version (8K): [in this window] [in a new window]   Figure 7. Biplot representation of environments following principal components performed on environmental indices. The analysis was performed for the two most extreme locations regarding latitude: La Estanzuela (LEer, LEmd, LElt = early, medium, and late sowing date respectively) and Paysandú (PY), during the 1991 to 1995 period. Trials are represented by squares. The last two digits of the trial labels correspond to the year of evaluation. The amount of variation accounted for by the PCA axes is given as a percentage of the total sum of squares.

Interpreting Genotype x Environment Interaction
Additive Main Effect and Multiplicative Interaction Model
The first two axes of the AMMI model explained just 30% of the GEI (Fig. 8 ), pointing to the complexity of the pattern of interactions. This complexity was to be expected from the large magnitude of the cultivar by sowing date within trial (year x location combination) variance component. Judging from the distance between biplot origin and environmental markers and relative positions on opposite sides of the origin, the more interactive testing environments were LE_{er_}95 on the one side, and PY_95 and LE_{lt_}94, on the other. Another important contrast was evident for LE_{md_}95 and LE_{md_}93. Looking at the cultivars, the contrast between LE_{er_}95 and PY_95 was mostly due to the differential behavior of Bowman and Defra, which could be attributed, among other reasons, to the large differences in season length and/or disease resistance to the prevalent diseases. The mean severity of leaf rust was low both for LE_{er_}95 (3%) and PY_95 (2%), making it unlikely that the differential yield performance was due to differences in leaf rust resistance between the two cultivars. For the set of years studied here, the average severity of leaf blotch was always equal or higher for trials grown in early environments at LE than for trials at PY. The opposite happened during 1995, when the severity of leaf blotch was higher for PY_95 (34%) than for LE_{er_}95 (22%). Both the earlier sowing date and the lower severity of leaf blotch resulted in an environment where Defra, a European long-season cultivar with low resistance to leaf blotch, could have performed relatively better. The results of the AMMI analysis strongly suggested that the GEI present in our data set was mainly due to cultivar x trial interaction, arising from differential genotypic responses to environmental factors that were specific for each trial.

View larger version (13K): [in this window] [in a new window]   Figure 8. Biplot representation following additive main effect and multiplicative interaction effect (AMMI) analysis of grain yield over years. The analysis was performed for the two most extreme locations regarding latitude: La Estanzuela (LEer, LEmd, LElt = early, medium, and late sowing date) and Paysandú (PY). Trials are represented by squares. The last two digits of the trial labels correspond to the year of evaluation. Cultivars are represented by circles. Five check cultivars are depicted by dark gray circles, while four widely grown or released cultivars during the evaluation period are depicted by light gray circles. The amount of variation accounted for by the AMMI axes is given as a percentage of the total sum of squares for genotype x environment interaction (GEI).

View this table: [in this window] [in a new window]   Table 3. Analysis of variance table for the results of factorial regressions to explain cultivar x environment interaction.

	DISCUSSION

TOP NOTES ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS DISCUSSION REFERENCES

In addition to determining an appropriate number of years, locations and sowing dates, it is important to prevent repetition of testing environments with similar environmental responses, to avoid the risk of redundancy and waste of resources (Brandle and Arthur, 1992). In this sense, the GGE biplot analysis by year allowed us to identify redundant test environments, as it was the case for the two sowing dates used at YG (Fig. 5).

The GGE biplot analysis by year evidenced the lack of consistency in the correlations among the trials at OL, TR, CS, and LE (Fig. 5a vs. 5d). This indicated rapid changes in the values of the relevant environmental variables for geographically close environments. These environmental changes can be partially attributed to differences in the date of sowing within years, which could have affected the timing of biotic and abiotic environmental stresses in relation to the developmental stage of the cultivars and thus contributing to the presence of differential responses. Most likely, the environmental differences may be related to the particular characteristics of the Uruguayan agriculture. In Uruguay, agriculture requires crop rotation. Therefore, even at the same location from one year to the next, a different experimental field is used. As the experimental fields within the same location represent different cropping histories, GLY would increase at the expense of GL. Cooper et al. (2001) studied the importance of management practices, that is, cropping history, planting date, N fertilizer rate and irrigation, on the expression of GEI in grain yield for wheat in the northern grain region of Australia. They concluded that for their wheat system, it is likely that part of previously documented genotype x site x year interactions can be explained by differences in the sampled management conditions. Furthermore, under Uruguayan conditions, and especially at LE, soil physical and chemical properties can change rapidly in relation to topographic position. When this kind of spatial variation interacts with genotypes, it gives rise to large and complex interactions for grain yield (Cooper et al., 1999). We concluded that for the Uruguayan barley trials, variation in management regime and spatial variability together with fluctuations in the values of meteorological variables (precipitation, temperature) sampled across years, locations, and sowing dates contributed to the presence of a large amount of unrepeatable interactions (cultivar x year, cultivar x year x location, cultivar x sowing date within location x year).

The same variation in environmental conditions that was found at the trial level is also likely to be found at the farmers' level. While it indicates potential opportunities for identifying cultivars specifically adapted to particular management regimes, exploitation of specific adaptation will not be possible without a better insight into the nature and repeatability of the environmental stresses typical for particular management practices. To achieve such insights, additional investments may be required. Trials under managed environmental conditions (e.g., disease protections, N levels) should be implemented in addition to the actual sampling of years, locations, and sowing dates.

The GGE analysis across years emphasized the differential behavior of early trials at LE compared with the rest of the testing environments. Under Uruguayan conditions, the temperature increases from early to late sowing dates as well as from south (LE) to north (PY). Therefore, temperature differences could, at least partially, explain the separation of the early (colder) environments at LE from the rest of the test environments via the GGE biplot analysis across years. This temperature hypothesis was confirmed by a factorial regression analysis. Both the mean temperature during the emergence-heading period and the minimum temperature within a brief period of ±7 d around heading partly explained the observed GEI in our barley data. This finding is in agreement with previous physiological studies to the effect of temperature stress on grain yield and yield components in wheat. High temperature during vegetative stages reduces the final number of tillers, while low temperature allows more tillers to be initiated and to survive later on (Kirby and Riggs, 1978; Garcia del Moral and Garcia del Moral, 1995). Calderini et al. (1999) found that high temperature between heading and anthesis reduced individual grain weight, an important source of variation in grain yield. A similar effect of temperature stress, under field conditions, on individual grain weight of wheat immediately before anthesis has been reported from Argentina and Mexico (Calderini et al., 2001).

For barley, high temperature during grain filling has been reported to affect individual grain weight and grain yield (Sofield et al., 1977; Savin and Nicolas, 1996). Voltas et al. (1999b) reported the effect of high temperature during grain filling on the individual responses of barley cultivars for individual grain weight under Mediterranean conditions. Unfortunately, we could not study the effect of meteorological variables during the grain-filling period because the date of maturity was not recorded in our study.

The results from the factorial regression analysis indicated that part of the GEI can be attributed to differential responses of cultivars to an abiotic stress factor, namely the prevalent temperature conditions, that may be seen as predictable in the Uruguayan conditions, suggesting the possibility of selecting for specific adaptation to contrasting sowing dates or sowing date-location combinations. Evidence for specific adaptation can be observed in GGE biplot (Fig. 5), where the cultivar Defra (long season) performed better in early sowing date trials (lower mean temperature), whereas the cultivar Bowman (short season) showed a better performance in late sowing date trials as well as in the northern environments (higher average temperature). This differential response could be partly based on the capacity of cultivars with different season lengths to escape from the temperature stress associated with sowing date and latitude.

Cultivars' differential responses may also be related to differences in the genetic resistance to main diseases in the crop growing area. Yan and Hunt (2001) reported that better resistance of winter wheat cultivars to leaf blotch was frequently associated with better overall performance. Yet, disease records could not explain cultivar x environment interaction in our barley data set.

Photoperiod response in barley (Tew and Rasmusson, 1978; Barnham and Rasmusson, 1981; Kernich et al., 1995) has been proposed as an important trait for yield stability under Uruguayan conditions (German et al., 2000). Photoperiod response would allow relaxation of planting dates, concentrating anthesis from the end of September to mid-October, thus allowing escape from temperature stress during heading-anthesis as well as during grain filling (German, 2004). An example is CLE 169, which was high yielding and stable, as it performed above average in all of the test environments represented in Fig. 6. Comparably high yielding and stable was cultivar Dayman (Fig. 6), but this cultivar was a medium-season cultivar with low photoperiod response, which represents another strategy for yield stability, like the presence of resistance/tolerance to temperature stress instead of escape mechanisms.

For cultivar recommendation purposes, presently the design of the Uruguayan barley trials network should be oriented toward selecting broadly adapted cultivars (high yielding and stable), while enhancing the chances for identifying cultivars showing specific adaptation to different sowing dates, or sowing date by location combinations (prevalent temperature conditions). If, according to the results of the variance components approach, only three locations with two sowing dates each are to be retained, emphasis should be given to sampling contrasting sowing dates (early June vs. late July) at the most distant locations regarding latitude (LE, YG, and PY). Better environmental characterizations (soil physical and chemical properties, N status) and genotypic descriptions (length of the grain-filling process, yield components, photoperiod sensitivity) can contribute to improved interpretation of GEI and to a further optimization of the trial network.

All rights reserved. No part of this periodical may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopying, recording, or any information storage and retrieval system, without permission in writing from the publisher. Permission for printing and for reprinting the material contained herein has been obtained by the publisher.

Received for publication June 27, 2006.

	REFERENCES

TOP NOTES ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS DISCUSSION REFERENCES

 

• Annicchiarico, P. 1997. Joint regression vs. AMMI analysis of genotype–environment interactions for cereals in Italy. Euphytica 94:53–62.[CrossRef][Web of Science]
• Atlin, G.N., K.B. McRae, and X. Lu. 2000. Genotype x region interaction for two-row barley yield in Canada. Crop Sci. 40:1–6.[Abstract/Free Full Text]
• Barnham, R.W., and D.C. Rasmusson. 1981. Inheritance of photoperiod response in barley. Crop Sci. 21:454–456.[Abstract/Free Full Text]
• Brandle, J.E., and L.M. Arthur. 1992. A comparative evaluation of two methods of selecting locations used for testing spring wheat cultivars. Theor. Appl. Genet. 83:301–304.[Web of Science]
• Calderini, D.F., L.G. Abeledo, R. Savin, and G.A. Slafer. 1999. Final grain weight in wheat as affected by short periods of high temperature during pre and postanthesis under field conditions. Aust. J. Plant Physiol. 26:453–458.[Web of Science]
• Calderini, D.F., R. Savín, L.G. Abeledo, M.P. Reynolds, and G.A. Slafer. 2001. The importance of the period immediately preceding anthesis for grain weight determination in wheat. Euphytica 119:199–204.[CrossRef][Web of Science]
• Campbell, L.G., and H.N. Lafever. 1977. Cultivar x environment interactions in soft red winter wheat yield tests. Crop Sci. 17:604–608.[Abstract/Free Full Text]
• Chapman, S.C., M. Cooper, D.G. Butler, and R.G. Henzell. 2000. Genotype by environment interactions affecting grain sorghum: I. Characteristics that confound interpretation of hybrid yield. Aust. J. Agric. Res. 51:197–207.[CrossRef][Web of Science]
• Comstock, R.E. 1977. Quantitative genetics and the design of breeding programs. p. 705–718. In E. Pollack et al (ed.) Proc. Int. Conf. on Quantitative Genetics, Ames, IA. 16–21 Aug. 1976. Iowa State Univ. Press, Ames.
• Cooper, M., S. Fukai, and L.J. Wade. 1999. How can breeding contribute to more productive and sustainable rainfed lowland rice systems? Field Crops Res. 64:199–209.[CrossRef]
• Cooper, M., D.R. Woodruff, R.L. Eisemann, P.S. Brennan, and I.H. DeLacy. 1995. A selection strategy to accommodate genotype-by-environment interaction for grain yield of wheat: Managed-environments for selection among genotypes. Theor. Appl. Genet. 90:492–502.[Web of Science]
• Cooper, M., D.R. Woodruff, I.G. Phillips, K.E. Basford, and A.R. Gilmour. 2001. Genotype-by-management interactions for grain yield and grain protein concentration of wheat. Field Crops Res. 69:47–67.[CrossRef]
• Crossa, J., and P. Cornelius. 2002. Linear–bilinear models for the analysis of genotype–environment interaction. p. 305–322. In M.S. Kang (ed.) Quantitative genetics, genomics and plant breeding. CAB Int., Wallingford, UK.
• Denis, J.B. 1988. Two-way analysis using covariates. Statistics 19:123–132.[CrossRef]
• Denis, J.B. 1991. Ajustement de modèles linéaires et bilinéaires sous contraintes linéaires avec données manquantes. Rev. Stat. Appl. 39:5–24.
• Gabriel, K.R. 1971. The biplot graphic display of matrices with applications to principal components analysis. Biometrika 58:453–467.[Abstract/Free Full Text]
• Gabriel, K.R. 1978. Least squares approximation of matrices by additive and multiplicative models. J. R. Stat. Soc. Ser. B 40:186–196.
• Garcia del Moral, M.B., and L.F. Garcia del Moral. 1995. Tiller production and survival in relation to grain yield in winter and spring barley. Field Crops Res. 44:85–93.[CrossRef]
• Gauch, H.G., Jr. 1988. Model selection and validation for yield trials with interaction. Biometrics 44:705–715.[CrossRef][Web of Science]
• Gauch, H.G., Jr. 2006. Statistical analysis of yield trials by AMMI and GGE. Crop Sci. 46:1488–1500.[Abstract/Free Full Text]
• Genstat 5 Committee. 1993. Genstat 5 reference manual, 3rd. release. Oxford Univ. Press, Oxford, UK.
• German, S. 2004. Breeding malting barley under stress conditions in South America. Czech J. Genet. Plant Breed. 40:140–147.
• German, S., M. Arbelbide, T. Abadie, R. Romero, and A. Peculio. 2000. Characterization of photoperiod response of barley genotypes from diverse origin. p. 211–214. In S. Logue (ed.) Proc. 8th. Int. Barley Genet. Symp., Adelaide, South Australia. 22–27 Oct. 2000. Vol. 3. Adelaide Univ., SA, Australia.
• Gollob, H.F. 1968. A statistical model which combines features of factor analysis and analysis of variance techniques. Psychometrika 33:73–115.[CrossRef][Web of Science][Medline]
• Graffelman, J., and F.A. van Eeuwijk. 2005. Calibration of multivariate scatter plots for exploratory analysis of relations within and between sets of variables in genomic research. Biom. J. 47:863–879.[CrossRef][Web of Science][Medline]
• Kempton, R.A. 1984. The use of bi-plots in interpreting variety by environment interactions. J. Agric. Sci. 103:123–135.
• Kernich, G.C., G.M. Halloran, and R.G. Flood. 1995. Relative effects of photoperiod and irradiance on preanthesis development in spring barley (Hordeum vulgare, L.). Cereal Res. Commun. 23:425–431.
• Kirby, E.J.M., and T.J. Riggs. 1978. Developmental consequences of two row and six-row ear type in spring barley: 2. Shoot apex, leaf and tiller development. J. Agric. Sci. 91:207–216.
• Lin, C.S., and M.J. Morrison. 1992. Selection of test locations for regional trials of barley. Theor. Appl. Genet. 82:968–972.
• Mandel, J. 1969. The partitioning of interaction in analysis of variance. J. Res. Natl. Bur. Stand. 73B:309–328.
• Patterson, H.D., V. Silvey, M. Talbot, and S.T.C. Weatherup. 1977. Variability of yields of cereals varieties in U.K. trials. J. Agric. Sci. 89:239–245.
• Piepho, H.P., and J. Möhring. 2005. Best linear unbiased prediction of cultivar effects for subdivided target regions. Crop Sci. 45:1151–1159.[Abstract/Free Full Text]
• Robinson, D. 1984. A study of sequential variety selection systems. J. Agric. Sci. 102:119–126.
• Savin, R., and M.E. Nicolas. 1996. Effects of short periods of drought and high temperature on grain of two malting barley cultivars. Aust. J. Plant Physiol. 23:201–210.[Web of Science]
• Sofield, I., L.T. Evans, M.G. Cook, and I.F. Wardlaw. 1977. Factors influencing the rate and duration of grain filling in wheat. Aust. J. Plant Physiol. 3:785–797.
• Talbot, M. 1984. Yield variability of crop varieties in the U.K. J. Agric. Sci. 102:315–321.
• Tew, T.L., and D.C. Rasmusson. 1978. Scoring photoperiod sensitivity in barley. Crop Sci. 18:897–898.[Abstract/Free Full Text]
• van Eeuwijk, F.A. 1995. Linear and bilinear models for the analysis of multi-environment trials: I. An inventory of models. Euphytica 84:1–7.[CrossRef][Web of Science]
• van Eeuwijk, F.A., J.B. Denis, and M.S. Kang. 1996. Incorporating additional information on genotypes and environments in models for two-way genotype by environment tables. p. 15–49. In M.S. Kang and H.G. Gauch, Jr. (ed.) Genotype-by-environment interaction: New perspectives. CRC Press, Boca Raton, FL.
• van Eeuwijk, F.A., L.C.P. Keizer, and J.J. Bakker. 1995. Linear and bilinear models for the analysis of multi-environment trials: II. An application to data from the Dutch maize variety trials. Euphytica 84:9–22.[CrossRef][Web of Science]
• Voltas, J., F.A. van Eeuwijk, J.L. Araus, and I. Romagosa. 1999a. Integrating statistical and ecophysiological analyses of genotype by environment interaction for grain filling of barley: II. Grain growth. Field Crops Res. 62:75–84.[CrossRef]
• Voltas, J., F.A. van Eeuwijk, A. Sombrero, A. Lafarga, E. Igartua, and I. Romagosa. 1999b. Integrating statistical and ecophysiological analysis of genotype by environment interaction for grain filling of barley: I. Individual grain weight. Field Crops Res. 62:63–74.[CrossRef]
• Yan, W., and L.A. Hunt. 2001. Interpretation of genotype x environment interaction for winter wheat yield in Ontario. Crop Sci. 4:19–25.
• Yan, W., L.A. Hunt, Q. Sheng, and Z. Szlavnics. 2000. Cultivar evaluation and mega-environment investigation based on the GGE biplot. Crop Sci. 40:597–605.[Abstract/Free Full Text]
• Yan, W., and M.S. Kang. 2003. GGE biplot analysis: A graphical tool for breeders, geneticists and agronomists. CRC Press, Boca Raton, FL.
• Yan, W., and N.A. Tinker. 2006. Biplot analysis of multi-environment trial data: Principles and applications. Can. J. Plant Sci. 83:623–645.

This Article Abstract Figures Only Full Text (PDF) Free Alert me when this article is cited Alert me if a correction is posted Services Similar articles in this journal Alert me to new issues of the journal Download to citation manager Citing Articles Citing Articles via Google Scholar Google Scholar Articles by Ceretta, S. Articles by van Eeuwijk, F. Search for Related Content PubMed Articles by Ceretta, S. Articles by van Eeuwijk, F. Agricola Articles by Ceretta, S. Articles by van Eeuwijk, F. Related Collections Biometrics Other Crops Data Management