- Email: [email protected]
Commercially available wheat cultivars are broadly adapted to location and time of sowing in Australia’s grain zone
Europ. J. Agronomy 77 (2016) 38–46
Contents lists available at ScienceDirect
European Journal of Agronomy journal homepage: www.elsevier.com/locate/eja
Commercially available wheat cultivars are broadly adapted to location and time of sowing in Australia’s grain zone R.A. Lawes a,∗ , N.D. Huth b , Z. Hochman c a b c
CSIRO Agriculture, 147 Underwood Avenue, Floreat, WA 6014, Australia CSIRO Agriculture, PO Box 2583, Brisbane, Qld 4001, Australia CSIRO Agriculture, 203 Tor St, Toowoomba, Qld 4350, Australia
a r t i c l e
i n f o
Article history: Received 9 October 2015 Received in revised form 8 February 2016 Accepted 29 March 2016 Available online 9 April 2016 Keywords: G×E×M Wheat Broad adaptation
a b s t r a c t Farmers must choose which cultivar to grow based on the phenology of the cultivar and anticipated season length. The current study investigated the established doctrine of sowing fast maturing cultivars late, and slow maturing cultivars early. This was explored by quantifying the genotype (G) × environment (E) × management (M) available to farmers using commercially released cultivars, where management relates to the time of sowing. Nineteen cultivars of spring wheat (Triticum aestivum) were sown at 3 times of sowing (early, conventional and late) at 13 sites in 2011 and 2012. Sites were located throughout the Australian grain growing region in Queensland, New South Wales, Victoria, South Australia and Western Australia from latitudes 27◦ 34 S to 35◦ 09 S where annual rainfall ranged from 237 mm to 747 mm. In general, the three way interaction between G, E and M for yield was small and cultivar could not overcome the yield penalty associated with a late time of sowing. At 11 of the 13 sites, fast to moderately fast maturing cultivars sown early generated the highest yields. Fast maturing cultivars sown late could not compensate for a late time of sowing. Commercial cultivars were broadly adapted to environment and management, and with these cultivars, the Australian grain growing region could be split into just two environments, south and north. Even then, season appears to be the main arbiter of environment, rather than location per se as individual sites moved from one group to the other, depending on season. There was no evidence to suggest farmers could exploit a cultivar by management interaction for time of sowing with commercial cultivars, as the outcome of the season is unpredictable, and with current technology farmers should simply choose the best performing cultivar for their region and sow it as early as possible. Crown Copyright © 2016 Published by Elsevier B.V. All rights reserved.
1. Introduction The goal of plant breeders is to produce new cultivars that are better adapted to a targeted production environment than their predecessors and to release these cultivars for commercial use by farmers. The farmer must then choose a cultivar for their particular environment and sow it at a particular time. The time of sowing immediately introduces a management component into the choice of cultivar. These objectives, enunciated by Comstock (1977) and discussed by Chapman (2008) highlight the importance of breeding cultivars for a particular environment, and the response of the cultivar in a particular environment can be further influenced by management. Therefore the farmer and the breeder must consider genotype (G), environment (E), management (M) and the interactions of those components.
∗ Corresponding author. E-mail address: [email protected] (R.A. Lawes). http://dx.doi.org/10.1016/j.eja.2016.03.009 1161-0301/Crown Copyright © 2016 Published by Elsevier B.V. All rights reserved.
Agronomists and plant breeders are both conscious of environment, but the two disciplines have evolved differently. For an agronomist, the environment has historically been defined in terms of the stresses applied to a plant through its life. These may include water stress (Nix and Fitzpatrick, 1969), nutrient stress, and temperature stress (Asseng et al., 2011). The importance of these stresses are summarised by frameworks such as those described by (Passioura and Angus, 2010), who highlight the importance of matching the phasic development of the plant to the water supply and temperature stresses of a particular region. For breeders, the definition of environment may be defined statistically, where cultivars will be grown across a wide range of environments (Cooper and Delacy, 1994). The variance components from an analysis of variance may be employed to determine the extent of variation in a trait that can be attributed to cultivar, the environment and the cultivar by environment interaction. Since large numbers of cultivars are often screened, statistical techniques have evolved to group cultivars into those that may be broadly adapted to an environment or specifically adapted to
R.A. Lawes et al. / Europ. J. Agronomy 77 (2016) 38–46
39
Table 1 Cultivars and the number of plots of each cultivar sown at each site. Maturity
Fast
Fast–Moderate
Moderate
Slow
a
State
Queensland
New South Wales
Victoria
South Australia
Location
Bungunya
Nangwee
Spring Ridge
Temora
Walpeup
Minnipa
Turretfield
Corrigin
Year
2012
2012
2011
2011
2011
2011
2012
2011
2012
2011
2012
2011
2012
3
9
9
9
9
9 9
9
9
9
Axe Westonia Lincoln Livingston Wyalkatchem Crusader Carnamah Gladius Yitpi Scout Janz Derrimut Correll Magenta Lang Gregory Wylie Endure Bolac
9
9
9
Western Australia Eradu
6 9 3
6
6
6
6
9
9
9
9
9
9 6
9
6 3 3
6 3
6 3 3
6 3
9
9
9
6
3
3
3
6
9 6
9 6
6
6
6
6
6
6
6 6
6
6 6
6
3 6
6
3 6
9 6 3
9 3
9 3
9 6
9 6
9
9
9
9
3
6
6
3
3
3
3
3
3
6
6
6
6
9 9
8a 9
9 9
9
9
9
9
9
Missing plot TOS 1.
an environment. Therefore, concepts such as the regression on the environment mean (Finlay and Wilkinson, 1963), and various multivariate techniques have been employed to detect G × E interactions (Cooper and Delacy, 1994). The analysis of multienvironment trials has evolved and is now quite flexible (Smith et al., 2005). For example, the nuances of individual trials can be accommodated where some sites may have strong spatial processes that need to be statistically modelled. Assumptions about the covariance structure of cultivars and environment at particular locations can also be explored using either compound symmetry, diagonal or factor analytic models (Smith et al., 2001). On occasions, environmental covariates such as rainfall can be introduced into such a model to further define environment (Eagles et al., 2010). The attempt to define cultivar performance for specific environments is partly driven by the need to communicate to farmers which cultivars will be suited to their farm. Furthermore, seasonal conditions and logistics may influence management decisions such as the time of sowing, where conventional wisdom suggests longer season cultivars should be sown early, and shorter season cultivars should be sown later. Some fast maturing cultivars are adapted to regions where the cold and heat stresses occur in close temporal proximity to each other. Slower maturing cultivars have been developed for environments with higher rainfall, where there is an opportunity for the plant to remain in the vegetative stage longer, produce more biomass, produce a higher number of grains, spend longer during grain fill and produce grains of equivalent kernel weight to the shorter season cultivar (Richards et al., 2014). Therefore the farmer and the plant breeder are trying to grow a crop where the timing and duration of vegetative and reproductive growth are optimised to use all the available resources on offer and avoid temperature stress. Unfortunately, the water deficit patterns and extent of the stresses all vary with season and location (Chenu et al., 2011), and this complicates the development of cultivars for niche environments and management actions. Since the timing of stress on a crop can be influenced by phenology (cultivar), the location and season (environment) and the time of sowing (management), the interaction between G × E × M may be important to farmers. The three way interactions are increasingly being examined for factors such as tillage practice (Rebetzke et al., 2014), organic and conventional systems (Kamran et al., 2014)
and weed management (Lemerle et al., 2001). Many of these earlier studies evaluated genetic material that had yet to be released to the industry, and these earlier studies were all attempting to identify whether G × E × M existed and should become part of the breeding process. Our goal here is to evaluate the G × E × M concept as it applies to farmers with regard to time of sowing using current, commercially available cultivars of varying maturity. To that end, the performance of 19 elite commercially available wheat cultivars of varying phenology sown at 3 different times in 13 locations across the entire Australian grain growing region were evaluated. Trials were located in Western Australia, South Australia, Victoria, New South Wales and Queensland. These data are used to determine whether farmers should change cultivars, given the time of sowing, where long season cultivars should be sown early and shorter season cultivars sown later. The current study quantifies the extent of the genetic variation available to farmers, and also considers whether niche environments exist across the Australian continent that farmers can realistically exploit with currently available genetic material. 2. Materials and methods Nineteen cultivars of spring wheat (Triticum aestivum) were chosen and selected in consultation with plant breeders to ensure cultivars were suited to a particular location, thus by definition, not all cultivars were grown at each location. Cultivars with maturities varying from early through to late were grown at each location (Table 1). At every location cultivars were sown at three times of sowing (TOS), early (late April to early May; TOS 1), conventional time of sowing (mid May to early June; TOS 2) and late (late June to early July; TOS 3). Specific times of sowing at each location are presented in Table 2. Sites were managed so that nutrients were non-limiting. Fertiliser was applied at sowing using formulations of S, K, and trace elements appropriate to the region and the soil type. If the season was favourable and soil fertility was considered limiting, additional N was applied to ensure the crops could achieve their highest possible yield potential as assessed in August. At each site total N applied is presented in Table 3. The sites had various soil types ranging from black vertosols of heavy texture at Nangwee in Queens-
40
R.A. Lawes et al. / Europ. J. Agronomy 77 (2016) 38–46
Table 2 Time of sowing for each site. Site
Time of sowing 1
Time of sowing 2
Time of sowing 3
Bungunya Nangwee Spring Ridge Temora Walpeup Turretfield Turretfield Minnipa Minnipa Corrigin Corrgin Eradu Eradu
10/05/2012 17/05/2012 09/05/2011 05/05/2011 29/04/2011 30/05/2012 18/05/2011 25/05/2012 13/05/2011 02/05/2012 30/04/2011 02/5/2012 29/04/2011
22/05/2012 31/05/2012 06/06/2011 26/05/2011 31/05/2011 15/06/2012 08/06/2011 08/06/2012 27/05/2011 21/05/2012 24/05/2011 21/05/2012 24/05/2011
23/06/2012 23/06/2012 23/06/2011 20/06/2011 01/07/2011 05/07/2012 28/06/2011 25/06/2012 24/06/2011 21/06/2012 21/06/2011 21/06/2012 23/06/2011
Table 3 Total N applied at each site, starting soil nitrate and initial plant available water (PAW) at sowing. Site
N applied (kg/ha)
Initial PAW (mm)
Initial NO3 (kg/ha)
Bungunya 2012 Nangwee 2012 Spring Ridge 2011 Temora 2011 Walpeup 2011 Minnipa 2011 Minnipa 2012a Turretfield 2011 Turretfield 2012 Corrigin 2011 Corrigin 2012 Eradu 2011 Eradu 2012
9.0 9.0 100.8 57.0 31.2 10.8 60.5 28.8 28.8 56.0 56.0 61.0 58.0
38.3 186.8 167.1 135.3 (to 150 cm) 104.4 69.9 35.3 43.4 65.2 23.2 (to 85 cm) 0b 42.5 0b
63.9 608.7 183.5 130.4 (to 150 cm) 256.1 89.3 49.5 205.6 13.2 50.6 (to 60 cm) 11 (to 30 cm) 113.0 17.7 (to 30 cm)
a b
TOS 1 and 2 had 60.5 kg N/ha applied, TOS 3 had 35.6 kg N/ha applied. Initial soil moisture not measured, but soils were dry and plant available water was assumed to be 0.
land to the deep yellow-brown loamy sand at Eradu in Western Australia. Some soils were acidic (Eradu, Corrigin), while others were alkaline (Turretfield, Walpeup). The plant available soil water holding capacities ranged from 77 mm at Turretfield to 246 mm at Nangwee. Rainfall patterns also varied considerably, with 8 sites (Bungunya, Eradu 2011, Minnipa 2011, Nangwee, Temora 2011, Turretfield 2012, Walpeup 2011 and Spring Ridge 2011) receiving more than 100 mm outside the traditional April to October rainfall. Growing season rainfall (April to October) ranged from just 136 mm at Minnipa in 2012 to 340 mm in Corrigin in 2011 while annual rainfall ranged from 281 mm at Eradu in 2012 to 747 mm at Temora in 2011 (Table 4). These variations reflect the diversity of the environments expressed across the Australian wheatbelt, but are not necessarily a representative sample of every environment in every season. At each site hand harvests from 4 quadrats (0.25 m2 ) were taken and the yield components of grain number, grain weight and biomass at harvest were measured. From these components, the harvest index was calculated. Grain yield was estimated by hand harvest of the plots. Phenology was monitored for each cultivar to determine the date of flowering. Flowering date (Zadoks growth stage 65; (Zadoks et al., 1974)) was estimated by eye for each cultivar at each time of sowing from multiple sequential observations during, before and after anthesis.
2.1. Design and analysis The experimental design is an unbalanced incomplete split plot block design, where there are two blocks (replicates) in each trial. The main plots are sowing times (early, normal and late) and the split plots are cultivars (7–9) which are incompletely replicated within and across sowing times. That is, at a trial a cultivar may be
replicated 3 times at each time of sowing (9 plots), twice at each time of sowing (6 plots) or once at each time of sowing (3 plots), i.e. within a block, a cultivar at a particular TOS may be replicated zero, once or twice. Three cultivars, Janz, Gregory and Gladius were sown in all trials following the principals of a p-rep design (Williams et al., 2011). The trial was orientated as a single long row and the dimensions were 1 row and 60 columns. With this layout, modelling the spatial variability along the row may be important. The design was then replicated across 13 sites, where different cultivars were used at each site depending on their geographical location (Table 4).
2.2. Statistical analysis Data were analysed using ASREML, a linear mixed effects model, in the R statistical package (R Core Team, 2013). Initially, each trial was analysed as a separate entity, to quantify the site level genetic variance, ascertain whether there was a spatial process that needed to be modelled separately at that particular site and determine whether the genetic variance at a particular site was bounded. For each site, the yield variance components were estimated. Time of sowing (the environment), was fixed, as is often recommended (Smith et al., 2005). The design attributes of replicate and time of sowing within replicate were fit as random effects. Cultivar and the interaction between cultivar and time of sowing were also fit as random effects. An AR1 autoregressive covariance structure was evaluated by AIC and BIC at each location (Table 5). If the AR1 covariance structure that accounted for spatial processes improved the model fit at that site, it was retained. Following these investigations, a Multi Environment Trial (MET) analysis was conducted, excluding those trials with a bounded genetic variance. The approach adopted followed the methods outlined by Smith et al. (2001, 2015). Two models were fitted to explore
R.A. Lawes et al. / Europ. J. Agronomy 77 (2016) 38–46
41
Table 4 Site description of soil type, location, altitude and climate. Site and year
Soil type
Latitude, Longitude
Altitude m asl
Annual Rainfall (January–December)
Growing Season Rainfall (Time of Sowing 1 to Harvest)
Plant Available Water Holding Capacity (mm)
Bungunya 2012
Grey Vertosol
28◦ 25 S, 149◦ 39 E
189
439
137
130
Corrigin 2011
Loamy Duplex
32◦ 22 S, 117◦ 55 E
313
406
340
79
32◦ 22 S, 117◦ 55 E
313
238
170
79
28 40 S, 115 10 E
266
494
326
96
28◦ 40 S, 115◦ 10 E
266
281
245
96
32◦ 49 S, 135◦ 09 E
196
391
221
139
32 49 S, 135 09 E
196
237
136
139
Corrigin 2012 Eradu 2011
Deep Yellowish-Brown Loamy Sand
Eradu 2012 Minnipa 2011
Sandy Clay Loam
◦
◦
Minnipa 2012
◦
◦
Nangwee 2012
Black Vertosol
27◦ 34 S, 151◦ 19 E
362
480
169
246
Temora 2011
Red Chromosol
34◦ 24 S, 147◦ 31 E
274
747
294
147
34◦ 32 S, 138◦ 47 E
129
528
310
77
34 32 E, 138 47 S
129
382
243
77
35◦ 09 E, 142◦ 01 S
77
448
152
105
31.19 E, 150◦ 11 S
325
596
266
183
Turretfield 2011 Calcareous Loam
◦
Turretfield 2012 Walpeup 2011
Loamy Sand
Spring Ridge 2011Grey-black Vertosol
◦
Table 5 AIC for standard and AR1 models for yield and variance components for each trait at each site for cultivar and the interaction between cultivar and time of sowing. Site
AIC AIC AR 1 Cultivar yield
Cultivar × TOS
Cultivar biomass
Cultivar × TOS
Cultivar grain number
Cultivar × TOS
Cultivar Grain weight
Cultivar × TOS
Bungunya 2012 Corrigin 2011 Corrigin 2012 Eradu 2011 Eradu 2012 Minnipa 2011 Minnipa 2012* Nangwee 2012 SpringRidge 2011 Temora 2011 Turretfield 2011 Turretfield 2012 Walpeup 2011
712 681 686 675 603 667 631 701 716 701 685 685 645
9.2% 0.0% 0.0% 11.3% 6.4% 0.0% 1.0% 0.0% 0.0% 0.0% 18.5% 0.0% 65.6%
13.6% 0.0% 4.9% 36.3% 0.0% 14.7% 31.3% 0.0% 46.8% 23.0% 1.1% 0.0% 7.9%
9.6% 0.0% 0.0% 5.2% 0.0% 0.0% 0.6% 0.0% 0.0% 0.0% 0.0% 0.0% 44.3%
14.9% 1.4% 1.4% 8.1% 0.0% 47.6% 21.4% 6.0% 35.2% 35.9% 10.4% 10.2% 10.5%
5.5% 0.0% 3.4% 34.8% 5.5% 0.0% 0.0% 0.0% 0.2% 0.0% 48.9% 0.0% 58.7%
68.9% 36.4% 22.5% 57.4% 61.8% 81.5% 35.3% 67.9% 56.0% 56.6% 35.0% 28.5% 23.7%
14.1% 17.9% 1.2% 30.9% 27.2% 6.2% 5.2% 9.1% 20.6% 0.0% 38.9% 0.0% 42.8%
714 683 672 677 605 666 633 702 715 701 683 687 647
11.5% 15.4% 5.8% 53.0% 11.2% 44.5% 34.9% 0.0% 47.0% 23.2% 0.0% 0.0% 0.0%
the site by cultivar by time of sowing interactions. The models chosen were a diagonal model structure (Diag), and a Factor Analytic structure (FA). In both cases the cultivar was not explicitly fitted. With the diagonal structure, trials were assumed independent with heterogeneous variances. With the FA structure the trial variance heterogeneity and between trial correlation was accommodated through a hypothetical factor (k), which in this case was 1 (Smith et al., 2001). Again AIC was used to select the best model out of the diagonal and FA alternative. Site and TOS were fixed and represent the environments. The random model included the interaction between site and replicate; the 3 way interaction between site, replicate and time of sowing; the interaction between site and cultivar; the interaction between site and time of sowing and the 3 way interaction between site, cultivar and TOS. In the random model, site was assumed to have a diagonal covariance structure, where the covariance between sites was assumed to be 0. An AR1 covariance structure was used for 3 sites (Corrigin 2012, Spring Ridge 2011 and Turretfield 2011), where a spatial process had previously been identified. The output
generated from the diagonal model was equivalent to analysing each trial separately. For the FA model, an FA structure where k = 1 was employed for the site by cultivar interaction. All other terms were fitted as described with the diagonal model. Predicted values for all site by TOS by cultivar combinations were then generated for yield (t/ha), biomass (t/ha), the square root of grain number and 1000 grain weight (mg), using the predict command in ASREML. The predicted values for yield were plotted against the day of flowering for cultivar at each location across all times of sowing. Similarly, the relationship between grain number, grain weight and biomass were all considered for each cultivar. The predicted values were subjected to an hierarchical cluster analysis where the columns were hand harvested yield, grain number, grain weight biomass and the rows were the 700 combinations of site, cultivar and time of sowing. Data were then standardised for each variable (column). Agglomeration was performed using the Manhattan metric. The objective here was to explore whether the accumulation of yield through the various yield components varied across the Aus-
42
R.A. Lawes et al. / Europ. J. Agronomy 77 (2016) 38–46
Fig. 1. The predicted values for grain weight (mg) and grain number for each cultivar across all sites and times of sowing.
tralian wheat belt and whether groups formed around locations, times of sowing or particular cultivars.
3. Results Yields varied considerably between sites, TOS and cultivar. Low yielding sites, where the best performing cultivar yielded less than 2.5 t/ha were at Corrigin in 2012 (2.1 t/ha), Eradu in 2012 (1.0 t/ha) and Minnipa in 2012 (2.3 t/ha). Moderate yields were generated by the best performing cultivar at Minnipa in 2011 (4.1 t/ha), Eradu in 2011(3.1 t/ha) and Bungunya in 2012 (3.3 t/ha). High yields, greater than 4 t/ha, were produced at Spring Ridge in 2011 (5.6 t/ha), Temora (4.4 t/ha) and Corrigin in 2011 (4.4 t/ha). Within a site, the expected decline in yield from the best performing cultivar at time of sowing 1 to time of sowing 2 and 3 varied between locations. From time of sowing 1 to time of sowing 3 yields typically declined by between 0.8 and 1.3 t/ha from Corrigin (2011, 2012), Minnipa (2012), Eradu (2012), Bungunya (2012). More moderate declines in yield occurred at Eradu in 2012 (0.2 t/ha), and Temora in 2011 (0.7 t/ha) and the range of yields at each site and time of sowing are presented in Table 6. In Spring Ridge, yields increased by 1.5 t/ha from time of sowing 1 to time of sowing 3. Similarly the yield decline between time of sowing 1 and time of sowing 2 ranged from 0.1 t/ha at Temora to 0.5 t/ha in Minnipa in 2012. At all other sites yields declined by between 0.2 and 0.4 t/ha from time of sowing 1 to time of sowing 2. The exception was the increase of 1.6 t/ha from time of sowing 1 to time of sowing 2 at Spring Ridge. Therefore, across sites and times of sowing, yields ranged from 0.7 t/ha to 5.6 t/ha, and at all sites except Spring Ridge, delayed time of sowing generally reduced yield (Table 6). Cultivar and the interaction between cultivar and time of sowing at the site level were comparatively small for grain yield, with some exceptions. At Bungunya, Corrigin in 2011 and 2012 and Eradu in 2012 the genetic variance was approximately 10% and the genotype by management component was approximately 9% or less (Table 5). The genetic variance at Eradu, Minnipa, Spring Ridge and Temora were all 23% or greater. However, providing the main cultivar effect was unbounded, the genotype by environment component for yield was greater than 5% at every location except Minnipa in 2012 (Table 5). Therefore, before the meta-analysis of data was conducted, the
Fig. 2. The predicted values for grain weight (mg) and grain number for each site across all cultivars and times of sowing.
G × M that can be exploited using commercial cultivars was small. Within a site the variance component for cultivar and the variance component for the interaction between cultivar and time of sowing for biomass and grain number were commensurate with the equivalent variance components for yields at most sites (Table 5). One exception occurred at Minnipa in 2011, where the variance component for cultivar yield was 44%, but the variance component for cultivar biomass was just 14%. The variance components for cultivar and the interaction between cultivar and TOS for grain weight were often larger than the corresponding variance components for the other 3 traits. The outputs from the meta-analysis confirmed the findings from the initial analysis of data at the individual sites with regard to the small amount of genetic variation and even smaller amount of genotype by management interaction within a site. To that end, the best performing genotype did vary between sites, but rarely varied within a site between TOS 1, TOS 2 and TOS 3 (Table 7). For example, Wyalkatchem, an early to mid season cultivar was the best performing, or equal best performing at Corrigin in 2011 and 2012, Eradu in 2011 and 2012 and Minnipa in 2011. Wyalkatchem was bested by the faster maturing Axe at Minnipa at 2012. However across these southern and western sites, Wyalkatchem dominated at all times of sowing. While Crusader was identified as the highest ranked cultivar at TOS 3 at Eradu, its performance at that time of sowing was not significantly different from Wyalkatchem. At the eastern sites, in Temora, Bungunya and Spring Ridge, the cultivar Gregory was the best performing or equal best performing, regardless of time of sowing. Gregory was characterised as a longer season cultivar, which should theoretically yield less than faster maturing cultivars when sown late. This did not occur at any of these locations. 3.1. Relationships between traits Overall, across sites and cultivars there were strong correlations between grain yield biomass (r = 0.94) and grain number (r = 0.94). The relationship between grain yield and grain weight was comparatively poor (r = 0.58).
R.A. Lawes et al. / Europ. J. Agronomy 77 (2016) 38–46
43
Table 6 The range of yields (kg/ha) for each site and corresponding time of sowing. TOS 1 Range Bungunya 2012 Corrigin 2011 Corrigin 2012 Eradu 2011 Eradu 2012 Minnipa 2011 Minnipa 2012 Spring Ridge 2011 Temora 2011
3034 3852 1602 1974 759 3357 1878 3055 3626
TOS 2 Range 3504 4494 2305 3516 1043 4080 2361 4104 4418
2757 3550 1299 1818 740 3219 1380 4656 3377
TOS 3 Range 3141 4191 2002 3160 1024 3942 1853 5705 4169
2170 3002 751 729 540 2163 1319 4543 2731
2622 3643 1454 1917 824 2885 1791 5592 3523
Table 7 Best performing cultivar with the predicted yield and standard error from the Multi Environment Trial Factor Analytic model.
Corrigin 2011 Corrigin 2012 Eradu 2011 Eradu 2012 Minnipa 2011 Minnipa 2012 Temora 2011 SpringRidge 2011 Bungunya 2012 a
Time of Sowing 1
Time of Sowing 2
Time of Sowing 3
4.5 ± 0.1 Wyalkatchem Crusader 2.3 ± 0.3 Wyalkatchem 3.5 ± 0.2 Wyalkatchem 1.0 ± 0.4 Wyalkatchem Crusader Axe 4.1 ± 0.2 Wyalkatchem Crusader Axe 2.4 ± 0.1 Axe 4.4 ± 0.2 Gregory 4.1 ± 0.2 Gregory Janz 3.5 ± 0.16 Gregory
4.2 ± 0.1 Wyalkatchem Crusader 2.0 ± 0.3 Wyalkatchem 3.5 ± 0.2 Wyalkatchem 1.0 ± 0.4 Wyalkatchem Crusader Axe 3.9 ± 0.2 Wyalkatchem Crusader 1.9 ± 0.1, Axe 4.2 ± 0.2 Gregory 5.7 ± 0.2 Gregory Janz 3.1 ± 0.16 Gregory
3.6 ± 0.1 Wyalkatchem Crusader 1.5 ± 0.3 Wyalkatchem 1.9 ± 0.3 Crusadera 0.8 ± 0.4 Wyalkatchem Crusader Axe 2.9 ± 0.2 Wyalkatchem Crusader 1.8 ± 0.1 Axe 3.5 ± 0.2 Gregory 5.6 ± 0.2 Gregory Janz 2.6 ± 0.16 Gregory
Crusader was not grown at the site.
Fig. 3. The predicted values for biomass and grain number for each cultivar across all sites and times of sowing.
Plots for the predicted values of grain number vs grain weight for each cultivar across all sites and times of sowing demonstrate that some cultivars, such as Axe, Wyalkatchem and Carnamah have grain weights that remain relatively stable across a wide range of grain numbers (Fig. 1). Conversely, grain weight varies more for cultivars such as Crusader, Scout, Janz, Lang, and Bolac across a wide range of grain numbers (Fig. 1). When the predicted values of grain weight and grain number for each site are examined across all cultivars and times of sowing, the driver of yield varies (Fig. 2). At Eradu in 2012, yields were heavily influenced by grain weight within a time of sowing. At Bungunya and Corrigin (both years), both traits vary. For sites such as Spring Ridge and Minnipa in 2011, the ability to assemble large numbers of grains appeared to be important as there was comparatively little variation in grain weight. Therefore the “best” mechanism for assembling yield varied with site and season. The linear relationship between biomass and grain number did not vary between cultivars (Fig. 3), where 1 kg/ha of biomass produced 11.3 grains/ha. However, the slope and intercept varied between cultivars for the relationship between biomass and grain weight, where Crusader, Lang, Wylie, Endure and Bolac all produced heavier grains per unit of biomass, but with lower intercepts (Fig. 4). One interpretation of this finding is that biomass influences the determination of grain number more than grain weight and grain weight is determined by other factors. 3.2. Does phenology influence yield?
The best linear unbiased estimates of grain number for time of sowing showed a decline of just 63 and 327 grains/ha from TOS 1 to TOS 2 and TOS 3, which represented a change of just 0.3% from the intercept. In contrast, grain weights declined by 1.5 and 3.6 mg from TOS 1 to TOS 2 and TOS 3 respectively. This decline for TOS 3 represented a reduction of 9.3%. The variance components for cultivar and the interaction between cultivar and time of sowing for grain weight (Table 5), were often larger than those for grain number. One interpretation of these data is that grain number is determined earlier in the season and does not vary greatly with TOS within a site, and cultivars then assemble yield differently and express this difference through grain weight more than grain number.
Flowering dates varied between cultivars, dates of sowing and sites. The cultivar Axe was earlier than every other cultivar, while Bolac flowered later than every other cultivar. In general, Axe, Westonia, Wyalkatchem and Crusader flowered earlier than cultivars such as Lang, Gregory, Wylie, Endure and Bolac. Cultivars including Gladius, Yitpi, Scout, Correll and Magenta were of intermediate flowering time. There was a negative relationship between flowering date and yield, where on average, yield decline by 39 ± 7.7 kg/ha per day after the optimum. This rate of yield decline occurred at Bungunya, Corrigin (both years), Eradu in 2011, Minnipa (both years) and Temora. The decline in yield was less at Eradu in 2012 where
44
R.A. Lawes et al. / Europ. J. Agronomy 77 (2016) 38–46
Fig. 4. The predicted values for biomass and grain weight (mg) for each cultivar across all sites and times of sowing.
Fig. 5. Grain yield vs the day of flowering for the 9 sites and 3 times of sowing.
3.3. Number of distinct environments in Australia yields declined by just 9 kg/ha/day and at Spring Ridge, where yields increased with delayed flowering by 73 kg/ha/day (Fig. 5). While linear representations define a trend over a 30 day flowering window, at smaller temporal scales (∼10 days) there was considerable variation between the cultivars where a slightly (2–5 day) later maturing cultivar out yielded the earlier cultivar. This difference may be due to other attributes of the cultivar (Fig. 5). Regardless, at 6 of the 9 sites sowing early and flowering early produced higher yields.
Cultivars, sites and times of sowing separated into just two groups when the dendrogram (not shown) was cut at a height of 200 and these groups could be loosely described as high yielding (group 1) with 249 entities and low yielding (group 2) with 264 entities. Group 1 had a median yield of 3.8 t/ha, median biomass of 8.4 t/ha, median grain number of 95,000/ha and a median grain weight of 40 mg. Conversely, group 2 had a median yield of 1.7 t/ha, median biomass of 4.7 t/ha, median grain number of 55,000 g/ha,
Fig. 6. Boxplots of traits for the two groups identified by the hierarchical cluster analysis for a) yield, b) biomass, c) grain number and d) grain weight.
R.A. Lawes et al. / Europ. J. Agronomy 77 (2016) 38–46
45
Table 8 Group membership for sites, times of sowing and individual cultivars.
Bungunya Corrgin 2011 Corrigin 2012 Eradu 2011 Eradu 2012 Minnipa 2011 Minnipa 2012 Spring Ridge 2011 Temora
Time of Sowing 1
Time of Sowing 2
Time of Sowing 3
1 1 2 2 (Axe, Westonia, Wyalkatchem, 1) 2 1 2 1 1
1 (Lang 2) 1 2 2 (Axe, Wyalkatchem 1) 2 1 2 1 1
2 1 2 2 2 2 2 1 1 (Crusader Lang 2)
and median grain weight of 31 mg (Fig. 6a–d). At Corrigin in 2011 and 2012, Eradu in 2012 and Spring Ridge in 2011, neither the time of sowing or cultivar had any impact on group membership. At Minnipa in 2011, the first two times of sowing separated into the first group, and the third time of sowing separated into the second. There were two instances, at Bungunya at time of sowing 2, and Temora at time of sowing 3, where cultivar alone moved from the high yielding group to the low yielding group. These cultivars were Lang at Bungunya and Crusader and Lang at Temora (Table 8). Finally, at Eradu in 2011 the cultivars Axe, Wyalkatchem and Westonia moved from group 2 to group 1 for time of sowing 1. At time of sowing 2 at Eradu in 2011 Axe and Wyalkatchem also moved from group 2 to group 1. Importantly, sites could move from group 1 to group 2 depending on season. 4. Discussion The selected sites provided a wide range of environments where the site mean yields ranged from 0.8 to 4.5 t/ha. The range of environments was further extended by varying the time of sowing. This effect alone varied from 2.41 t/ha to 3.27 t/ha and the decline in yield with time of sowing supports agronomic results found elsewhere (Sharma et al., 2008; Fletcher et al., 2015). This wide range of environments was established to test whether farmers should alter their cultivar choice based on time of sowing, given the phenology of the cultivar. The analysis of 19 commercially available cultivars with varying phenology grown across 13 sites with 3 times of sowing demonstrated that the cultivar by environment by management interaction that could be exploited commercially was small. The environmental and management effects (time of sowing and site) drive the variability in biomass and that translates into variability in yield. Cultivar has a much greater influence on grain number and grain weight, where the variance components and therefore the structure of this variability is quite different to that of grain yield and biomass (Table 5). The implication is that individual cultivars produce a yield with varying combinations of grain number and grain weight, and the two mechanisms can compensate for each other (Fig. 2). Similar findings were reported by de Oliveira et al. (2013), who compared high vigour wheat to Janz under low and high CO2 environments and found that despite possessing vastly different physiological attributes, the yield differences were minimal. Sharma et al. (2008) also found little evidence of useable G × E in Western Australia for time of sowing. The implication of this study and others is that commercially released cultivars of wheat are plastic and broadly adapted to the wide range of environments in the Australian wheat zone. The partitioning of environment into site by time of sowing demonstrated that early sowing was more favourable than later times of sowing and in some instances, very late sown crops could cause a considerable yield penalty. However, the environment of a particular location changed with season. In one season a site may have been in a high yielding group, and in another it would be in the
low yielding group. This suggests the stresses imposed on the crop were the main driver of environment, rather than geographic location or soil type. Furthermore the timing of those stresses could be altered by time of sowing, but the impact timing of stress on grain yield could not be altered by cultivar to the point where a useable G × E × M interaction that brought about a significant change in cultivar rank could be found. Chenu et al. (2011) also demonstrated that the environment for a particular location defined by a stress index also varied considerably and while a particular location had a dominant environment, other environment types were expressed and these ‘other’ environments often accounted for 50% or more of the seasons at that location. Lawes et al. (2009) and Hayman et al. (2010) also used crop simulation modelling and classification to define environmental classes for regions in the southern and western Australian grain belt. The implication of environments changing with season is that it makes the target production environment difficult to define for plant breeders. Therefore, to succeed, cultivars must broadly adapt to these variable environments. Statistically, such an agglomeration of broadly adapted cultivars will also homogenise the definition of environments and result in environments clustering together. For this reason, with commercially released material, the Australian grain belt segregated into just two production environments. Like Chapman (2008), the data imply that the target production environment is constantly moving and this presents breeders with considerable challenges if they wish to define and then target niche production environments for G × E × M solutions to agronomic problems. At one level, the findings from this research are not surprising. The time of sowing literature has routinely demonstrated a yield penalty of 15–50 kg/ha/day after an optimal sowing time (summarised in Sharma et al. (2008)). Here the yield penalty was 39 kg/ha/day, but in two locations it declined to just 9 kg/ha/day and at one location yields improved with later sowing. It has been hypothesised that such a penalty could be avoided by growing fast maturing cultivars and these opinions are based on the frameworks described by Passioura and Angus (2010). The data presented here demonstrate the difficulty of using such a framework to precisely engineer an ideal compromise between time of sowing, phenology, harvest index, grain number, grain weight and yield because the outcome of the season is unknown when cultivar selection is made. One interpretation of the analysis relating to cultivar performance and flowering day (Fig. 5), would be that with the exception of Spring Ridge, farmers should sow a fast to moderately fast maturing cultivar as early as possible to maximise yield. This interpretation only applies to spring wheat cultivars and the sowing window investigated. The data also showed that the key relationships about stress hold, where stress decreases biomass production, reduces grain number, can reduce grain weight, and reduces yield. Breeders may consider how multiple traits interact as the environment changes, rather than concentrate on individual traits. Indeed, Chapman (2008) recommended that cultivars must be tested across as wide a range of environments as possible to ensure that potentially advantageous cultivars are not accidently
46
R.A. Lawes et al. / Europ. J. Agronomy 77 (2016) 38–46
discarded because they performed poorly in an average to below average environment. To that end, the cultivars were broadly adapted, and the most broadly adapted of all was Wyalkatchem. Environment appears to be defined by stress and the extent of terminal drought, and frameworks such as those presented by Chapman et al. (2000), that describe the extent of such a stress are intrinsically useful in broadly defining an environment. Complicating the definition of environment is arguably of little value, because the stresses are influenced by season more than location. Therefore it may be difficult to develop G × E × M strategies that rely on the currently available commercial cultivars. Hammer et al. (2014) postulated that breeding programmes and agronomic development occur independently from each other appears to hold here where the data suggest modern breeding strategies may be unintentionally producing broadly adapted cultivars. Chapman (2008) also points out that breeding programmes are faced with an extraordinary challenge, where farmers demand cultivars that tolerate periods of stress, yet yield well in seasons with more rainfall, as this is when farmers generate a profit. The analysis here suggests the breeding programmes have achieved this by releasing broadly adapted material to the industry. Finally, from a management perspective, with current commercially available cultivars, the cultivar by management interaction was not important and can be ignored. The highest ranking cultivar should be sown regardless of sowing date, and often that cultivar was Wyalkatchem or Gregory, or a cultivar that did not yield significantly more than one of these two cultivars (Table 6). 5. Conclusion The analysis of G × E × M across Australia with commercially released cultivars of spring wheat with varying phenology demonstrated that there was little scope to use this variation from a farmers perspective. Commercial cultivars are broadly adapted to a wide range of environments and times of sowing. This occurred, despite finding that the yield components of individual cultivars did vary, and implies that the plasticity of the crop meant that variation in yield components did not translate into variation in yield. This outcome is a result of a breeding strategy designed to give farmers cultivars that yield well in drought seasons and also perform well in favourable seasons. Similarly, for the wide range of cultivars, sites and times of sowing investigated, environments across Australia can be simplified into two zones, north and south, and this simplification is also a by-product of the broadly adapted cultivars currently available to farmers. Acknowledgements We thank GRDC for funding this research. We also thank Dr Alison Kelly from SAGI for providing the trial designs and Dr Ky Mathews for advice on implementing the Factor Analytic method and commentary on the statistical analysis. References Asseng, S., Foster, I., Turner, N.C., 2011. The impact of temperature variability on wheat yields. Global Change Biol. 17, 997–1012.
Chapman, S.C., 2008. Use of crop models to understand genotype by environment interactions for drought in real-world and simulated plant breeding trials. Euphytica 161, 195–208. Chapman, S.C., Cooper, M., Hammer, G.L., Butler, D.G., 2000. Genotype by environment interactions affecting grain sorghum. II. Frequencies of different seasonal patterns of drought stress are related to location effects on hybrid yields. Aust. J. Agric. Res. 51, 209–221. Chenu, K., Cooper, M., Hammer, G.L., Mathews, K.L., Dreccer, M.F., Chapman, S.C., 2011. Environment characterization as an aid to wheat improvement: interpreting genotype-environment interactions by modelling water-deficit patterns in North-Eastern Australia. J. Exp. Bot. 62, 1743–1755. Comstock, R.E., 1977. Quantitative genetics and the design of breeding programs. In: Proceedings of the International Conference on Quantitative Genetics, August 16–21, 1976, pp. 705–718. Cooper, M., Delacy, I.H., 1994. Relationships among analytical methods used to study genotype variation and genotype-by-environment interaction in plant-breeding multi environment experiments. Theor. Appl. Genet. 88, 561–572. Eagles, H.A., Cane, K., Kuchel, H., Hollamby, G.J., Vallance, N., Eastwood, R.F., Gororo, N.N., Martin, P.J., 2010. Photoperiod and vernalization gene effects in southern Australian wheat. Crop & Pasture Sci. 61, 721–730. Finlay, K.W., Wilkinson, G.N., 1963. Analysis of adaptation in a plant breeding proramme. Aust. J. Agric. Res. 14, 742–754. Fletcher, A.L., Robertson, M.J., Abrecht, D.G., Sharma, D.L., Holzworth, D.P., 2015. Dry sowing increases farm level wheat yields but not production risks in a Mediterranean environment. Agric. Syst. 136, 114–124. Hammer, G.L., McLean, G., Chapman, S., Zheng, B., Doherty, A., Harrison, M.T., Oosterom, E.v., Jordan, D., 2014. Crop design for specific adaptation in variable dryland production environments. Crop Pasture Sci. 65, 614–626. Hayman, P.T., Whitbread, A.M., Gobbett, D.L., 2010. The impact of El Nino Southern Oscillation on seasonal drought in the southern Australian grainbelt. Crop Pasture Sci. 61, 528–539. Kamran, A., Kubota, H., Yang, R., Randhawa, H.S., Spaner, D., 2014. Relative performance of Canadian spring wheat cultivars under organic and conventional field conditions. Euphytica 196, 13–24. Lawes, R.A., Oliver, Y.M., Robertson, M.J., 2009. Integrating the effects of climate and plant available soil water holding capacity on wheat yield. Field Crops Res. 113, 297–305. Lemerle, D., Gill, G.S., Murphy, C.E., Walker, S.R., Cousens, R.D., Mokhtari, S., Peltzer, S.J., Coleman, R., Luckett, D.J., 2001. Genetic improvement and agronomy for enhanced wheat competitiveness with weeds. Aust. J. Agric. Res. 52, 527–548.Nix, H.A., Fitzpatrick, E.A., 1969. An index of crop water stress related to wheat and grain sorghum yields. Agric. Meteorol. 6, 321–337. Passioura, J.B., Angus, J.F., 2010. Improving productivity of crops in water-limited environments. Adv. Agron. 106, 37–75. R Core Team, 2013. R: A Language and Environment for Statistical Computing. R Foundation for Statistical Computing, Vienna http://www.R-project.org. Rebetzke, G.J., Kirkegaard, J.A., Watt, M., Richards, R.A., 2014. Genetically vigorous wheat genotypes maintain superior early growth in no-till soils. Plant Soil 377, 127–144. Richards, R.A., Hunt, J.R., Kirkegaard, J.A., Passioura, J.B., 2014. Yield improvement and adaptation of wheat to water-limited environments in Australia-a case study. Crop Pasture Sci. 65, 676–689. Sharma, D.L., D’Antuono, M.F., Anderson, W.K., Shackley, B.J., Zaicou-Kunesch, C.M., Amjad, M., 2008. Variability of optimum sowing time for wheat yield in Western Australia. Aust. J. Agric. Res. 59, 958–970. Smith, A., Cullis, B., Thompson, R., 2001. Analyzing variety by environment data using multiplicative mixed models and adjustments for spatial field trend. Biometrics 57, 1138–1147. Smith, A.B., Cullis, B.R., Thompson, R., 2005. The analysis of crop cultivar breeding and evaluation trials: an overview of current mixed model approaches. J. Agric. Sci. 143, 449–462. Smith, A.B., Ganesalingam, A., Kuchel, H., Cullis, B.R., 2015. Factor analytic mixed models for the provision of grower information from national crop variety testing programs. Theor. Appl. Genet. 128, 55–72. Williams, E., Piepho, H.-P., Whitaker, D., 2011. Augmented p-rep designs. Biom. J. 53, 19–27. Zadoks, J.C., Chang, T.T., Konzak, C.F., 1974. Decimal code for growth stage of cereals. Weed Res. 14, 415–421. de Oliveira, E.D., Bramley, H., Siddique, K.H.M., Henty, S., Berger, J., Palta, J.A., 2013. Can elevated CO2 combined with high temperature ameliorate the effect of terminal drought in wheat? Funct. Plant Biol. 40, 160–171.
Contents lists available at ScienceDirect
European Journal of Agronomy journal homepage: www.elsevier.com/locate/eja
Commercially available wheat cultivars are broadly adapted to location and time of sowing in Australia’s grain zone R.A. Lawes a,∗ , N.D. Huth b , Z. Hochman c a b c
CSIRO Agriculture, 147 Underwood Avenue, Floreat, WA 6014, Australia CSIRO Agriculture, PO Box 2583, Brisbane, Qld 4001, Australia CSIRO Agriculture, 203 Tor St, Toowoomba, Qld 4350, Australia
a r t i c l e
i n f o
Article history: Received 9 October 2015 Received in revised form 8 February 2016 Accepted 29 March 2016 Available online 9 April 2016 Keywords: G×E×M Wheat Broad adaptation
a b s t r a c t Farmers must choose which cultivar to grow based on the phenology of the cultivar and anticipated season length. The current study investigated the established doctrine of sowing fast maturing cultivars late, and slow maturing cultivars early. This was explored by quantifying the genotype (G) × environment (E) × management (M) available to farmers using commercially released cultivars, where management relates to the time of sowing. Nineteen cultivars of spring wheat (Triticum aestivum) were sown at 3 times of sowing (early, conventional and late) at 13 sites in 2011 and 2012. Sites were located throughout the Australian grain growing region in Queensland, New South Wales, Victoria, South Australia and Western Australia from latitudes 27◦ 34 S to 35◦ 09 S where annual rainfall ranged from 237 mm to 747 mm. In general, the three way interaction between G, E and M for yield was small and cultivar could not overcome the yield penalty associated with a late time of sowing. At 11 of the 13 sites, fast to moderately fast maturing cultivars sown early generated the highest yields. Fast maturing cultivars sown late could not compensate for a late time of sowing. Commercial cultivars were broadly adapted to environment and management, and with these cultivars, the Australian grain growing region could be split into just two environments, south and north. Even then, season appears to be the main arbiter of environment, rather than location per se as individual sites moved from one group to the other, depending on season. There was no evidence to suggest farmers could exploit a cultivar by management interaction for time of sowing with commercial cultivars, as the outcome of the season is unpredictable, and with current technology farmers should simply choose the best performing cultivar for their region and sow it as early as possible. Crown Copyright © 2016 Published by Elsevier B.V. All rights reserved.
1. Introduction The goal of plant breeders is to produce new cultivars that are better adapted to a targeted production environment than their predecessors and to release these cultivars for commercial use by farmers. The farmer must then choose a cultivar for their particular environment and sow it at a particular time. The time of sowing immediately introduces a management component into the choice of cultivar. These objectives, enunciated by Comstock (1977) and discussed by Chapman (2008) highlight the importance of breeding cultivars for a particular environment, and the response of the cultivar in a particular environment can be further influenced by management. Therefore the farmer and the breeder must consider genotype (G), environment (E), management (M) and the interactions of those components.
∗ Corresponding author. E-mail address: [email protected] (R.A. Lawes). http://dx.doi.org/10.1016/j.eja.2016.03.009 1161-0301/Crown Copyright © 2016 Published by Elsevier B.V. All rights reserved.
Agronomists and plant breeders are both conscious of environment, but the two disciplines have evolved differently. For an agronomist, the environment has historically been defined in terms of the stresses applied to a plant through its life. These may include water stress (Nix and Fitzpatrick, 1969), nutrient stress, and temperature stress (Asseng et al., 2011). The importance of these stresses are summarised by frameworks such as those described by (Passioura and Angus, 2010), who highlight the importance of matching the phasic development of the plant to the water supply and temperature stresses of a particular region. For breeders, the definition of environment may be defined statistically, where cultivars will be grown across a wide range of environments (Cooper and Delacy, 1994). The variance components from an analysis of variance may be employed to determine the extent of variation in a trait that can be attributed to cultivar, the environment and the cultivar by environment interaction. Since large numbers of cultivars are often screened, statistical techniques have evolved to group cultivars into those that may be broadly adapted to an environment or specifically adapted to
R.A. Lawes et al. / Europ. J. Agronomy 77 (2016) 38–46
39
Table 1 Cultivars and the number of plots of each cultivar sown at each site. Maturity
Fast
Fast–Moderate
Moderate
Slow
a
State
Queensland
New South Wales
Victoria
South Australia
Location
Bungunya
Nangwee
Spring Ridge
Temora
Walpeup
Minnipa
Turretfield
Corrigin
Year
2012
2012
2011
2011
2011
2011
2012
2011
2012
2011
2012
2011
2012
3
9
9
9
9
9 9
9
9
9
Axe Westonia Lincoln Livingston Wyalkatchem Crusader Carnamah Gladius Yitpi Scout Janz Derrimut Correll Magenta Lang Gregory Wylie Endure Bolac
9
9
9
Western Australia Eradu
6 9 3
6
6
6
6
9
9
9
9
9
9 6
9
6 3 3
6 3
6 3 3
6 3
9
9
9
6
3
3
3
6
9 6
9 6
6
6
6
6
6
6
6 6
6
6 6
6
3 6
6
3 6
9 6 3
9 3
9 3
9 6
9 6
9
9
9
9
3
6
6
3
3
3
3
3
3
6
6
6
6
9 9
8a 9
9 9
9
9
9
9
9
Missing plot TOS 1.
an environment. Therefore, concepts such as the regression on the environment mean (Finlay and Wilkinson, 1963), and various multivariate techniques have been employed to detect G × E interactions (Cooper and Delacy, 1994). The analysis of multienvironment trials has evolved and is now quite flexible (Smith et al., 2005). For example, the nuances of individual trials can be accommodated where some sites may have strong spatial processes that need to be statistically modelled. Assumptions about the covariance structure of cultivars and environment at particular locations can also be explored using either compound symmetry, diagonal or factor analytic models (Smith et al., 2001). On occasions, environmental covariates such as rainfall can be introduced into such a model to further define environment (Eagles et al., 2010). The attempt to define cultivar performance for specific environments is partly driven by the need to communicate to farmers which cultivars will be suited to their farm. Furthermore, seasonal conditions and logistics may influence management decisions such as the time of sowing, where conventional wisdom suggests longer season cultivars should be sown early, and shorter season cultivars should be sown later. Some fast maturing cultivars are adapted to regions where the cold and heat stresses occur in close temporal proximity to each other. Slower maturing cultivars have been developed for environments with higher rainfall, where there is an opportunity for the plant to remain in the vegetative stage longer, produce more biomass, produce a higher number of grains, spend longer during grain fill and produce grains of equivalent kernel weight to the shorter season cultivar (Richards et al., 2014). Therefore the farmer and the plant breeder are trying to grow a crop where the timing and duration of vegetative and reproductive growth are optimised to use all the available resources on offer and avoid temperature stress. Unfortunately, the water deficit patterns and extent of the stresses all vary with season and location (Chenu et al., 2011), and this complicates the development of cultivars for niche environments and management actions. Since the timing of stress on a crop can be influenced by phenology (cultivar), the location and season (environment) and the time of sowing (management), the interaction between G × E × M may be important to farmers. The three way interactions are increasingly being examined for factors such as tillage practice (Rebetzke et al., 2014), organic and conventional systems (Kamran et al., 2014)
and weed management (Lemerle et al., 2001). Many of these earlier studies evaluated genetic material that had yet to be released to the industry, and these earlier studies were all attempting to identify whether G × E × M existed and should become part of the breeding process. Our goal here is to evaluate the G × E × M concept as it applies to farmers with regard to time of sowing using current, commercially available cultivars of varying maturity. To that end, the performance of 19 elite commercially available wheat cultivars of varying phenology sown at 3 different times in 13 locations across the entire Australian grain growing region were evaluated. Trials were located in Western Australia, South Australia, Victoria, New South Wales and Queensland. These data are used to determine whether farmers should change cultivars, given the time of sowing, where long season cultivars should be sown early and shorter season cultivars sown later. The current study quantifies the extent of the genetic variation available to farmers, and also considers whether niche environments exist across the Australian continent that farmers can realistically exploit with currently available genetic material. 2. Materials and methods Nineteen cultivars of spring wheat (Triticum aestivum) were chosen and selected in consultation with plant breeders to ensure cultivars were suited to a particular location, thus by definition, not all cultivars were grown at each location. Cultivars with maturities varying from early through to late were grown at each location (Table 1). At every location cultivars were sown at three times of sowing (TOS), early (late April to early May; TOS 1), conventional time of sowing (mid May to early June; TOS 2) and late (late June to early July; TOS 3). Specific times of sowing at each location are presented in Table 2. Sites were managed so that nutrients were non-limiting. Fertiliser was applied at sowing using formulations of S, K, and trace elements appropriate to the region and the soil type. If the season was favourable and soil fertility was considered limiting, additional N was applied to ensure the crops could achieve their highest possible yield potential as assessed in August. At each site total N applied is presented in Table 3. The sites had various soil types ranging from black vertosols of heavy texture at Nangwee in Queens-
40
R.A. Lawes et al. / Europ. J. Agronomy 77 (2016) 38–46
Table 2 Time of sowing for each site. Site
Time of sowing 1
Time of sowing 2
Time of sowing 3
Bungunya Nangwee Spring Ridge Temora Walpeup Turretfield Turretfield Minnipa Minnipa Corrigin Corrgin Eradu Eradu
10/05/2012 17/05/2012 09/05/2011 05/05/2011 29/04/2011 30/05/2012 18/05/2011 25/05/2012 13/05/2011 02/05/2012 30/04/2011 02/5/2012 29/04/2011
22/05/2012 31/05/2012 06/06/2011 26/05/2011 31/05/2011 15/06/2012 08/06/2011 08/06/2012 27/05/2011 21/05/2012 24/05/2011 21/05/2012 24/05/2011
23/06/2012 23/06/2012 23/06/2011 20/06/2011 01/07/2011 05/07/2012 28/06/2011 25/06/2012 24/06/2011 21/06/2012 21/06/2011 21/06/2012 23/06/2011
Table 3 Total N applied at each site, starting soil nitrate and initial plant available water (PAW) at sowing. Site
N applied (kg/ha)
Initial PAW (mm)
Initial NO3 (kg/ha)
Bungunya 2012 Nangwee 2012 Spring Ridge 2011 Temora 2011 Walpeup 2011 Minnipa 2011 Minnipa 2012a Turretfield 2011 Turretfield 2012 Corrigin 2011 Corrigin 2012 Eradu 2011 Eradu 2012
9.0 9.0 100.8 57.0 31.2 10.8 60.5 28.8 28.8 56.0 56.0 61.0 58.0
38.3 186.8 167.1 135.3 (to 150 cm) 104.4 69.9 35.3 43.4 65.2 23.2 (to 85 cm) 0b 42.5 0b
63.9 608.7 183.5 130.4 (to 150 cm) 256.1 89.3 49.5 205.6 13.2 50.6 (to 60 cm) 11 (to 30 cm) 113.0 17.7 (to 30 cm)
a b
TOS 1 and 2 had 60.5 kg N/ha applied, TOS 3 had 35.6 kg N/ha applied. Initial soil moisture not measured, but soils were dry and plant available water was assumed to be 0.
land to the deep yellow-brown loamy sand at Eradu in Western Australia. Some soils were acidic (Eradu, Corrigin), while others were alkaline (Turretfield, Walpeup). The plant available soil water holding capacities ranged from 77 mm at Turretfield to 246 mm at Nangwee. Rainfall patterns also varied considerably, with 8 sites (Bungunya, Eradu 2011, Minnipa 2011, Nangwee, Temora 2011, Turretfield 2012, Walpeup 2011 and Spring Ridge 2011) receiving more than 100 mm outside the traditional April to October rainfall. Growing season rainfall (April to October) ranged from just 136 mm at Minnipa in 2012 to 340 mm in Corrigin in 2011 while annual rainfall ranged from 281 mm at Eradu in 2012 to 747 mm at Temora in 2011 (Table 4). These variations reflect the diversity of the environments expressed across the Australian wheatbelt, but are not necessarily a representative sample of every environment in every season. At each site hand harvests from 4 quadrats (0.25 m2 ) were taken and the yield components of grain number, grain weight and biomass at harvest were measured. From these components, the harvest index was calculated. Grain yield was estimated by hand harvest of the plots. Phenology was monitored for each cultivar to determine the date of flowering. Flowering date (Zadoks growth stage 65; (Zadoks et al., 1974)) was estimated by eye for each cultivar at each time of sowing from multiple sequential observations during, before and after anthesis.
2.1. Design and analysis The experimental design is an unbalanced incomplete split plot block design, where there are two blocks (replicates) in each trial. The main plots are sowing times (early, normal and late) and the split plots are cultivars (7–9) which are incompletely replicated within and across sowing times. That is, at a trial a cultivar may be
replicated 3 times at each time of sowing (9 plots), twice at each time of sowing (6 plots) or once at each time of sowing (3 plots), i.e. within a block, a cultivar at a particular TOS may be replicated zero, once or twice. Three cultivars, Janz, Gregory and Gladius were sown in all trials following the principals of a p-rep design (Williams et al., 2011). The trial was orientated as a single long row and the dimensions were 1 row and 60 columns. With this layout, modelling the spatial variability along the row may be important. The design was then replicated across 13 sites, where different cultivars were used at each site depending on their geographical location (Table 4).
2.2. Statistical analysis Data were analysed using ASREML, a linear mixed effects model, in the R statistical package (R Core Team, 2013). Initially, each trial was analysed as a separate entity, to quantify the site level genetic variance, ascertain whether there was a spatial process that needed to be modelled separately at that particular site and determine whether the genetic variance at a particular site was bounded. For each site, the yield variance components were estimated. Time of sowing (the environment), was fixed, as is often recommended (Smith et al., 2005). The design attributes of replicate and time of sowing within replicate were fit as random effects. Cultivar and the interaction between cultivar and time of sowing were also fit as random effects. An AR1 autoregressive covariance structure was evaluated by AIC and BIC at each location (Table 5). If the AR1 covariance structure that accounted for spatial processes improved the model fit at that site, it was retained. Following these investigations, a Multi Environment Trial (MET) analysis was conducted, excluding those trials with a bounded genetic variance. The approach adopted followed the methods outlined by Smith et al. (2001, 2015). Two models were fitted to explore
R.A. Lawes et al. / Europ. J. Agronomy 77 (2016) 38–46
41
Table 4 Site description of soil type, location, altitude and climate. Site and year
Soil type
Latitude, Longitude
Altitude m asl
Annual Rainfall (January–December)
Growing Season Rainfall (Time of Sowing 1 to Harvest)
Plant Available Water Holding Capacity (mm)
Bungunya 2012
Grey Vertosol
28◦ 25 S, 149◦ 39 E
189
439
137
130
Corrigin 2011
Loamy Duplex
32◦ 22 S, 117◦ 55 E
313
406
340
79
32◦ 22 S, 117◦ 55 E
313
238
170
79
28 40 S, 115 10 E
266
494
326
96
28◦ 40 S, 115◦ 10 E
266
281
245
96
32◦ 49 S, 135◦ 09 E
196
391
221
139
32 49 S, 135 09 E
196
237
136
139
Corrigin 2012 Eradu 2011
Deep Yellowish-Brown Loamy Sand
Eradu 2012 Minnipa 2011
Sandy Clay Loam
◦
◦
Minnipa 2012
◦
◦
Nangwee 2012
Black Vertosol
27◦ 34 S, 151◦ 19 E
362
480
169
246
Temora 2011
Red Chromosol
34◦ 24 S, 147◦ 31 E
274
747
294
147
34◦ 32 S, 138◦ 47 E
129
528
310
77
34 32 E, 138 47 S
129
382
243
77
35◦ 09 E, 142◦ 01 S
77
448
152
105
31.19 E, 150◦ 11 S
325
596
266
183
Turretfield 2011 Calcareous Loam
◦
Turretfield 2012 Walpeup 2011
Loamy Sand
Spring Ridge 2011Grey-black Vertosol
◦
Table 5 AIC for standard and AR1 models for yield and variance components for each trait at each site for cultivar and the interaction between cultivar and time of sowing. Site
AIC AIC AR 1 Cultivar yield
Cultivar × TOS
Cultivar biomass
Cultivar × TOS
Cultivar grain number
Cultivar × TOS
Cultivar Grain weight
Cultivar × TOS
Bungunya 2012 Corrigin 2011 Corrigin 2012 Eradu 2011 Eradu 2012 Minnipa 2011 Minnipa 2012* Nangwee 2012 SpringRidge 2011 Temora 2011 Turretfield 2011 Turretfield 2012 Walpeup 2011
712 681 686 675 603 667 631 701 716 701 685 685 645
9.2% 0.0% 0.0% 11.3% 6.4% 0.0% 1.0% 0.0% 0.0% 0.0% 18.5% 0.0% 65.6%
13.6% 0.0% 4.9% 36.3% 0.0% 14.7% 31.3% 0.0% 46.8% 23.0% 1.1% 0.0% 7.9%
9.6% 0.0% 0.0% 5.2% 0.0% 0.0% 0.6% 0.0% 0.0% 0.0% 0.0% 0.0% 44.3%
14.9% 1.4% 1.4% 8.1% 0.0% 47.6% 21.4% 6.0% 35.2% 35.9% 10.4% 10.2% 10.5%
5.5% 0.0% 3.4% 34.8% 5.5% 0.0% 0.0% 0.0% 0.2% 0.0% 48.9% 0.0% 58.7%
68.9% 36.4% 22.5% 57.4% 61.8% 81.5% 35.3% 67.9% 56.0% 56.6% 35.0% 28.5% 23.7%
14.1% 17.9% 1.2% 30.9% 27.2% 6.2% 5.2% 9.1% 20.6% 0.0% 38.9% 0.0% 42.8%
714 683 672 677 605 666 633 702 715 701 683 687 647
11.5% 15.4% 5.8% 53.0% 11.2% 44.5% 34.9% 0.0% 47.0% 23.2% 0.0% 0.0% 0.0%
the site by cultivar by time of sowing interactions. The models chosen were a diagonal model structure (Diag), and a Factor Analytic structure (FA). In both cases the cultivar was not explicitly fitted. With the diagonal structure, trials were assumed independent with heterogeneous variances. With the FA structure the trial variance heterogeneity and between trial correlation was accommodated through a hypothetical factor (k), which in this case was 1 (Smith et al., 2001). Again AIC was used to select the best model out of the diagonal and FA alternative. Site and TOS were fixed and represent the environments. The random model included the interaction between site and replicate; the 3 way interaction between site, replicate and time of sowing; the interaction between site and cultivar; the interaction between site and time of sowing and the 3 way interaction between site, cultivar and TOS. In the random model, site was assumed to have a diagonal covariance structure, where the covariance between sites was assumed to be 0. An AR1 covariance structure was used for 3 sites (Corrigin 2012, Spring Ridge 2011 and Turretfield 2011), where a spatial process had previously been identified. The output
generated from the diagonal model was equivalent to analysing each trial separately. For the FA model, an FA structure where k = 1 was employed for the site by cultivar interaction. All other terms were fitted as described with the diagonal model. Predicted values for all site by TOS by cultivar combinations were then generated for yield (t/ha), biomass (t/ha), the square root of grain number and 1000 grain weight (mg), using the predict command in ASREML. The predicted values for yield were plotted against the day of flowering for cultivar at each location across all times of sowing. Similarly, the relationship between grain number, grain weight and biomass were all considered for each cultivar. The predicted values were subjected to an hierarchical cluster analysis where the columns were hand harvested yield, grain number, grain weight biomass and the rows were the 700 combinations of site, cultivar and time of sowing. Data were then standardised for each variable (column). Agglomeration was performed using the Manhattan metric. The objective here was to explore whether the accumulation of yield through the various yield components varied across the Aus-
42
R.A. Lawes et al. / Europ. J. Agronomy 77 (2016) 38–46
Fig. 1. The predicted values for grain weight (mg) and grain number for each cultivar across all sites and times of sowing.
tralian wheat belt and whether groups formed around locations, times of sowing or particular cultivars.
3. Results Yields varied considerably between sites, TOS and cultivar. Low yielding sites, where the best performing cultivar yielded less than 2.5 t/ha were at Corrigin in 2012 (2.1 t/ha), Eradu in 2012 (1.0 t/ha) and Minnipa in 2012 (2.3 t/ha). Moderate yields were generated by the best performing cultivar at Minnipa in 2011 (4.1 t/ha), Eradu in 2011(3.1 t/ha) and Bungunya in 2012 (3.3 t/ha). High yields, greater than 4 t/ha, were produced at Spring Ridge in 2011 (5.6 t/ha), Temora (4.4 t/ha) and Corrigin in 2011 (4.4 t/ha). Within a site, the expected decline in yield from the best performing cultivar at time of sowing 1 to time of sowing 2 and 3 varied between locations. From time of sowing 1 to time of sowing 3 yields typically declined by between 0.8 and 1.3 t/ha from Corrigin (2011, 2012), Minnipa (2012), Eradu (2012), Bungunya (2012). More moderate declines in yield occurred at Eradu in 2012 (0.2 t/ha), and Temora in 2011 (0.7 t/ha) and the range of yields at each site and time of sowing are presented in Table 6. In Spring Ridge, yields increased by 1.5 t/ha from time of sowing 1 to time of sowing 3. Similarly the yield decline between time of sowing 1 and time of sowing 2 ranged from 0.1 t/ha at Temora to 0.5 t/ha in Minnipa in 2012. At all other sites yields declined by between 0.2 and 0.4 t/ha from time of sowing 1 to time of sowing 2. The exception was the increase of 1.6 t/ha from time of sowing 1 to time of sowing 2 at Spring Ridge. Therefore, across sites and times of sowing, yields ranged from 0.7 t/ha to 5.6 t/ha, and at all sites except Spring Ridge, delayed time of sowing generally reduced yield (Table 6). Cultivar and the interaction between cultivar and time of sowing at the site level were comparatively small for grain yield, with some exceptions. At Bungunya, Corrigin in 2011 and 2012 and Eradu in 2012 the genetic variance was approximately 10% and the genotype by management component was approximately 9% or less (Table 5). The genetic variance at Eradu, Minnipa, Spring Ridge and Temora were all 23% or greater. However, providing the main cultivar effect was unbounded, the genotype by environment component for yield was greater than 5% at every location except Minnipa in 2012 (Table 5). Therefore, before the meta-analysis of data was conducted, the
Fig. 2. The predicted values for grain weight (mg) and grain number for each site across all cultivars and times of sowing.
G × M that can be exploited using commercial cultivars was small. Within a site the variance component for cultivar and the variance component for the interaction between cultivar and time of sowing for biomass and grain number were commensurate with the equivalent variance components for yields at most sites (Table 5). One exception occurred at Minnipa in 2011, where the variance component for cultivar yield was 44%, but the variance component for cultivar biomass was just 14%. The variance components for cultivar and the interaction between cultivar and TOS for grain weight were often larger than the corresponding variance components for the other 3 traits. The outputs from the meta-analysis confirmed the findings from the initial analysis of data at the individual sites with regard to the small amount of genetic variation and even smaller amount of genotype by management interaction within a site. To that end, the best performing genotype did vary between sites, but rarely varied within a site between TOS 1, TOS 2 and TOS 3 (Table 7). For example, Wyalkatchem, an early to mid season cultivar was the best performing, or equal best performing at Corrigin in 2011 and 2012, Eradu in 2011 and 2012 and Minnipa in 2011. Wyalkatchem was bested by the faster maturing Axe at Minnipa at 2012. However across these southern and western sites, Wyalkatchem dominated at all times of sowing. While Crusader was identified as the highest ranked cultivar at TOS 3 at Eradu, its performance at that time of sowing was not significantly different from Wyalkatchem. At the eastern sites, in Temora, Bungunya and Spring Ridge, the cultivar Gregory was the best performing or equal best performing, regardless of time of sowing. Gregory was characterised as a longer season cultivar, which should theoretically yield less than faster maturing cultivars when sown late. This did not occur at any of these locations. 3.1. Relationships between traits Overall, across sites and cultivars there were strong correlations between grain yield biomass (r = 0.94) and grain number (r = 0.94). The relationship between grain yield and grain weight was comparatively poor (r = 0.58).
R.A. Lawes et al. / Europ. J. Agronomy 77 (2016) 38–46
43
Table 6 The range of yields (kg/ha) for each site and corresponding time of sowing. TOS 1 Range Bungunya 2012 Corrigin 2011 Corrigin 2012 Eradu 2011 Eradu 2012 Minnipa 2011 Minnipa 2012 Spring Ridge 2011 Temora 2011
3034 3852 1602 1974 759 3357 1878 3055 3626
TOS 2 Range 3504 4494 2305 3516 1043 4080 2361 4104 4418
2757 3550 1299 1818 740 3219 1380 4656 3377
TOS 3 Range 3141 4191 2002 3160 1024 3942 1853 5705 4169
2170 3002 751 729 540 2163 1319 4543 2731
2622 3643 1454 1917 824 2885 1791 5592 3523
Table 7 Best performing cultivar with the predicted yield and standard error from the Multi Environment Trial Factor Analytic model.
Corrigin 2011 Corrigin 2012 Eradu 2011 Eradu 2012 Minnipa 2011 Minnipa 2012 Temora 2011 SpringRidge 2011 Bungunya 2012 a
Time of Sowing 1
Time of Sowing 2
Time of Sowing 3
4.5 ± 0.1 Wyalkatchem Crusader 2.3 ± 0.3 Wyalkatchem 3.5 ± 0.2 Wyalkatchem 1.0 ± 0.4 Wyalkatchem Crusader Axe 4.1 ± 0.2 Wyalkatchem Crusader Axe 2.4 ± 0.1 Axe 4.4 ± 0.2 Gregory 4.1 ± 0.2 Gregory Janz 3.5 ± 0.16 Gregory
4.2 ± 0.1 Wyalkatchem Crusader 2.0 ± 0.3 Wyalkatchem 3.5 ± 0.2 Wyalkatchem 1.0 ± 0.4 Wyalkatchem Crusader Axe 3.9 ± 0.2 Wyalkatchem Crusader 1.9 ± 0.1, Axe 4.2 ± 0.2 Gregory 5.7 ± 0.2 Gregory Janz 3.1 ± 0.16 Gregory
3.6 ± 0.1 Wyalkatchem Crusader 1.5 ± 0.3 Wyalkatchem 1.9 ± 0.3 Crusadera 0.8 ± 0.4 Wyalkatchem Crusader Axe 2.9 ± 0.2 Wyalkatchem Crusader 1.8 ± 0.1 Axe 3.5 ± 0.2 Gregory 5.6 ± 0.2 Gregory Janz 2.6 ± 0.16 Gregory
Crusader was not grown at the site.
Fig. 3. The predicted values for biomass and grain number for each cultivar across all sites and times of sowing.
Plots for the predicted values of grain number vs grain weight for each cultivar across all sites and times of sowing demonstrate that some cultivars, such as Axe, Wyalkatchem and Carnamah have grain weights that remain relatively stable across a wide range of grain numbers (Fig. 1). Conversely, grain weight varies more for cultivars such as Crusader, Scout, Janz, Lang, and Bolac across a wide range of grain numbers (Fig. 1). When the predicted values of grain weight and grain number for each site are examined across all cultivars and times of sowing, the driver of yield varies (Fig. 2). At Eradu in 2012, yields were heavily influenced by grain weight within a time of sowing. At Bungunya and Corrigin (both years), both traits vary. For sites such as Spring Ridge and Minnipa in 2011, the ability to assemble large numbers of grains appeared to be important as there was comparatively little variation in grain weight. Therefore the “best” mechanism for assembling yield varied with site and season. The linear relationship between biomass and grain number did not vary between cultivars (Fig. 3), where 1 kg/ha of biomass produced 11.3 grains/ha. However, the slope and intercept varied between cultivars for the relationship between biomass and grain weight, where Crusader, Lang, Wylie, Endure and Bolac all produced heavier grains per unit of biomass, but with lower intercepts (Fig. 4). One interpretation of this finding is that biomass influences the determination of grain number more than grain weight and grain weight is determined by other factors. 3.2. Does phenology influence yield?
The best linear unbiased estimates of grain number for time of sowing showed a decline of just 63 and 327 grains/ha from TOS 1 to TOS 2 and TOS 3, which represented a change of just 0.3% from the intercept. In contrast, grain weights declined by 1.5 and 3.6 mg from TOS 1 to TOS 2 and TOS 3 respectively. This decline for TOS 3 represented a reduction of 9.3%. The variance components for cultivar and the interaction between cultivar and time of sowing for grain weight (Table 5), were often larger than those for grain number. One interpretation of these data is that grain number is determined earlier in the season and does not vary greatly with TOS within a site, and cultivars then assemble yield differently and express this difference through grain weight more than grain number.
Flowering dates varied between cultivars, dates of sowing and sites. The cultivar Axe was earlier than every other cultivar, while Bolac flowered later than every other cultivar. In general, Axe, Westonia, Wyalkatchem and Crusader flowered earlier than cultivars such as Lang, Gregory, Wylie, Endure and Bolac. Cultivars including Gladius, Yitpi, Scout, Correll and Magenta were of intermediate flowering time. There was a negative relationship between flowering date and yield, where on average, yield decline by 39 ± 7.7 kg/ha per day after the optimum. This rate of yield decline occurred at Bungunya, Corrigin (both years), Eradu in 2011, Minnipa (both years) and Temora. The decline in yield was less at Eradu in 2012 where
44
R.A. Lawes et al. / Europ. J. Agronomy 77 (2016) 38–46
Fig. 4. The predicted values for biomass and grain weight (mg) for each cultivar across all sites and times of sowing.
Fig. 5. Grain yield vs the day of flowering for the 9 sites and 3 times of sowing.
3.3. Number of distinct environments in Australia yields declined by just 9 kg/ha/day and at Spring Ridge, where yields increased with delayed flowering by 73 kg/ha/day (Fig. 5). While linear representations define a trend over a 30 day flowering window, at smaller temporal scales (∼10 days) there was considerable variation between the cultivars where a slightly (2–5 day) later maturing cultivar out yielded the earlier cultivar. This difference may be due to other attributes of the cultivar (Fig. 5). Regardless, at 6 of the 9 sites sowing early and flowering early produced higher yields.
Cultivars, sites and times of sowing separated into just two groups when the dendrogram (not shown) was cut at a height of 200 and these groups could be loosely described as high yielding (group 1) with 249 entities and low yielding (group 2) with 264 entities. Group 1 had a median yield of 3.8 t/ha, median biomass of 8.4 t/ha, median grain number of 95,000/ha and a median grain weight of 40 mg. Conversely, group 2 had a median yield of 1.7 t/ha, median biomass of 4.7 t/ha, median grain number of 55,000 g/ha,
Fig. 6. Boxplots of traits for the two groups identified by the hierarchical cluster analysis for a) yield, b) biomass, c) grain number and d) grain weight.
R.A. Lawes et al. / Europ. J. Agronomy 77 (2016) 38–46
45
Table 8 Group membership for sites, times of sowing and individual cultivars.
Bungunya Corrgin 2011 Corrigin 2012 Eradu 2011 Eradu 2012 Minnipa 2011 Minnipa 2012 Spring Ridge 2011 Temora
Time of Sowing 1
Time of Sowing 2
Time of Sowing 3
1 1 2 2 (Axe, Westonia, Wyalkatchem, 1) 2 1 2 1 1
1 (Lang 2) 1 2 2 (Axe, Wyalkatchem 1) 2 1 2 1 1
2 1 2 2 2 2 2 1 1 (Crusader Lang 2)
and median grain weight of 31 mg (Fig. 6a–d). At Corrigin in 2011 and 2012, Eradu in 2012 and Spring Ridge in 2011, neither the time of sowing or cultivar had any impact on group membership. At Minnipa in 2011, the first two times of sowing separated into the first group, and the third time of sowing separated into the second. There were two instances, at Bungunya at time of sowing 2, and Temora at time of sowing 3, where cultivar alone moved from the high yielding group to the low yielding group. These cultivars were Lang at Bungunya and Crusader and Lang at Temora (Table 8). Finally, at Eradu in 2011 the cultivars Axe, Wyalkatchem and Westonia moved from group 2 to group 1 for time of sowing 1. At time of sowing 2 at Eradu in 2011 Axe and Wyalkatchem also moved from group 2 to group 1. Importantly, sites could move from group 1 to group 2 depending on season. 4. Discussion The selected sites provided a wide range of environments where the site mean yields ranged from 0.8 to 4.5 t/ha. The range of environments was further extended by varying the time of sowing. This effect alone varied from 2.41 t/ha to 3.27 t/ha and the decline in yield with time of sowing supports agronomic results found elsewhere (Sharma et al., 2008; Fletcher et al., 2015). This wide range of environments was established to test whether farmers should alter their cultivar choice based on time of sowing, given the phenology of the cultivar. The analysis of 19 commercially available cultivars with varying phenology grown across 13 sites with 3 times of sowing demonstrated that the cultivar by environment by management interaction that could be exploited commercially was small. The environmental and management effects (time of sowing and site) drive the variability in biomass and that translates into variability in yield. Cultivar has a much greater influence on grain number and grain weight, where the variance components and therefore the structure of this variability is quite different to that of grain yield and biomass (Table 5). The implication is that individual cultivars produce a yield with varying combinations of grain number and grain weight, and the two mechanisms can compensate for each other (Fig. 2). Similar findings were reported by de Oliveira et al. (2013), who compared high vigour wheat to Janz under low and high CO2 environments and found that despite possessing vastly different physiological attributes, the yield differences were minimal. Sharma et al. (2008) also found little evidence of useable G × E in Western Australia for time of sowing. The implication of this study and others is that commercially released cultivars of wheat are plastic and broadly adapted to the wide range of environments in the Australian wheat zone. The partitioning of environment into site by time of sowing demonstrated that early sowing was more favourable than later times of sowing and in some instances, very late sown crops could cause a considerable yield penalty. However, the environment of a particular location changed with season. In one season a site may have been in a high yielding group, and in another it would be in the
low yielding group. This suggests the stresses imposed on the crop were the main driver of environment, rather than geographic location or soil type. Furthermore the timing of those stresses could be altered by time of sowing, but the impact timing of stress on grain yield could not be altered by cultivar to the point where a useable G × E × M interaction that brought about a significant change in cultivar rank could be found. Chenu et al. (2011) also demonstrated that the environment for a particular location defined by a stress index also varied considerably and while a particular location had a dominant environment, other environment types were expressed and these ‘other’ environments often accounted for 50% or more of the seasons at that location. Lawes et al. (2009) and Hayman et al. (2010) also used crop simulation modelling and classification to define environmental classes for regions in the southern and western Australian grain belt. The implication of environments changing with season is that it makes the target production environment difficult to define for plant breeders. Therefore, to succeed, cultivars must broadly adapt to these variable environments. Statistically, such an agglomeration of broadly adapted cultivars will also homogenise the definition of environments and result in environments clustering together. For this reason, with commercially released material, the Australian grain belt segregated into just two production environments. Like Chapman (2008), the data imply that the target production environment is constantly moving and this presents breeders with considerable challenges if they wish to define and then target niche production environments for G × E × M solutions to agronomic problems. At one level, the findings from this research are not surprising. The time of sowing literature has routinely demonstrated a yield penalty of 15–50 kg/ha/day after an optimal sowing time (summarised in Sharma et al. (2008)). Here the yield penalty was 39 kg/ha/day, but in two locations it declined to just 9 kg/ha/day and at one location yields improved with later sowing. It has been hypothesised that such a penalty could be avoided by growing fast maturing cultivars and these opinions are based on the frameworks described by Passioura and Angus (2010). The data presented here demonstrate the difficulty of using such a framework to precisely engineer an ideal compromise between time of sowing, phenology, harvest index, grain number, grain weight and yield because the outcome of the season is unknown when cultivar selection is made. One interpretation of the analysis relating to cultivar performance and flowering day (Fig. 5), would be that with the exception of Spring Ridge, farmers should sow a fast to moderately fast maturing cultivar as early as possible to maximise yield. This interpretation only applies to spring wheat cultivars and the sowing window investigated. The data also showed that the key relationships about stress hold, where stress decreases biomass production, reduces grain number, can reduce grain weight, and reduces yield. Breeders may consider how multiple traits interact as the environment changes, rather than concentrate on individual traits. Indeed, Chapman (2008) recommended that cultivars must be tested across as wide a range of environments as possible to ensure that potentially advantageous cultivars are not accidently
46
R.A. Lawes et al. / Europ. J. Agronomy 77 (2016) 38–46
discarded because they performed poorly in an average to below average environment. To that end, the cultivars were broadly adapted, and the most broadly adapted of all was Wyalkatchem. Environment appears to be defined by stress and the extent of terminal drought, and frameworks such as those presented by Chapman et al. (2000), that describe the extent of such a stress are intrinsically useful in broadly defining an environment. Complicating the definition of environment is arguably of little value, because the stresses are influenced by season more than location. Therefore it may be difficult to develop G × E × M strategies that rely on the currently available commercial cultivars. Hammer et al. (2014) postulated that breeding programmes and agronomic development occur independently from each other appears to hold here where the data suggest modern breeding strategies may be unintentionally producing broadly adapted cultivars. Chapman (2008) also points out that breeding programmes are faced with an extraordinary challenge, where farmers demand cultivars that tolerate periods of stress, yet yield well in seasons with more rainfall, as this is when farmers generate a profit. The analysis here suggests the breeding programmes have achieved this by releasing broadly adapted material to the industry. Finally, from a management perspective, with current commercially available cultivars, the cultivar by management interaction was not important and can be ignored. The highest ranking cultivar should be sown regardless of sowing date, and often that cultivar was Wyalkatchem or Gregory, or a cultivar that did not yield significantly more than one of these two cultivars (Table 6). 5. Conclusion The analysis of G × E × M across Australia with commercially released cultivars of spring wheat with varying phenology demonstrated that there was little scope to use this variation from a farmers perspective. Commercial cultivars are broadly adapted to a wide range of environments and times of sowing. This occurred, despite finding that the yield components of individual cultivars did vary, and implies that the plasticity of the crop meant that variation in yield components did not translate into variation in yield. This outcome is a result of a breeding strategy designed to give farmers cultivars that yield well in drought seasons and also perform well in favourable seasons. Similarly, for the wide range of cultivars, sites and times of sowing investigated, environments across Australia can be simplified into two zones, north and south, and this simplification is also a by-product of the broadly adapted cultivars currently available to farmers. Acknowledgements We thank GRDC for funding this research. We also thank Dr Alison Kelly from SAGI for providing the trial designs and Dr Ky Mathews for advice on implementing the Factor Analytic method and commentary on the statistical analysis. References Asseng, S., Foster, I., Turner, N.C., 2011. The impact of temperature variability on wheat yields. Global Change Biol. 17, 997–1012.
Chapman, S.C., 2008. Use of crop models to understand genotype by environment interactions for drought in real-world and simulated plant breeding trials. Euphytica 161, 195–208. Chapman, S.C., Cooper, M., Hammer, G.L., Butler, D.G., 2000. Genotype by environment interactions affecting grain sorghum. II. Frequencies of different seasonal patterns of drought stress are related to location effects on hybrid yields. Aust. J. Agric. Res. 51, 209–221. Chenu, K., Cooper, M., Hammer, G.L., Mathews, K.L., Dreccer, M.F., Chapman, S.C., 2011. Environment characterization as an aid to wheat improvement: interpreting genotype-environment interactions by modelling water-deficit patterns in North-Eastern Australia. J. Exp. Bot. 62, 1743–1755. Comstock, R.E., 1977. Quantitative genetics and the design of breeding programs. In: Proceedings of the International Conference on Quantitative Genetics, August 16–21, 1976, pp. 705–718. Cooper, M., Delacy, I.H., 1994. Relationships among analytical methods used to study genotype variation and genotype-by-environment interaction in plant-breeding multi environment experiments. Theor. Appl. Genet. 88, 561–572. Eagles, H.A., Cane, K., Kuchel, H., Hollamby, G.J., Vallance, N., Eastwood, R.F., Gororo, N.N., Martin, P.J., 2010. Photoperiod and vernalization gene effects in southern Australian wheat. Crop & Pasture Sci. 61, 721–730. Finlay, K.W., Wilkinson, G.N., 1963. Analysis of adaptation in a plant breeding proramme. Aust. J. Agric. Res. 14, 742–754. Fletcher, A.L., Robertson, M.J., Abrecht, D.G., Sharma, D.L., Holzworth, D.P., 2015. Dry sowing increases farm level wheat yields but not production risks in a Mediterranean environment. Agric. Syst. 136, 114–124. Hammer, G.L., McLean, G., Chapman, S., Zheng, B., Doherty, A., Harrison, M.T., Oosterom, E.v., Jordan, D., 2014. Crop design for specific adaptation in variable dryland production environments. Crop Pasture Sci. 65, 614–626. Hayman, P.T., Whitbread, A.M., Gobbett, D.L., 2010. The impact of El Nino Southern Oscillation on seasonal drought in the southern Australian grainbelt. Crop Pasture Sci. 61, 528–539. Kamran, A., Kubota, H., Yang, R., Randhawa, H.S., Spaner, D., 2014. Relative performance of Canadian spring wheat cultivars under organic and conventional field conditions. Euphytica 196, 13–24. Lawes, R.A., Oliver, Y.M., Robertson, M.J., 2009. Integrating the effects of climate and plant available soil water holding capacity on wheat yield. Field Crops Res. 113, 297–305. Lemerle, D., Gill, G.S., Murphy, C.E., Walker, S.R., Cousens, R.D., Mokhtari, S., Peltzer, S.J., Coleman, R., Luckett, D.J., 2001. Genetic improvement and agronomy for enhanced wheat competitiveness with weeds. Aust. J. Agric. Res. 52, 527–548.Nix, H.A., Fitzpatrick, E.A., 1969. An index of crop water stress related to wheat and grain sorghum yields. Agric. Meteorol. 6, 321–337. Passioura, J.B., Angus, J.F., 2010. Improving productivity of crops in water-limited environments. Adv. Agron. 106, 37–75. R Core Team, 2013. R: A Language and Environment for Statistical Computing. R Foundation for Statistical Computing, Vienna http://www.R-project.org. Rebetzke, G.J., Kirkegaard, J.A., Watt, M., Richards, R.A., 2014. Genetically vigorous wheat genotypes maintain superior early growth in no-till soils. Plant Soil 377, 127–144. Richards, R.A., Hunt, J.R., Kirkegaard, J.A., Passioura, J.B., 2014. Yield improvement and adaptation of wheat to water-limited environments in Australia-a case study. Crop Pasture Sci. 65, 676–689. Sharma, D.L., D’Antuono, M.F., Anderson, W.K., Shackley, B.J., Zaicou-Kunesch, C.M., Amjad, M., 2008. Variability of optimum sowing time for wheat yield in Western Australia. Aust. J. Agric. Res. 59, 958–970. Smith, A., Cullis, B., Thompson, R., 2001. Analyzing variety by environment data using multiplicative mixed models and adjustments for spatial field trend. Biometrics 57, 1138–1147. Smith, A.B., Cullis, B.R., Thompson, R., 2005. The analysis of crop cultivar breeding and evaluation trials: an overview of current mixed model approaches. J. Agric. Sci. 143, 449–462. Smith, A.B., Ganesalingam, A., Kuchel, H., Cullis, B.R., 2015. Factor analytic mixed models for the provision of grower information from national crop variety testing programs. Theor. Appl. Genet. 128, 55–72. Williams, E., Piepho, H.-P., Whitaker, D., 2011. Augmented p-rep designs. Biom. J. 53, 19–27. Zadoks, J.C., Chang, T.T., Konzak, C.F., 1974. Decimal code for growth stage of cereals. Weed Res. 14, 415–421. de Oliveira, E.D., Bramley, H., Siddique, K.H.M., Henty, S., Berger, J., Palta, J.A., 2013. Can elevated CO2 combined with high temperature ameliorate the effect of terminal drought in wheat? Funct. Plant Biol. 40, 160–171.