Abstract

Genotype–environment interactions (GEI) limit genetic gain for complex traits such as tolerance to drought. Characterization of the crop environment is an important step in understanding GEI. A modelling approach is proposed here to characterize broadly (large geographic area, long-term period) and locally (field experiment) drought-related environmental stresses, which enables breeders to analyse their experimental trials with regard to the broad population of environments that they target. Water-deficit patterns experienced by wheat crops were determined for drought-prone north-eastern Australia, using the APSIM crop model to account for the interactions of crops with their environment (e.g. feedback of plant growth on water depletion). Simulations based on more than 100 years of historical climate data were conducted for representative locations, soils, and management systems, for a check cultivar, Hartog. The three main environment types identified differed in their patterns of simulated water stress around flowering and during grain-filling. Over the entire region, the terminal drought-stress pattern was most common (50% of production environments) followed by a flowering stress (24%), although the frequencies of occurrence of the three types varied greatly across regions, years, and management. This environment classification was applied to 16 trials relevant to late stages testing of a breeding programme. The incorporation of the independently-determined environment types in a statistical analysis assisted interpretation of the GEI for yield among the 18 representative genotypes by reducing the relative effect of GEI compared with genotypic variance, and helped to identify opportunities to improve breeding and germplasm-testing strategies for this region.

Introduction

Progress in plant improvement for yield relies on the identification of genotypes better adapted to their environment. However, the complexity of genotype–environment interactions (GEI), in particular in drought-prone environments, can reduce heritability of yield, i.e. the ability statistically to identify superior genotypes. In the ‘target population of environments’ (TPE), i.e. the set of environments in which cultivars can be grown within the geographical area targeted by a breeding programme (Comstock, 1977), GEI impede plant breeding progress for both (i) broad adaptation across the TPE and (ii) adaptation to specific types of environments within the TPE (Cooper and Hammer, 1996; Löffler et al., 2005; Dreccer et al., 2008; Mathews et al., 2008).

The effectiveness of genotype evaluation as part of breeding is influenced by (i) understanding of GEI and (ii) the degree to which the environments sampled in the multi-environment testing trials (MET) represent the TPE (Podlich and Cooper, 1998). While statistical approaches to estimate GEI have developed over more than 50 years, it has only been relatively recently that the cause of GEI has been investigated, with specific environment variables being included in mixed models (Voltas et al., 1999; van Eeuwijk et al., 2005) and with environment classifications being identified in principal component analysis (Annicchiarico and Perenzin, 1994; Yan et al., 2000). In parallel with these methodological improvements to analyse GEI within a set of trials (MET), characterization of the complete TPE has been conducted based on (i) climate and soil data (Runge, 1968; Pollak and Corbett, 1993; Hodson and White, 2007), (ii) physiological traits that integrate various stresses that plants experience, for example, flowering date or yield (Chauhan et al. 2008; Hernandez-Segundo et al., 2009), and (iii) stress index determined with crop models (Chapman et al., 2000a, b).

In practice, breeders typically characterize their trials based on check-variety performance rather than environmental data, undertaking as many trials as resources allow. While this approach provides a solid platform for statistical analysis of GEI within their MET, it may often misrepresent the TPE, mainly due to cost limitation and the impracticality of large-scale MET to sample the TPE. By contrast, modelling tools applied to multi-site long-term studies allow a more comprehensive environmental sampling (Löffler et al., 2005; Hammer and Jordan, 2007). Furthermore, these tools can also estimate local environmental variables and account for the interactions between the plants and their environment (e.g.Chelle, 2005; Chenu et al., 2008), which thus allow more detailed physiological interpretation of GEI basis (Reymond et al., 2004; Chenu et al., 2007).

Crop models capture feedback between plant growth and soil water depletion and have been shown to characterize water-limited environments better when compared with standard indices computed from climatic data (Muchow et al., 1996). Applying this method, Chapman et al. (2000,a) characterized the water-deficit patterns experienced by sorghum crops over a long-term period in six locations, and grouped them into environment types. The frequency of these environment types at specific locations correlated with the patterns of discrimination (principal component analysis) for yield among hybrids at those locations over c. 15 years of testing. Chapman et al. (2000,b) also used these environment types to characterize the environment of the Australian sorghum-growing area during the 20th century and concluded, as did Podlich and Cooper (1998), that weighting the multi-environment trials of breeding programmes, based on their representativeness in the TPE, would improve the rate of yield gain. Furthermore, a recent study on wheat showed that the negative influence of GEI on ranking of wheat genotypes could be reduced when trials were classified based on crop drought patterns (Chapman, 2008).

This study aimed to characterize the water-deficit patterns of an entire TPE and to demonstrate the utility of this broad characterization for local breeding trials. The focus here is on the north-eastern wheat-growing area of Australia, as this extensive TPE (more than 1500 km N–S and c. 500 km E–W) experiences large spatial and temporal variability in the seasonal pattern of water availability due to contrasting soil types, management systems, and inter-annual rainfall (Potgieter et al., 2002). In such environments, genotype ranking varies greatly among locations and seasons, which limits the potential rate of genetic gain by decreasing the effective heritability for selection (Brennan et al., 1981; Basford and Cooper, 1998). Here, the goals were to (i) use historical weather data to characterize the nature of the main water-deficit patterns occurring in the TPE, (ii) determine the frequency of these main environment types for each region within this area, for typical management and over a long-term period, and (iii) analyse a set of trials to demonstrate that this typology of the TPE can aid interpretation of GEI. Simulations were performed over 119 years, for 12 regions (14 locations) distributed across north-eastern Australia, with representative soil types for each region. Weighting factors were assigned to each site (location×soil combination) so that all sites corresponded to similar fractions of the total area of wheat planted. The nature and frequency of major environmental types were defined based on seasonal water-deficit patterns that were centred around flowering. These environment types were used to analyse and interpret the genotypic variability for yield of 18 lines selected from previous broad-scale evaluations in 16 irrigated and rain-fed trials (MET).

Materials and methods

Overview

The wheat-cropping area of north-eastern Australia was divided into 12 regions, each represented by 1–3 sites (combinations of weather station×soil; Table 1). These sites were chosen to have long-term and good quality climatic weather data, and to have soil characteristics as representative as possible of the region (soil information had been collected by local experts from the Agricultural Production Systems Research Unit (APSRU), http://www.asris.csiro.au/grapher/APSRUkml.aspx and from Daniells et al., 2002). A weighting factor (Table 1) was assigned so that each site represented between 73 000 and 91 000 ha of wheat planted area (average data over 1975–1999; Australian Bureau of Statistics, http://www.abs.gov.au/).

Table 1.

Characteristics of the sites and soils chosen to represent the north-eastern wheat-cropping area of Australia

Cropping area Region Weather station Latitude Longitude Soil PAWC (mm) PAWC at sowing (mm) Weighting factor 
Central Queensland Central Highlands Emerald –23.53 148.16 134 107-125-131-134-134 
 Dawson Callide Biloela –24.38 150.52 198 121-154-186-196-198 
South Queensland Western Darling Downs Roma –26.57 148.79 119 94-109-118-119-119 
 Central Darling Downs Miles –26.66 150.180 159 105-134-149-159-159 
     201 105-137-167-191-201 
 Eastern Darling Downs Dalby –27.17 151.27 205 119-164-193-203-205 
     285 124-169-206-234-282 
 St George St George –28.04 148.58 169 91-120-147-164-169 
 Southern Darling Downs Goondiwindi –28.55 150.31 159 94-121-146-157-159 
     222 102-129-158-189-222 
Northern New South Wales Moree Moree –29.50 149.90 153 90-116-141-150-153 
     194 93-122-149-180-194 
     238 100-124-155-190-233 
 Narrabri Narrabri –30.34 149.76 232 107-151-181-220-232 
  Walgett –30.04 148.12 232 93-109-132-165-225 
 Coonamble Coonamble –30.98 148.38 197 87-106-134-160-191 
 Dubbo Dubbo –32.21 148.57 128 109-128-128-128-128 
     142 104-131-141-142-142 
 Gunnedah Coonabarabran –31.27 149.27 142 128-139-142-142-142 
  Gunnedah –30.98 150.25 153 92-121-142-151-153 
     282 102-135-169-206-262 
Cropping area Region Weather station Latitude Longitude Soil PAWC (mm) PAWC at sowing (mm) Weighting factor 
Central Queensland Central Highlands Emerald –23.53 148.16 134 107-125-131-134-134 
 Dawson Callide Biloela –24.38 150.52 198 121-154-186-196-198 
South Queensland Western Darling Downs Roma –26.57 148.79 119 94-109-118-119-119 
 Central Darling Downs Miles –26.66 150.180 159 105-134-149-159-159 
     201 105-137-167-191-201 
 Eastern Darling Downs Dalby –27.17 151.27 205 119-164-193-203-205 
     285 124-169-206-234-282 
 St George St George –28.04 148.58 169 91-120-147-164-169 
 Southern Darling Downs Goondiwindi –28.55 150.31 159 94-121-146-157-159 
     222 102-129-158-189-222 
Northern New South Wales Moree Moree –29.50 149.90 153 90-116-141-150-153 
     194 93-122-149-180-194 
     238 100-124-155-190-233 
 Narrabri Narrabri –30.34 149.76 232 107-151-181-220-232 
  Walgett –30.04 148.12 232 93-109-132-165-225 
 Coonamble Coonamble –30.98 148.38 197 87-106-134-160-191 
 Dubbo Dubbo –32.21 148.57 128 109-128-128-128-128 
     142 104-131-141-142-142 
 Gunnedah Coonabarabran –31.27 149.27 142 128-139-142-142-142 
  Gunnedah –30.98 150.25 153 92-121-142-151-153 
     282 102-135-169-206-262 

The main cropping areas (Central Queensland, South Queensland, and Northern New South Wales) were subdivided into smaller regions. Weighting factors were imposed for the sites (combination of weather station×soil) so that the number of situations per region was proportional to its average area planted to wheat. Plant available water content (PAWC) is indicated for the different soils used in the simulations. The five different levels of PAWC used at sowing in each site each corresponded to 20% of the sowing soil-water conditions estimated from an initial set of simulations.

Using the APSIM crop model (Wang et al., 2002), an initial set of simulations was performed over 119 years of historical-climatic data to characterize sowing date and soil water profile at planting, which would be encountered at each site. Based on these initial conditions, a second set of simulations was run to characterize the seasonal water-deficit patterns occurring at the different sites for the 119 years. Cluster analysis was applied to this dataset to identify the major environment types of the wheat-growing area of north-eastern Australia.

A third set of simulations was performed to characterize the water-deficit pattern of 16 irrigated and rain-fed trials and to identify the environment type of each trial. This environmental classification was used in genetic analyses to interpret the GEI for yield in this MET.

Determination of the initial conditions (sowing date and soil water at sowing) in the target population of environments (TPE)

To be successful in this summer-dominant rainfall area, wheat crops require a rainfall event before planting, and sufficient stored water for access later during their development. To characterize the initial conditions (sowing date and soil water content at planting) occurring in each location×soil combination (called ‘site’), cropping system simulations were run using the APSIM-Wheat crop model (Keating et al., 2003) for each site for the years 1889 to 2008. In each simulation, a fallow (5–8 months) followed by a wheat crop (Hartog cultivar, no nitrogen limitation, minimum tillage, and no weed, pest or disease) and followed by another fallow was simulated to mimic farmer practices and estimate (i) the timing of planting events and (ii) the moisture retained in the soil at these dates. The planting opportunities were estimated within an 8-week planting window specific to each region, as consistent with local farmer practice. Based on consultations with experienced agronomists and previous studies (Hammer et al., 1987), a planting opportunity was defined to occur when (i) the fraction of extractable water was between 0.6 and 0.9 in the surface soil layer, so that seeds could germinate, but the soil was dry enough for sowing traffic, (ii) the extractable stored soil water in the entire profile was higher than 80 mm, and (iii) cumulative rainfall in the previous 10 d was at least 10 mm.

This initial set of simulations revealed that sowing opportunities were uniformly distributed throughout the planting window. Given this uniform distribution in planting opportunities, sowing dates for the second set of simulations (see next section) were chosen to be 2-weeks apart within the planting window, with each date representing around 20% of the sowing opportunities occurring at each site. Sowing dates were: 15 April, 1 May, 15 May, 1 June, and 15 June for Central Queensland; 1 May, 15 May, 1 June, 15 June, and 1 July for South Queensland; and 15 May, 1 June, 15 June, 1 July, and 15 July for northern New South Wales. For each site, the initial soil water conditions were found to be similar for all the sowing opportunities, showing that rainfall occurring during the previous fallow was more important than the small rainfall events occurring during the planting window. Hence, the second set of simulations used five different initial soil water conditions for each site (without distinction across sowing dates; Table 1), each representing 20% of the conditions encountered during the planting window for the considered sites over 119 years.

Crop simulation and environmental characterization in the TPE

Simulations were performed for a standard check variety, Hartog, at the 25 sites distributed over the actual wheat-cropping area of north-eastern Australia (Table 1). Simulations were performed in APSIM, for the years 1889 to 2008, with no biotic constraint, no nitrogen limitation (180 kg ha−1 of nitrogen in the soil at sowing), with minimum tillage, for the five planting dates and five initial soil water conditions identified above (25 simulations per year and per site), i.e. c. 75 000 simulations in total.

Each environment (location×year×sowing date×initial soil water combination) was characterized based on a daily crop water stress index, corresponding to the ratio between water supply and demand. This index indicates the degree to which the potential water supply that depends on the volume and wetness of soil explored by roots is able to match the water demand of the canopy, which is influenced by radiation and temperature and air humidity conditions (Chapman et al., 1993, 2000a, b).

For each simulation, the temporal pattern of this water stress index was centred at flowering and averaged every 100 °Cd (between emergence and maturity). A cluster analysis was applied to these 75 000 water-deficit patterns using the partitioning clustering function clara in the R statistical package (R Development Core Team, 2008) in order to identify the three major environment types (‘ET1-2-3’) encountered in the TPE. This method is based on minimizing the sum of the dissimilarities (i.e. pair-wise distances) between the pattern of water stress index of (i) each environment of the TPE and (ii) the median situation of the environment type considered. The partitioning clustering method is better suited to this large dataset than the hierarchical method that was used for smaller datasets in previous studies. An average pattern of water deficit was calculated to describe each environment type. The occurrence of different environment types was interpreted with respect to different regions, years, starting soil water conditions, and sowing date.

Multi-environment trials (MET): experimental conditions and simulations

Eighteen wheat lines (Triticum aestivum L.; Cooper et al., 1997) were grown in 16 field trials (Qiao et al., 2000, 2004) in the north-eastern wheat-growing area of Australia, between 1994 and 1996 (Table 2). These lines were selected in a breeding programme to represent advanced CIMMYT lines and a few local check genotypes, for yield performance and GEI for yield in Queensland drought environments (Cooper et al., 1994,a, b). The released cultivars (‘named’ genotypes, e.g. ‘Hartog’) and broadly-adapted wheat lines from the 17th International bread wheat nursery distributed by CIMMYT, Mexico (‘numbered’ genotypes, e.g. ‘17IB206’) that were studied here are: Hartog (‘G1’), Cunningham (‘G2’), Banks (‘G3’), Pelsart (‘G4’), 17IB7 (‘G5’), 17IB30 (‘G6’), 17IB31 (‘G7’), 17IB38 (‘G8’), 17IB53 (‘G9’), 17IB64 (‘G10’), 17IB92 (‘G11’), 17IB129 (‘G12’), Seri (17IB206, ‘G13’), 11IB50 (‘G14’), Genaro (‘G15’), 17IB173 (‘G16’), Sunco (‘G17’), and Kite (‘G18’).

Table 2.

Characteristics of the multi-environment trials (MET), which comprised 16 trials carried out across the north-eastern wheat-cropping area of Australia (experiments described by Qiao et al., 2000, 2004)

Trial Year Location Latitude Longitude Sowing date Fertilizationa (kg ha-1Irrigation treatment Total water supply (mm)b Environment type 
E1 1994 Gatton –27.55 152.33 10 May 174 Irrigated 473.4 ET1 
E2 1994 Jimbour –6.94 151.13 19 Jun 150 – 111.3 ET3 
E3 1994 Meandarra –27.32 149.87 21 Jun 150 – 75.4 ET3 
E4 1994 Oakleigh –27.32 151.44 31 May 150 – 145.3 ET3 
E5 1994 Fernless –23.92 148.39 08 May 106 – ET3 
E6 1994 The Gum –27.33 150.20 13 Jun 150 – 79.8 ET3 
E7 1995 Billa Billa –28.10 150.31 28 May 150 – 86.8 ET3 
E8 1995 Biloela –24.40 150.50 22 May 130 Partially irrigated 336.2 ET3 
E9 1995 Bungunya –28.42 149.65 28 May 128 – 133.7 ET3 
E10 1995 Gatton –27.55 152.33 14 May 174 Partially irrigated 248.7 ET2 
E11 1995 Tummaville –27.87 151.48 25 Jun 150 – 406.3 ET2 
E12 1996 Bungunya –28.42 149.65 27 May 120 – 152.3 ET2 
E13 1996 Gatton –27.55 152.33 23 May 196 Irrigated 288.7 ET1 
E14 1996 Jimbour –26.94 151.13 06 Jun 150 – 239.9 ET1 
E15 1996 Moonie –28.33 150.30 02 Jun 150 – 254.9 ET2 
E16 1996 Fernless –23.92 148.39 10 May 111 – 100.8 ET2 
Trial Year Location Latitude Longitude Sowing date Fertilizationa (kg ha-1Irrigation treatment Total water supply (mm)b Environment type 
E1 1994 Gatton –27.55 152.33 10 May 174 Irrigated 473.4 ET1 
E2 1994 Jimbour –6.94 151.13 19 Jun 150 – 111.3 ET3 
E3 1994 Meandarra –27.32 149.87 21 Jun 150 – 75.4 ET3 
E4 1994 Oakleigh –27.32 151.44 31 May 150 – 145.3 ET3 
E5 1994 Fernless –23.92 148.39 08 May 106 – ET3 
E6 1994 The Gum –27.33 150.20 13 Jun 150 – 79.8 ET3 
E7 1995 Billa Billa –28.10 150.31 28 May 150 – 86.8 ET3 
E8 1995 Biloela –24.40 150.50 22 May 130 Partially irrigated 336.2 ET3 
E9 1995 Bungunya –28.42 149.65 28 May 128 – 133.7 ET3 
E10 1995 Gatton –27.55 152.33 14 May 174 Partially irrigated 248.7 ET2 
E11 1995 Tummaville –27.87 151.48 25 Jun 150 – 406.3 ET2 
E12 1996 Bungunya –28.42 149.65 27 May 120 – 152.3 ET2 
E13 1996 Gatton –27.55 152.33 23 May 196 Irrigated 288.7 ET1 
E14 1996 Jimbour –26.94 151.13 06 Jun 150 – 239.9 ET1 
E15 1996 Moonie –28.33 150.30 02 Jun 150 – 254.9 ET2 
E16 1996 Fernless –23.92 148.39 10 May 111 – 100.8 ET2 
a

Fertilization at sowing.

b

Cumulative rain and irrigation for the sowing-harvest period.

In the 16 trials, plants were sown at a density of 85–100 plants m−2, in c. 15 m2 plots, with a triple lattice, three replicate design (Qiao et al., 2000, 2004). In order to test the model, above-ground biomass was measured twice before flowering and flowering date was recorded for the standard Hartog cultivar. Yield was measured at maturity for the 18 lines.

MET trials were simulated for cultivar Hartog using climatic and soil data collected at each site, and their water-deficit patterns were determined as described above for the TPE. The trials were classified based on which environment type from the TPE they were the most similar to, i.e. based on the minimum sum of squared differences for the trial water-deficit pattern compared with the water-deficit pattern of the TPE environment types.

MET: statistical analyses and heritability

The variance components for the genotype and GEI were estimated for both:

  • (i)

    the classical compound symmetry model:  

    (1)
    graphic

  • (ii)

    and a nested compound symmetry model to test the G and GEI contribution within each environment type:  

    (2)
    graphic

where P is the phenotype (yield); E is the fixed environment effect; G and Gwithin each ET are the random genotype effects in all environments and within each environment type, respectively; GEI and Gwithin each ET are the random genotype-by-environment effects in all environments and within each environment type, respectively; and ε is the residual trial error.

The genotype mean heritability (h2) was estimated for each environment type and for all trials of the MET, as:

 
(3)
graphic

where G, GEI, and ε are the variance components and the residual error described above, nE is the number of environments (i.e. number of trials for the considered environment type or for the MET), and nR is the number of replications per trial (three for these trials).

In order to estimate the number of trials needed to have an acceptable heritability in the different environment groups (as presented in Fig. 7), an expected heritability was also calculated for a number of trials (nE) varying from 1 to 10 for each group of environments, with the values of G, GEI, and ε as observed in the MET (Table 3), and with two repetitions (nR=2; as in a usual breeding trial).

Table 3.

Heritability (A) and mixed model analyses of yield in the MET either without (B), or with (C), the environment characterization. (B) Classical compound symmetry model (equation 1). (C) Compound symmetry which nests the components for the genotypes (G) and for the genotype–environment interactions (GEI) within independently assigned environment types.

(A) Heritability
 
Environment group Heritability 
All trials 0.592 
ET1 trials 0.564 
ET2 trials 0.408 
ET3 trials 0.467 
(A) Heritability
 
Environment group Heritability 
All trials 0.592 
ET1 trials 0.564 
ET2 trials 0.408 
ET3 trials 0.467 
(B) Classical GEI model 
Source Variance component Standard error 
0.053 0.022 
GEI 0.193 0.018 
Residual variance 0.073  
(B) Classical GEI model 
Source Variance component Standard error 
0.053 0.022 
GEI 0.193 0.018 
Residual variance 0.073  
(C) Model with the environment characterization
 
Source Variance component Standard error 
Gwithin ET1 0.124 0.072 
Gwithin ET2 0.054 0.039 
Gwithin ET3 0.036 0.018 
GEIwithin ET1 0.214 0.057 
GEIwithin ET2 0.268 0.049 
GEIwithin ET3 0.130 0.018 
Pooled residual variance 0.073  
(C) Model with the environment characterization
 
Source Variance component Standard error 
Gwithin ET1 0.124 0.072 
Gwithin ET2 0.054 0.039 
Gwithin ET3 0.036 0.018 
GEIwithin ET1 0.214 0.057 
GEIwithin ET2 0.268 0.049 
GEIwithin ET3 0.130 0.018 
Pooled residual variance 0.073  

Analyses were performed with genotype means (BLUEs, Best Linear Unbiased Estimates) and relative weights from existing analyses (Qiao et al., 2000), using the ASREML software (Gilmor et al., 1997, 2002) and the methods described by Cullis et al. (1996) and Chapman et al. (2000a).

Phenotypic correlations among the original BLUEs were calculated for all possible pairs of trials, both within and across environment types, with R (R Development Core Team, 2008). A principal component analysis was applied to the G×E matrix of BLUEs with the environment standardized data transformation (Fox and Rosielle, 1982; Cooper and DeLacy, 1994). The analysis and the resulting biplot were produced using R.

Results

Three main environment types for the wheat-cropping area of north-eastern Australia

Long-term simulations in the wheat-cropping area of north-eastern Australia highlighted the high variability of wheat yield across years and regions (Fig. 1), in accordance with wheat production data (Potgeiter et al., 2002). Average yield ranged between 2.0 t ha−1 (Central Highlands and Western Darling Downs) and 3.7 t ha−1 (Eastern Darling Downs), with an inter-annual range of more than 4 t ha−1 between the most extreme years in each region. The occurrence of water-deficit patterns was also highly variable across regions and seasons, ranging from stress-free environments where the simulated yield reached close to 7 t ha−1, to severely stressed crops that yielded 0.5 t ha−1. Given the initial soil-water conditions required for planting, there was no crop failures. Note that in the real-world, crop failures can occur as the result of a catastrophic event due to pests, flooding etc (not simulated here) or when farmers choose to plant crops in high-risk conditions (e.g. low initial soil water).

Fig. 1.

Distribution of the simulated yield of cv. Hartog over 119 years for the different regions, ordered north to south. Upper and lower dimensions of the boxes represent the third and first quartiles, the medians are indicated by bars inside the boxes, and the circles represent outlying values below the 10th and above the 90th percentiles.

Fig. 1.

Distribution of the simulated yield of cv. Hartog over 119 years for the different regions, ordered north to south. Upper and lower dimensions of the boxes represent the third and first quartiles, the medians are indicated by bars inside the boxes, and the circles represent outlying values below the 10th and above the 90th percentiles.

Using a cluster analysis, three major water-deficit patterns were identified, explaining 65% of the variation in the c. 75 000 simulations (Fig. 2): the first environment type (ET1) comprised situations where plants were effectively not limited by water or experienced only short-term water deficit; the second environment type (ET2) was characterized by mid-season water deficit, starting during the vegetative period and relieved by rainfall events during grain filling; and the third environment type (ET3) also had a mid-season water deficit beginning at around the same time as in ET2, but continuing through to plant maturity.

Fig. 2.

Water-stress index throughout the crop cycle, represented as a function of thermal time relative to flowering, for three environment types (ET1-2-3) identified in the north-eastern wheat-growing area of Australia, for cv. Hartog. Inset, coefficient of determination for the clustering of the different number of environment types. This water stress index corresponds to a water supply/demand ratio that integrates the crop demand (potential transpiration depending on the climate and plant status) and resource supply available to the roots on the considered date. A ratio of 1 indicates no water stress, while a ratio of 0 corresponds to a full-stress, with no water available to the crop. Data of all the environments of the TPE were centred at flowering, averaged every 100 °Cd and clustered in three environment types. The data presented here correspond to the average of the water stress index of the different environments of the TPE that belonged to ET1-2-3, respectively.

Fig. 2.

Water-stress index throughout the crop cycle, represented as a function of thermal time relative to flowering, for three environment types (ET1-2-3) identified in the north-eastern wheat-growing area of Australia, for cv. Hartog. Inset, coefficient of determination for the clustering of the different number of environment types. This water stress index corresponds to a water supply/demand ratio that integrates the crop demand (potential transpiration depending on the climate and plant status) and resource supply available to the roots on the considered date. A ratio of 1 indicates no water stress, while a ratio of 0 corresponds to a full-stress, with no water available to the crop. Data of all the environments of the TPE were centred at flowering, averaged every 100 °Cd and clustered in three environment types. The data presented here correspond to the average of the water stress index of the different environments of the TPE that belonged to ET1-2-3, respectively.

Occurrence of the environment type highly varied across regions, over time and among management practices

The distribution of the three environment types varied across regions (Fig. 3). For the entire north-eastern Australian wheat-growing area, the frequency of occurrence of the environment types ET1:2:3 was 16:34:50%, respectively. The generally stress-free conditions (ET1) only represented between 6% and 38% of the years among the different regions, while the severe ET3 stresses occurred at 33–65% depending on the region. Except for ET3, which was more frequent in the driest years, the frequency of occurrence of the environment types was poorly correlated with the cumulative rainfall within the season (data not shown). For example, the Dawson Callide in the north is one of the regions with lowest rainfall, but it had a similar frequency of ET1 to the highest rainfall regions in the south (Dubbo and Gunnedah).

Fig. 3.

Map of the frequencies of each environment type (pie charts) and simulated yield distribution of cv. Hartog within each environment type (box plots), for the different regions of the north-eastern wheat-growing area of Australia. Data from simulations over 119 years of historical climate data. The size of the pie charts is proportional to the wheat-planted area in the associated region.

Fig. 3.

Map of the frequencies of each environment type (pie charts) and simulated yield distribution of cv. Hartog within each environment type (box plots), for the different regions of the north-eastern wheat-growing area of Australia. Data from simulations over 119 years of historical climate data. The size of the pie charts is proportional to the wheat-planted area in the associated region.

High variability in the frequency of different environment types has also been observed over the years (Fig. 4A). Decades when more than 50% of sites in the wheat-growing area experienced an ET1 were relatively rare (0–3 years per decade), while they frequently had more than 50% of sites with ET3 (2–7 years per decade). Note that the results presented here do not correspond with the historical occurrences of these environment types but to their occurrence when considering years as independent, using identical initial conditions (sowing date and soil water at sowing). This remark also applies to the simulated yield results. Nonetheless, the results reflected well the incidence of El Niño Southern Oscillation (ENSO) events and their effect on wheat yield (Potgieter et al., 2002).

Fig. 4.

Frequency of occurrence of the different environment types (A) over 119 years of climatic data in the north-eastern wheat-growing area of Australia, and (B) for each combination of sowing date and initial soil water used in the simulations. (A) Annual yield averaged for the area was superimposed as a line. Note that the yield and environment type simulated here does not correspond to historical yield and drought occurrence, as initial conditions in the simulations were calculated for the entire TPE and not for a particular year (i.e. the same set of sowing dates and initial soil water for each year in the simulations). (B) Sowing dates from the earliest (1) to the latest (5) were separated by 2 week intervals. Initial soil water increased from the lowest (1; most severe conditions) to highest (5), each representing 20% of the initial soil water encountered in the initial round of simulations. Frequency data presented here correspond to all the simulations performed (all sites over 119 years).

Fig. 4.

Frequency of occurrence of the different environment types (A) over 119 years of climatic data in the north-eastern wheat-growing area of Australia, and (B) for each combination of sowing date and initial soil water used in the simulations. (A) Annual yield averaged for the area was superimposed as a line. Note that the yield and environment type simulated here does not correspond to historical yield and drought occurrence, as initial conditions in the simulations were calculated for the entire TPE and not for a particular year (i.e. the same set of sowing dates and initial soil water for each year in the simulations). (B) Sowing dates from the earliest (1) to the latest (5) were separated by 2 week intervals. Initial soil water increased from the lowest (1; most severe conditions) to highest (5), each representing 20% of the initial soil water encountered in the initial round of simulations. Frequency data presented here correspond to all the simulations performed (all sites over 119 years).

Simulated yield was highly constrained by the water-deficit pattern (Figs 3, 4A). In each region, crops grown in ET1 seasons had the highest yield (Fig. 3). In ET2, crops tended to yield more than in ET3; however, the yield distribution in ET2 and ET3 was similar for many regions (e.g. Central Highlands). The range of simulated yield for each environment type remained broad, and the large overlaps in yield performance among environment types indicated (i) the importance of other environmental factors not included in this analysis (e.g. due to warmer temperatures and shorter times to flowering, mean yields in the ET1 seasons in the northern regions (Central Highlands and Dawson Callide) were at least 0.6 t ha−1 lower than in other regions), and (ii) the limitation of clustering so many environments into three groups, so that short-term stresses affecting yield were smoothed in the averaging and clustering process.

Management practice also had an impact on the frequency of occurrence of the different environment types. For the standard variety used (Hartog), delaying sowing reduced the occurrence of ET1, and increased the occurrence of ET2 (Fig. 4B): later-sown crops were subjected to more water stress around flowering, as this period of their development occurred under higher temperature and water demand conditions in the latter part of spring.

As might be expected, increased initial soil water at any sowing date led to an increase in the frequency of ET1 and a decrease in the frequency of ET2 and ET3. This impact of initial water was greatest at the earlier sowing dates, and decreased with the later sowing dates.

Environment characterization helped in understanding the genotype–environment interactions in the MET

Sixteen trials of a breeding programme (MET; Fig. 5A) were simulated and their water-deficit patterns were characterized for the cultivar Hartog. The simulations over-predicted biomass production slightly, but predicted final yield well for this genotype (Fig. 5B, C), suggesting that transpiration patterns and water-deficit responses were reasonably well approximated.

Fig. 5.

Map of the 16 wheat trials of the MET, and simulated versus observed biomass (A) and grain yield (B) for the standard cv. Hartog in the MET. Dotted line, 1:1 line; solid line, fitted linear relationship. (This figure is available in colour at JXB online.)

Fig. 5.

Map of the 16 wheat trials of the MET, and simulated versus observed biomass (A) and grain yield (B) for the standard cv. Hartog in the MET. Dotted line, 1:1 line; solid line, fitted linear relationship. (This figure is available in colour at JXB online.)

Compared with the entire TPE, the 16-trial MET, which included several partially-irrigated trials, slightly over-represented ET1 (22% versus 16% in the TPE), and slightly under-represented ET2 (28% versus 34%), while both the TPE and the MET had the same proportion of ET3 (50%).

In the MET, average yield decreased substantially from ET1 to ET3 for all lines (Fig. 6A), with the GEI variance component being more than three times that of the genotype component (Table 3). Three broadly-adapted lines were high yielding across most trials (lines 17B30, G6; Seri, G13, and Genaro, G15) but the ranking of the remaining genotypes varied substantially between trials (Fig. 6A).

Fig. 6.

Genotype–environment interactions for yield in the MET (18 lines in the 16 trials). (A) Yield of the lines in each trial. Trials of each environment type were ordered from high to low average yield among the genotypes. Three lines are in bold in order to highlight the genotype–environment interactions within and across environment types. (B) Distribution of the genetic correlations between trials and within and across environment types. (C) Biplot of the principal component analysis for the different trials (coloured by environment type) and genotypes.

Fig. 6.

Genotype–environment interactions for yield in the MET (18 lines in the 16 trials). (A) Yield of the lines in each trial. Trials of each environment type were ordered from high to low average yield among the genotypes. Three lines are in bold in order to highlight the genotype–environment interactions within and across environment types. (B) Distribution of the genetic correlations between trials and within and across environment types. (C) Biplot of the principal component analysis for the different trials (coloured by environment type) and genotypes.

The best phenotypic correlations were found for trials within each environment type, and between trials in ET1 and ET2 (Fig. 6B). Consistently, trials in ET1 and ET2 were grouped together in the principal component analysis (Fig. 6C), although the average yields in ET1 were substantially higher than in ET2 for all lines. Poorer correlations were observed between trials in ET2 and ET3, and between trials in ET1 and ET3. Accordingly, ET3 trials were quite distinct from the ET1–2 group in the principal component analysis (Fig. 6C).

Including the environment characterization in the statistical models assisted in interpreting GEI of the MET (Table 3). In the classical analysis (equation 1), the genotype component was 27% of the GEI component. When considering the different environment types (equation 2), this ratio increased substantially for the trials in ET1 (58%), slightly decreased in ET2 (20%), and remained unchanged in ET3 (27%). This indicated that genotype performance is more predictable in stress-free environments than under drought, although phenotypic correlations within ET1, 2, and 3 had similar median values (Fig. 6B). Accordingly, heritability was higher for the ET1 trials (h2 of 0.56) than for the other environment types (h2 of 0.41–0.47; Table 3).

A consequence of the differences in G and GEI variance components observed across environment types is that the number of trials required to obtain a given level of heritability depends on the environment type to which these trials belong (Fig. 7). For instance, breeders who target a heritability of 0.6, would need to harvest about 3, 9 or 7 trials in ET1, ET2 or ET3 environments, respectively. This interpretation should be valid across the stages of testing in this breeding programme, as the genotypic diversity of the lines is representative of that being evaluated by the breeding programme.

Fig. 7.

Expected heritability (h2) for 1–10 environments, given the variance estimates for genotype (G), genotype–environment interaction (GEI), and residual error observed in the MET (Table 3B, C). Heritability was calculated for environment of any type (i.e. random environment) and for environments of each environment type. Estimates assume two replicates per trial. (This figure is available in colour at JXB online.)

Fig. 7.

Expected heritability (h2) for 1–10 environments, given the variance estimates for genotype (G), genotype–environment interaction (GEI), and residual error observed in the MET (Table 3B, C). Heritability was calculated for environment of any type (i.e. random environment) and for environments of each environment type. Estimates assume two replicates per trial. (This figure is available in colour at JXB online.)

Discussion

The north-eastern wheat-growing area of Australia has a highly variable TPE in terms of drought patterns

Annual wheat production in north-eastern of Australia is highly dependent on rainfall, which varies greatly across years. From 75 000 simulations using historical weather data, three major environment types were identified to classify the water-deficit patterns occurring in this cropping area. Despite their classification by the main limiting factor (drought), high variability in yield remained within each environment type (Fig. 3), for two major reasons. Firstly, classifying environments of such a broad TPE inevitably led to a broad range of stresses being included in the same smoothed and averaged environment type. Secondly, only stresses related to water were characterized here, meaning that limitation due to extreme temperatures or other environmental factors were only included in the environment classification through their effect on the water-deficit index, regardless of their additional effects on yield.

Given the high variability in rainfall patterns in the studied TPE, none of the environment types clearly dominated within any of the regions. Specific adaptation to a single environment type would thus be risky in such a TPE, unless rainfall could be forecast long enough before the sowing date and the environment type of the season predicted. In practice, farmers of this region target varieties that provide high yield (income) in good years, and sufficient yield to cover more than production costs in poor years. Broad adaptation to drought for this complex TPE thus appears as an appropriate strategy from a breeding viewpoint.

Environment characterization as a direct aid to selection

As frequently observed in dry-land wheat trials (Brennan et al., 1981; van Ginkel et al. 1998; Rebetzke et al., 2002; Mathews et al., 2002; Dreccer et al., 2007), the effect of GEI in the MET contributed a greater fraction of the variance for yield than did the direct genotypic effect (Table 3). To increase our understanding of GEI, climatic or soil variables, such as average rainfall, can be incorporated in the data analysis via mixed models or structural equation models (Malosetti et al., 2004; Vargas et al., 2007). However, these statistical approaches do not account for the interactions between the plants and their environment (e.g. the feedback of plant growth on soil water depletion). To resolve this issue, the environment characterization in this study integrated the water-deficit patterns that plants experience over time, and aimed to assist breeders to deal with large GEI, as detailed below. In addition, the analysis was performed here for breeding trials classified against the TPE, i.e. against a central reference independent from the specific scenarios of breeding trials.

Including the environment types into the statistical analysis assisted in interpretation of the GEI. Here, the G/GEI ratio in the MET (Table 3A) doubled when only considering the generally stress-free ET1 trials (Table 3B), meaning that a greater proportion of the genotypic effect was explained when considering the ET1 trials. In a previous study, when defining the environment types within a set of wheat trials (i.e. without consideration of the TPE), more than 30% of GEI for yield in these trials was associated with genotypic interactions across environment types (Chapman, 2008). For maize in the USA, Löffler et al. (2005) also demonstrated the benefit of environmental characterization. There, environment types were defined by mega-environments mainly based on temperature and photoperiod, and enabled up to 30% of the GEI variance observed in large-scale trials (18 hybrids tested in 90 environments) to be explained.

Knowledge about the main environment types of the TPE can help breeders choose trial locations and design experiments to focus on genetic variability in the environment type(s) of interest. In the studied TPE, ET1 was the least-frequent environment type (Figs 3, 4) but was associated with a high heritability (Table 3; Fig. 7). Furthermore, ET1 also had good phenotypic correlation with ET2 trials in the MET (Fig. 6B), meaning that relatively few ET1 trials (3–5) should suffice to select for adaptation to both of these environment types (at least for the genotypic diversity represented by the lines studied here). This finding quantitatively supports the argument of Cooper et al. (1997) who argued that including irrigated nurseries would improve the selection response for broad adaptation in this region, especially during commonly occurring years when most or all of the on-farm breeding trials would experience only drought conditions (Fig. 4A). To ensure broad adaptation of the genotypes, breeders could then attempt to select in ET3 environments by growing trials in places where ET3 environments are more common, for example, in the more northern and western areas, especially on shallow soils.

An extension of these uses of environment characterization is to weight genotype performance by the representativeness of their growing environment with respect to the TPE. Brennan and Sheppard (1985) proposed that environments could be weighted based on proximity to a hypothetical most frequently encountered environment. Analysis of breeding-system simulations has demonstrated advantages of such a weighted selection strategy in variable environments, especially when GEI are high (Cooper and Podlich, 1998;,Podlich et al., 1999).

Characterization of the environment types could also be useful to test specific physiological and breeding hypothesis, for example, concerning the identification of key adaptive traits. In the present study similarity , was observed in the genotype rankings in the relatively stress-free ET1 environments and in the stress-around-flowering ET2 environments (Fig. 6B). This phenotypic similarity in the genotype responses could be due to a lack of genetic diversity for drought-adaptive traits (as only 18 genotypes were studied here): traits associated with tolerance or avoidance of stress at flowering (e.g. limited potential transpiration rate; Condon et al., 2002; Sinclair et al., 2005) would be expected to be advantageous for yield in ET2, but also to introduce some trade-offs for yield performance in the more favourable ET1. Given the results for the genotypic diversity represented by the set of lines tested, seeking new germplasm with well-adapted traits for ET2 could be an alternative to improve yield in these environments.

Finally, knowledge about the main environment types is also valuable in breeding-simulation studies. In this case, crop models are employed to predict the phenotypes associated with the allelic combinations explored by the breeding programmes (Chapman et al., 2003; Hammer et al., 2005; Chenu et al., 2009; Messina et al., 2009) and are used to assist decision-making about breeding strategies with respect to relevant traits and genomic regions for broad or specific adaptation in the TPE.

Improving environment characterization to increase GEI understanding?

There are several possibilities for improving and broadening the applicability of this approach to characterizing crop growing environments.

Firstly, the environment types presented here only characterized water deficit patterns. Other major stresses affecting trials in the TPE (e.g. frost/heat around flowering) or management factors (e.g. different levels of fertilization) could be included. In order to integrate all the environmental factors (biotic and abiotic), environment characterizations based on crop yield have been proposed (Chauhan et al. 2008; Hernandez-Segundo et al., 2009). However, such a level of integration does not necessarily help in the improvement of specific factors, such as for drought tolerance.

Secondly, the MET characterization was performed for a single genotype (Hartog), and could be improved by running simulations for all genotypes studied in the MET, or for groups of these genotypes. Traits such as flowering date or tillering significantly influence the stress pattern, and can improve stress tolerance (via modification of plant leaf area and transpiration), or allow stress avoidance (via early flowering). Accounting for the genotypic variability for such traits in the simulations would allow the different stress patterns that genotypes experienced within trials to be characterized (i.e. for the same climate and soil) and would enhance our capacity to understand GEI (Hammer and Jordan, 2007).

Finally, increasing the number of environment types would allow us to explain better the variation in the stress patterns. However, for a given MET, this would inevitably reduce the number of trials that belong to each environment type, thus increasing the probability of insufficient data to analyse properly the genotypic variability within single environment type(s). In this study, three environment types seemed to be an accurate trade-off, as adding a fourth environment type only explained 3% more of the variability for the TPE water-deficit patterns (Fig. 2, inset).

Conclusion

Understanding a TPE in terms of major environmental conditions that cause genotype discrimination is important for both conventional and molecular selection approaches. Although only water-deficit patterns were characterized in this example for the wheat-growing area of north-eastern Australia, the grouping of MET based on this environmental classification helped to explain the basis for the observed GEI for yield. As discussed in this paper, knowledge of the spatial and temporal distribution of environment types can help to improve the predictability of genotype performance in the TPE, in the design of breeding programmes, and in phenotyping strategies. Furthermore, extending such environmental characterization to other drought-prone regions of the world, and thus facilitating robust comparison of environmental challenges in diverse regions, could aid the exchange of adapted germplasm among breeders.

With climate change, the 10–15 years necessary to develop a new wheat variety conventionally presents a new challenge for breeders. Modelling tools offer an opportunity to consider possible changes in the frequency of different stress patterns, and to prepare for these changes. Integrated modelling approaches used to characterize TPE under a future climate could help breeders, geneticists, and physiologists target genotypes, traits, and genetic regions of specific interest for predicted climates.

Thanks to Greg McLean, Al Doherty, Peter deVoil, and Nick Hansen for assistance in software programming, to Douglas Lush for his help in undertaking simulations, to Richard Routley, Ian Heiner, and Jo Owens for providing expertise on local soil and management practices, to Andries Potgieter and the Australian Bureau of Statistics in Brisbane for supplying maps and data concerning wheat production and planted area, and to Enli Wang for his assistance in the MET simulation. This research was partially funded by GRDC projects UQ113, UQ123, CSP00053, and CSP00132.

References

Annicchiarico
P
Perenzin
M
Adaptation patterns and definition of macro-environments for selection and recommendation of common wheat genotypes in Italy
Plant Breeding
 , 
1994
, vol. 
113
 (pg. 
197
-
205
)
Basford
KE
Cooper
M
Genotype×environment interactions and some considerations of their implications for wheat breeding in Australia
Australian Journal of Agricultural Research
 , 
1998
, vol. 
49
 (pg. 
153
-
174
)
Brennan
PS
Byth
DE
Drake
DW
Delacy
IH
Butler
DG
Determination of the location and number of test environments for a wheat cultivar evaluation programme
Australian Journal of Agricultural Research
 , 
1981
, vol. 
32
 (pg. 
189
-
201
)
Brennan
PS
Sheppard
JA
Retrospective assessment of environments in the determination of an objective strategy for the evaluation of the relative yield of wheat cultivars
Euphytica
 , 
1985
, vol. 
34
 (pg. 
397
-
408
)
Chapman
S
Cooper
M
Podlich
D
Hammer
G
Evaluating plant breeding strategies by simulating gene action and dryland environment effects
Agronomy Journal
 , 
2003
, vol. 
95
 (pg. 
99
-
113
)
Chapman
SC
Use of crop models to understand genotype by environment interactions for drought in real-world and simulated plant breeding trials
Euphytica
 , 
2008
, vol. 
161
 (pg. 
195
-
208
)
Chapman
SC
Cooper
M
Hammer
GL
Butler
DG
Genotype by environment interactions affecting grain sorghum. II. Frequencies of different seasonal patterns of drought stress are related to location effects on hybrid yields
Australian Journal of Agricultural Research
 , 
2000
, vol. 
51
 (pg. 
209
-
221
)
Chapman
SC
Hammer
GL
Butler
DG
Cooper
M
Genotype by environment interactions affecting grain sorghum. III. Temporal sequences and spatial patterns in the target population of environments
Australian Journal of Agricultural Research
 , 
2000
, vol. 
51
 (pg. 
223
-
233
)
Chapman
SC
Hammer
GL
Meinke
H
A sunflower simulation model. I. Model development
Agronomy Journal
 , 
1993
, vol. 
85
 (pg. 
725
-
735
)
Chauhan
Y
Wright
G
Rachaputi
N
McCosker
K
Identifying chickpea homoclimes using the APSIM chickpea model
Australian Journal of Agricultural Research
 , 
2008
, vol. 
59
 (pg. 
260
-
269
)
Chelle
M
Phylloclimate or the climate perceived by individual plant organs. What is it? How to model it? What for?
New Phytologist
 , 
2005
, vol. 
166
 (pg. 
781
-
790
)
Chenu
K
Chapman
SC
Tardieu
F
McLean
G
Welcker
C
Hammer
GL
Simulating the yield impacts of organ-level quantitative trait loci associated with drought response in maize: a ‘gene-to-phenotype’ modeling approach
Genetics
 , 
2009
, vol. 
183
 (pg. 
1507
-
1523
)
Chenu
K
Franck
N
Lecoeur
J
Simulations of virtual plants reveal a role for SERRATE in the response of leaf development to light in Arabidopsis thaliana
New Phytologist
 , 
2007
, vol. 
175
 (pg. 
472
-
481
)
Chenu
K
Rey
H
Dauzat
J
Lydie
G
Lecoeur
J
Estimation of light interception in research environments: a joint approach using directional light sensors and 3D virtual plants applied to sunflower (Helianthus annuus) and Arabidopsis thaliana in natural and artificial conditions
Functional Plant Biology
 , 
2008
, vol. 
35
 (pg. 
850
-
866
)
Comstock
RE
Quantitative genetics and the design of breeding programs
Proceedings of the international conference on quantitative genetics
 , 
1977
 
16–21 August 1976. Ames, USA. Iowa State University Press, 1705–1718
Condon
AG
Richards
RA
Rebetzke
GJ
Farquhar
GD
Improving intrinsic water-use efficiency and crop yield
Crop Science
 , 
2002
, vol. 
42
 (pg. 
122
-
131
)
Cooper
M
Byth
DE
Woodruff
DR
An investigation of the grain-yield adaptation of advanced CIMMYT wheat lines to water-stress environments in Queensland. 1. Crop physiological analysis
Australian Journal of Agricultural Research
 , 
1994
, vol. 
45
 (pg. 
965
-
984
)
Cooper
M
Byth
DE
Woodruff
DR
An investigation of the grain-yield adaptation of advanced CIMMYT wheat lines to water-stress environments in Queensland. 2. Classification analysis
Australian Journal of Agricultural Research
 , 
1994
, vol. 
45
 (pg. 
985
-
1002
)
Cooper
M
Delacy
IH
Relationships among analytical methods used to study genotypic variation and genotype-by-environment interaction in plant-breeding multienvironment experiments
Theoretical and Applied Genetics
 , 
1994
, vol. 
88
 (pg. 
561
-
572
)
Cooper
M
Hammer
GL
Plant adaptation and crop improvement
 , 
1996
CAB International, International, Rice Research Institute, and the International Crops Research Institute for the Semi-Arid Tropics, Wallingford, UK
Cooper
M
Stucker
RE
DeLacy
IH
Harch
BD
Wheat breeding nurseries, target environments, and indirect selection for grain yield
Crop Science
 , 
1997
, vol. 
37
 (pg. 
1168
-
1176
)
Cullis
BR
Thomson
FM
Fisher
JA
Gilmour
AR
Thompson
R
The analysis of the NSW wheat variety database. II. Variance component estimation
Theoretical and Applied Genetics
 , 
1996
, vol. 
92
 (pg. 
28
-
39
)
Daniells
I
Manning
B
Pearce
B
Profile descriptions: district guidelines for managing soils in north west NSW
NSW Agriculture
 , 
2002
Australia
GRDC, Natural Heritage Trust
Dreccer
MF
Borgognone
MG
Ogbonnaya
FC
Trethowan
RM
Winter
B
CIMMYT-selected derived synthetic bread wheats for rainfed environments: yield evaluation in Mexico and Australia
Field Crops Research
 , 
2007
, vol. 
100
 (pg. 
218
-
228
)
Dreccer
MF
Chapman
SC
Ogbonnaya
FC
Borgognone
MG
Trethowan
RM
Crop and environmental attributes underpinning genotype by environment interaction in synthetic-derived bread wheat evaluated in Mexico and Australia
Australian Journal of Agricultural Research
 , 
2008
, vol. 
59
 (pg. 
447
-
460
)
Fox
PN
Rosielle
AA
Reducing the influence of environmental main-effects on pattern analysis of plant breeding environments
Euphytica
 , 
1982
, vol. 
31
 (pg. 
645
-
656
)
Gilmour
AR
Cullis
BR
Verbyla
AP
Gleeson
AC
Accounting for natural and extraneous variation in the analysis of field experiments
Journal of Agricultural, Biological, and Environmental Statistics
 , 
1997
, vol. 
2
 (pg. 
269
-
293
)
Gilmour
AR
Thompson
R
Cullis
BR
Welham
SJ
ASReml estimates variance matrices from multivariate data using the animal model
Proceedings of the 7th world congress on genetics applied to livestock production, August 2002, Montpellier, France
 , 
2002
 
Session 28, 1–2
Hammer
GL
Chapman
S
van Oosterom
E
Podlich
DW
Trait physiology and crop modelling as a framework to link phenotypic complexity to underlying genetic systems
Australian Journal of Agricultural Research
 , 
2005
, vol. 
56
 (pg. 
947
-
960
)
Hammer
GL
Jordan
DR
Spiertz
JHJ
Struik
PC
van Laar
HH
An integrated systems approach to crop improvement
Scale and complexity in plant systems research: gene–plant–crop relations
 , 
2007
 
Wageningen UR–Frontis Series No. 21. Dordecht, The Netherlands: Springer, 45–61
Hammer
GL
Woodruff
DR
Robinson
RB
Effects of climatic variability and possible climatic change on reliability of wheat cropping: a modelling approach
Agricultural and Forest Meteorology
 , 
1987
, vol. 
41
 (pg. 
123
-
142
)
Hernandez-Segundo
E
Capettini
F
Trethowan
R
van Ginkel
M
Mejia
A
Carballo
A
Crossa
J
Vargas
M
Balbuena-Melgarejo
A
Mega-environment identification for barley based on twenty-seven years of global grain yield data
Crop Science
 , 
2009
, vol. 
49
 (pg. 
1705
-
1718
)
Hodson
DP
White
JW
Use of spatial analyses for global characterization of wheat-based production systems
Journal of Agricultural Science
 , 
2007
, vol. 
145
 (pg. 
115
-
125
)
Keating
BA
Carberry
PS
Hammer
GL
, et al.  . 
An overview of APSIM, a model designed for farming systems simulation
European Journal of Agronomy
 , 
2003
, vol. 
18
 (pg. 
267
-
288
)
Löffler
CM
Wei
J
Fast
T
Gogerty
J
Langton
S
Bergman
M
Merrill
B
Cooper
M
Classification of maize environments using crop simulation and geographic information systems
Crop Science
 , 
2005
, vol. 
45
 (pg. 
1708
-
1716
)
Malosetti
M
Voltas
J
Romagosa
I
Ullrich
SE
van Eeuwijk
FA
Mixed models including environmental covariables for studying QTL by environment interaction
Euphytica
 , 
2004
, vol. 
137
 (pg. 
139
-
145
)
Mathews
KL
Chapman
SC
Butler
D
Cooper
M
DeLacy
IH
Sheppard
J
Sahama
T
McComb
JA
Inter-annual changes in genotypic and genotype by environment variance components for different stages of the northern wheat improvement programme
Plant breeding for the 11th millennium proceedings 12th Australasian plant breeding conference
 , 
2002
Perth
Mathews
KL
Malosetti
M
Chapman
S
McIntyre
L
Reynolds
M
Shorter
R
van Eeuwijk
F
Multi-environment QTL mixed models for drought stress adaptation in wheat
Theoretical and Applied Genetics
 , 
2008
, vol. 
117
 (pg. 
1077
-
1091
)
Messina
C
Hammer
GL
Dong
Z
Podlich
D
Cooper
M
Sadras
VO
Calderini
D
Modelling crop improvement in a G×E×M framework via gene-trait-phenotype relationships
Crop physiology: applications for genetic improvement and agronomy
 , 
2009
The Netherlands
Academic Press
(pg. 
235
-
265
)
Muchow
RC
Cooper
M
Hammer
GL
Cooper
M
Hammer
GL
Characterizing environmental challenges using models
Plant adaptation and crop improvement
 , 
1996
 
CAB International, Wallingford, 349–364
Podlich
DW
Cooper
M
QU-GENE: a simulation platform for quantitative analysis of genetic models
Bioinformatics
 , 
1998
, vol. 
14
 (pg. 
632
-
653
)
Podlich
DW
Cooper
M
Basford
KE
Computer simulation of a selection strategy to accommodate genotype-by-environment interaction in a wheat recurrent selection programme
Plant Breeding
 , 
1999
, vol. 
118
 (pg. 
17
-
28
)
Pollak
LM
Corbett
JD
Using GIS datasets to classify maize-growing regions in Mexico and Central America
Agronomy Journal
 , 
1993
, vol. 
85
 (pg. 
1133
-
1139
)
Potgieter
AB
Hammer
GL
Butler
D
Spatial and temporal patterns in Australian wheat yield and their relationship with ENSO
Australian Journal of Agricultural Research
 , 
2002
, vol. 
53
 (pg. 
77
-
89
)
Qiao
CG
Basford
KE
DeLacy
IH
Cooper
M
Evaluation of experimental designs and spatial analyses in wheat breeding trials
Theoretical and Applied Genetics
 , 
2000
, vol. 
100
 (pg. 
9
-
16
)
Qiao
CG
Basford
KE
DeLacy
IH
Cooper
M
Advantage of single-trial models for response to selection in wheat breeding multi-environment trials
Theoretical and Applied Genetics
 , 
2004
, vol. 
108
 (pg. 
1256
-
1264
)
R Development Core Team
R: A language and environment for statistical computing
2008
 
Vienna, Austria: R Foundation for Statistical Computing
Rebetzke
GJ
Condon
AG
Richards
RA
Farquhar
GD
Selection for reduced carbon isotope discrimination increases aerial biomass and grain yield of rainfed bread wheat
Crop Science
 , 
2002
, vol. 
42
 (pg. 
739
-
745
)
Reymond
M
Muller
B
Tardieu
F
Dealing with the genotype×environment interaction via a modelling approach: a comparison of QTLs of maize leaf length or width with QTLs of model parameters
Journal of Experimental Botany
 , 
2004
, vol. 
55
 (pg. 
2461
-
2472
)
Runge
ECA
Effects of rainfall and temperature interactions during growing season on corn yield
Agronomy Journal
 , 
1968
, vol. 
60
 (pg. 
503
-
507
)
Sinclair
TR
Hammer
GL
van Oosterom
EJ
Potential yield and water-use efficiency benefits in sorghum from limited maximum transpiration rate
Functional Plant Biology
 , 
2005
, vol. 
32
 (pg. 
945
-
952
)
van Eeuwijk
FA
Malosetti
M
Yin
XY
Struik
PC
Stam
P
Statistical models for genotype by environment data: from conventional ANOVA models to eco-physiological QTL models
Australian Journal of Agricultural Research
 , 
2005
, vol. 
56
 (pg. 
883
-
894
)
van Ginkel
M
Calhoun
DS
Gebeyehu
G
Miranda
A
Tian-you
C
Pargas Lara
R
Trethowan
RM
Sayre
K
Crossa
J
Rajaram
S
Plant traits related to yield of wheat in early, late, or continuous drought conditions
Euphytica
 , 
1998
, vol. 
100
 (pg. 
109
-
121
)
Vargas
M
Crossa
J
Reynolds
MP
Dhungana
P
Eskridge
KM
Structural equation modelling for studying genotype×environment interactions of physiological traits affecting yield in wheat
Journal of Agricultural Science
 , 
2007
, vol. 
145
 (pg. 
151
-
161
)
Voltas
J
van Eeuwijk
FA
Sombrero
A
Lafarga
A
Igartua
E
Romagosa
I
Integrating statistical and ecophysiological analyses of genotype by environment interaction for grain filling of barley. I. Individual grain weight
Field Crops Research
 , 
1999
, vol. 
62
 (pg. 
63
-
74
)
Wang
E
Robertson
MJ
Hammer
GL
Carberry
PS
Holzworth
D
Meinke
H
Chapman
SC
Hargreaves
JNG
Huth
NI
McLean
G
Development of a generic crop model template in the cropping system model APSIM
European Journal of Agronomy
 , 
2002
, vol. 
18
 (pg. 
121
-
140
)
Yan
W
Hunt
LA
Sheng
Q
Szlavnics
Z
Cultivar evaluation and mega-environment investigation based on the GGE biplot
Crop Science
 , 
2000
, vol. 
40
 (pg. 
597
-
605
)

Author notes

Present address: Dupont Pioneer-HiBred, Johnston, USA.
Present address: The University of Sydney, Sydney, Australia.

Comments

0 Comments