Use of a managed stress environment in breeding cotton for a variable rainfall environment

Field Crops Research xxx (xxxx) xxx–xxx

Contents lists available at ScienceDirect

Field Crops Research journal homepage: www.elsevier.com/locate/fcr

Use of a managed stress environment in breeding cotton for a variable rainfall environment ⁎

Warren C. Conaty , David B. Johnston, Alan J.E. Thompson, Shiming Liu, Warwick N. Stiller, Greg A. Constable CSIRO Agriculture & Food, Locked Bag 59, Narrabri, NSW 2390, Australia

A R T I C L E I N F O

A B S T R A C T

Keywords: Breeding target environment Dryland Genetic variability Gossypium hirsutium OZCOT simulation model Rainfed

Australian rainfed cotton is grown in regions with highly variable rainfall, with in-crop rainfall ranging from 100 mm to 800 mm. The CSIRO cotton breeding program conducts rainfed germplasm evaluations at its core research site, as well as a number of regional locations. As a direct result of our variable rainfall environment, yields < 550 kg ha−1 are observed in 20% of years. Historically, these low yielding seasons result in variable experimental data with an increased risk of experimental failure as a non-significant result. Therefore, the aim of this research was twofold. Firstly, to develop and evaluate a managed stress environment (MSE) protocol where ‘rainfed’ germplasm evaluations are irrigated when yield is expected to fall below the minimum threshold for our target breeding environment (550 kg ha−1). Secondly, to assess the reliability of germplasm performance under rainfed conditions and limited water situations, clarifying whether germplasm selected under very dry rainfed conditions has the ability to produce high lint yield in seasons with higher than average rainfall. It was hypothesised that applying one furrow irrigation in very dry seasons will reduce within experiment variability and increase experimental yields to better reflect our breeding target environment. As irrigation timing will impact its efficacy, the OZCOT simulation model for cotton crop management was used to determine the most effective irrigation date with respect to soil water content and crop growth stage. The simulation, conducted with weather data from a 151 year period, concluded that yields > 550 kg ha−1 were not achieved in 27 seasons (18%). These 27 seasons underwent further simulations to determine the most suitable soil water content and crop growth stage where yield was increased with a single irrigation. It was determined that an irrigation should be applied to a ‘rainfed’ experiment if plant available water (PAW) reached approximately 40% by peak flowering (100-110 DAS). Paired rainfed and MSE experiments with 21 genotypes were established in 2013/14, 2014/15 and 2015/ 16 to provide field validation of the developed protocol. Genotype performance was assessed in terms of lint yield and fibre quality. Results show that in dry seasons (2013/14) irrigating the ‘rainfed’ treatment was necessary to reduce within experiment variability and increase yields above 550 kg ha−1. However, once rainfed yield levels increase due to greater in-crop rainfall (2014/15 and 2015/16), irrigation was no longer necessary. This was further supported by the result that genotype yield ranking differed between rainfed and MSE treatments. Genotype changes in fibre quality between treatments were small. It was concluded that a MSE, designed to produce experimental data better matched to our breeding target environment as well as reducing the risk of experimental failure, would be a worthwhile addition to rainfed evaluations conducted in variable rainfall environments.

1. Introduction Australian rainfed cotton production occurs in warm regions with intermittent and highly variable rainfall throughout the growing season. Not only is the rainfall variable, it is also skewed such that a

smaller proportion of seasons have favourable rainfall patterns (Fig. 1). This environmental variability significantly impacts the area of rainfed cotton production, as well as crop yields (Fig. 2). Rainfed cotton production relies on the use of stored soil moisture as well as in-season rainfall, so production is targeted to long-season regions where soils

Abbreviations: ACRI, Australian Cotton Research Institute; Avg., average; CSIRO, Commonwealth Scientific and Industrial Research Organisation; CV, coefficient of variation; ETO, grass reference crop evapotranspiration; DAS, days after sowing; GxE, genotype-by-environment interaction; HVI, High Volume Instrument; MSE, managed stress environment; NAM, neutron attenuation metre; PAW, plant available water; Ta, air temperature; Tmax, maximum air temperature; v.r., variance ratio ⁎ Corresponding author. E-mail address: [email protected] (W.C. Conaty). http://dx.doi.org/10.1016/j.fcr.2017.10.012 Received 24 April 2017; Received in revised form 10 October 2017; Accepted 13 October 2017 0378-4290/ Crown Copyright © 2017 Published by Elsevier B.V. All rights reserved.

Please cite this article as: Conaty, W.C., Field Crops Research (2017), http://dx.doi.org/10.1016/j.fcr.2017.10.012

Field Crops Research xxx (xxxx) xxx–xxx

W.C. Conaty et al.

Fig. 1. (a) Frequency distribution of in-crop rainfall (October to March), where the dashed line represents the mean rainfall; and (b) box plot highlighting monthly variation in rainfall measured between 1885 and 2015 at Wee Waa (15 km west of ACRI). Data sourced from Aust. BOM (2015).

Fig. 2. In-crop rainfall variability alters: (a) the production area of Australian rainfed cotton, and; (b) the average industry wide rainfed lint yield. Data compiled from Australian Cotton Yearbooks (Dowling, 2015).

The CSIRO rainfed cotton breeding program conducts paired irrigated and rainfed germplasm evaluations at its core research site at the Australian Cotton Research Institute (ACRI). Additional data are collected from up to five alternative rainfed and 15 irrigated locations on commercial farms across Australian cotton production regions. An important aspect of the CSIRO rainfed breeding program is that although targeted parental selection for rainfed cultivar development occurs, only intermediate and advanced material is assessed under rainfed conditions, where early generation material is only evaluated under irrigated conditions. This ensures that yield potential is a consideration in the selection of material, as well as yield performance and reliability under water-limited conditions. The ability to conduct off-site evaluations is limited by the seasonal conditions and willingness of collaborators to sow a rainfed crop. It is widely established that an increase in water availability will decrease variability and increase crop yields. Furthermore, under water limited conditions the genetic potential of a genotype may not be expressed, which may limit the ability to resolve statistical genotype differences. This has been observed in the CSIRO rainfed Advanced Line Trials in the period between 1995 and 2014 (Fig. 3a). This effect can be highlighted by the observation that statistical variation as a percentage of the experiment mean rises significantly when rainfed yield is less than 550 kg ha−1 (Fig. 3b). In the period between 1995 and 2014, 20% of seasons at ACRI resulted in rainfed yields < 550 kg ha−1. Historically, these years result in variable experimental data and increased risk of experiment failure. Therefore, rainfed germplasm evaluated in these seasons may not confidently provide information for selection and progression of candidate lines through the rainfed breeding program.

have high water holding capacity (principally Vertosols) and in-season rainfall is more reliable. However, often long-season areas are those regions with increased rainfall variability. Agronomic management practices can reduce the risk associated with this variable in-crop rainfall. ‘Skip-row’ sowing configurations that decrease plant population can increase the soil water availability to individual plants (Bange et al., 2005). Sowing dates are varied according to soil water availability (Hearn, 1995; Bange et al., 2005). Modest nitrogen inputs, to control early season growth, and the use of limited/no-tillage and stubble retention, to increase the capture and storage of soil water, are employed (Bange et al., 2002). Finally, later-maturing cultivars with phenological plasticity, and the okra leaf trait, have been shown to provide benefit in this target environment (Stiller et al., 2004). As a direct result of the variable rainfall environment, rainfed cotton crops are not sown every season. Thus, the proportion of the Australian cotton industry that rainfed cotton represents is dynamic; between 5 and 25%, where sowing decisions are influenced by cotton prices, availability of stored soil moisture and long-range rainfall predictions (Ford and Forrester, 2002). The average lint yield for Australian rainfed cotton systems is approximately 800 kg ha−1, however yield potential can range from less than 300 to over 2000 kg ha−1, primarily depending on seasonal rainfall patterns (Ford and Forrester, 2002). Although it is important for rainfed crops to produce yield in seasons with limited in-season rainfall, rainfed cotton producers tend to only sow crops in more favourable seasons. Furthermore, most profit is made in seasons with higher than average in-crop rainfall. Therefore, in this target environment it may arguably be more important for a genotype to perform under water-limited, but not extreme drought, conditions.

2

Field Crops Research xxx (xxxx) xxx–xxx

W.C. Conaty et al.

Fig. 3. Historical variability in CSIRO rainfed cotton Advanced Line Trials between 1995 and 2014 as highlighted by: (a) Box plot outlining seasonal variation and the spread of genotype means, n.b. trials were not harvested in 2011 and 2013; and (b) Relationship between experiment mean yield and genotype coefficient of variation (CV, %), as described by logarithmic regression (y = if(x > 0, 61.1 −7.28*ln(abs(x)),0), R2 = 0.82, p < 0.001). The three open circles (not included in the regression) are the results of the rainfed experiments from this study.

detailed by Hearn (1994). Briefly, it is a dynamic simulation model of cotton growth, development and yield, where potential yield is the primary modelled output. It uses a daily time-step with growth and development driven by air temperature and intercepted radiation, and modified by soil water and nitrogen status (Bange et al., 2005). Central to the model is the fruit production and survival subroutine (Hearn and Da Roza, 1985), where the rate of fruit production, shedding and growth is determined by carbon supply. Daily carbon supply is estimated from light interception, and a crop level carbon assimilation rate, adjusted for respiration rates. Light interception is estimated using Beer’s Law (Monsi and Saeki, 2005) and leaf area is generated using a correlation between fruiting site production and leaf area (Jackson et al., 1988). The rate of photosynthesis, leaf expansion and fruiting is modified by ‘stress indices’ which scale the rate of these processes in relation to nitrogen supply and water availability, including the effects of waterlogging. The moisture extraction routine is based on increasing supply with increasing extraction depth over time, using the Ritchie water balance approach (Ritchie, 1972). The model requires site related (weather data, soil characteristics and starting soil moisture), agronomic (sowing date, row configuration, plant stand, irrigation dates and defoliation date), and genotype inputs. The model does not account for the effects of insect pests, diseases, weeds, management failures and soil nutrient limitations other than nitrogen. The model also does not simulate the effects of climate and management on fibre quality, which is an important consideration in rainfed cotton production. The MSE protocol aimed to apply a single furrow irrigation in seasons with high experimental variability, where yields fall below a minimum threshold lint yield (550 kg ha−1). It is hypothesised that this will ensure useful rainfed germplasm evaluations could be carried out every season. As the timing of this irrigation may have a significant impact on its efficacy, OZCOT simulations were run to optimise its timing, with respect to soil water content and crop phenology. These simulations used the cultivar Sicot 71BRF and unless specified, a sowing date of 15th Oct. This date was selected as it is the nominal target date for sowing cotton in the study region. The ability of the OZCOT simulation model to accurately simulate rainfed crop yield was confirmed using historic experiment data collected from ACRI between 1994 and 2011 (n.b. the sowing date was altered to reflect actual sowing dates in these simulations). The correlation between observed and simulated yields was conducted on rainfed experiment mean yield data, and data generated from the OZCOT simulation model (based on Sicot 71BRF). One data point from 2009 (represented by an open circle in Fig. 4) was excluded as in that instance OZCOT incorrectly simulated significant waterlogging induced fruit abscission during early flowering. As the simulation results correlated well with observed results, a further OZCOT simulation was conducted using the available historic

This loss of confidence in data can significantly impact the timeliness and/or effectiveness of rainfed cultivar development. Therefore, a minimum threshold lint yield associated with unacceptable experimental variability has been identified as 550 kg ha−1. To the best of our knowledge, no studies have published the use of managed water stress conditions for cotton germplasm development under limited water scenarios, particularly in the context of ensuring experimental results accurately reflect the breeding target environment by realising minimum yield thresholds. However, studies in other field crops have identified the potential advantages of managed water stress screening to assess the drought tolerance of germplasm. Verulkar et al. (2010) report the IRRI-India Drought Breeding Network, which aims to develop drought-tolerant rice cultivars by assessing grain yield at numerous locations where crops were grown under moderate and severe water stress, as well as fully irrigated conditions. They conclude that a detailed understanding of the target environment, the use of diverse sources of germplasm and direct selection for grain yield under irrigated and drought situations lead to the development of lines that combine high yield potential and improved drought tolerance. Our study hypothesised that applying a single furrow irrigation in very dry seasons would allow the expression of genetic yield variability better suited to our breeding target environment. The aim of this research was twofold: Firstly, to develop and evaluate a managed stress environment protocol where water is applied to ‘rainfed’ germplasm evaluations when crop yield is expected to fall below the threshold for conducting selections with confidence. Secondly, to assess the reliability of germplasm performance under very dry rainfed conditions and limited water situations, with respect to in-crop rainfall. This will clarify whether germplasm selected under rainfed conditions has the ability to respond to more favourable water conditions and produce high lint yield in seasons with above average rainfall. The study used a combination of crop simulation modelling and field validation experiments to develop and test a managed stress environment (MSE) protocol. This study is important as a breeding experiment that does not reflect its target environment and/or provide confident selection decisions is a wasted effort.

2. Materials and methods 2.1. Development of a rainfed breeding protocol using the OZCOT crop simulation model The OZCOT simulation model was employed to develop a managed stress environment (MSE) protocol. The model has been shown to accurately simulate cotton yield, with respect to nitrogen availability, sowing date, rainfed production and row configuration (Hearn, 1994; Bange et al., 2005). The structure and validation of the OZCOT model is 3

Field Crops Research xxx (xxxx) xxx–xxx

W.C. Conaty et al.

Table 1 Weather observed from sowing to crop maturity in the 2013/14, 2014/15 and 2015/16 cotton seasons.

Avg. maximum Ta (°C) Avg. minimum Ta (°C) Tmax > 35 °C (d) Total rainfall (mm) Days to maturity (d) Avg. solar radiation (MJ m−2 d−1) Grass reference ETO (mm)

2013/14

2014/15

2015/16

33.8 16.8 62 127 154 29.4 1011

34.2 17.9 64 199 147 28.3 933

33.7 17.9 61 350 149 27.2 895

Resolvable row-column designs derived from CycDesigN software, with four replicates were used in the MSE validation experiments (Whitaker et al., 2002; Liu et al., 2015). Paired rainfed and MSE experiments were conducted side-by-side. Lateral movement of irrigation from the MSE to the rainfed treatment was minimised by a 6 m gap of bare earth between the rainfed and MSE treatments, as well as two rows of buffer cotton on the margins of each treatment. Management for all field experiments followed current high-input commercial practices with respect to weed and insect control (Cotton Research and Development Corporation, 2016). Each experiment was managed according to its individual requirements, with all plots receiving the same management regime, except for the irrigation treatments imposed. The only exception was the date of the first defoliation, which differed across irrigation treatments where rainfed treatments had earlier maturity. At approximately 60% open bolls, crops were defoliated with thidiazuron, and mature un-opened bolls were conditioned with ethephon. A second application was applied 14–20 d later.

Fig. 4. Correlation between lint yields observed in experiments at ACRI and simulated using the OZCOT crop simulation model between 1994 and 2011 (y = 0.85x + 117.8, R2 = 0.6, p < 0.001). The open circle data point was excluded from the regression as OZCOT under predicted crop yield due to waterlogging and early crop cut-out.

weather data from 1861 to 2011 (151 y). This weather data was collected in Wee Waa, 15 km west of ACRI. It was observed that the simulated crops yields fell below 550 kg lint ha−1 in 18% of seasons (27 y). Using the weather data from these 27 seasons, further OZCOT simulations were conducted to determine the most effective timing of a single irrigation to increase the simulated crop yield above 550 kg lint ha−1. This was achieved by applying rules to the crop simulations where an irrigation was applied when PAW reached 51, 39, 27 and 15% PAW, by 61, 76, 92, 107, and 123 days after sowing (DAS).

2.2.2. Genotype selection Twenty-one genotypes were used in field experiments (Table 2). All of the genotypes studied were developed by the Commonwealth Scientific and Industrial Research Organisation (CSIRO). The genotypes studied included both nominal rainfed and irrigated, historic and current, as well as conventional and transgenic cultivars or genotypes. The transgenic genotypes all expressed Bt toxins for resistance to Helicoverpa spp. larvae damage (i.e. Bollgard II®Ô producing the Monsanto Cry1Ac and Cry2Ab proteins) as well as full season tolerance to the glyphosate (i.e. RoundupReady Flex®™) family of herbicides (Monsanto Australia Ltd., 2012a,b). As the cultivar Sicot 71BRF was used in the OZCOT simulations, it was selected as the control genotype in the field

2.2. Field validation of the developed protocol 2.2.1. Site description, experimental design and crop management Experiments were established in the summer growing seasons of 2013/14, 2014/15 and 2015/16 at ACRI (30° 12′S, 149° 36′E), 22 km north-west of Narrabri NSW, Australia. Experiments were sown on Oct. 15th 2013, Oct. 20th 2014 and Oct. 31st 2015. Experiments were sown following an 11 month fallow period which was preceded by a winter wheat crop. A row spacing of 1 m was used with a sowing density of 10–12 plants m−2. A single skip row sowing configuration was used (two rows in, 1 row out). Therefore, each plot consisted of two 13 m rows of cotton, with an adjacent empty row on each side of the plot. Nitrogen was applied as anhydrous ammonia approximately 12 weeks before sowing at a rate of 90 kg N ha−1. Experiments started with a full moisture profile supplied via furrow irrigation. The MSE treatment was furrow irrigated provided the threshold soil water content and crop stage were observed. The study region is semi-arid, characterised by mild winters, hot summers and summer-dominant rainfall patterns, with an annual average precipitation of 646 mm (Aust. BOM, 2015). The soil of the site is a uniform grey cracking clay (USDA soil taxonomy: Typic Haplustert; Australian soil taxonomy: Grey Vertosol). Plant available soil water to 1.2 m at the site is between 160 and 180 mm (Tennakoon and Hulugalle, 2006). Weather data from sowing to maturity in 2013/14, 2014/15, and 2015/16 was monitored with a weather station using the recommendations of the ASAE (Allen et al., 2005). The weather station was located above clipped grass, 2.5 km from the field experiments. Weather data are presented in Table 1.

Table 2 Genotypes used in the managed stress environment (MSE) experiment.

4

Cultivar/genotype

Release date

Nominal Rainfed

Siokra 1–4 CS 50 Siokra L23 Sicala V-2 Sicot 189 Siokra V-16 Sicot 71 Siokra 24 Sicala 60BRF Sicot 80BRF Sicot 81 Sicot 75 Sicot 71BRF Siokra V-18BRF Sicala 340BRF Sicot 74BRF Siokra 24BRF Sicot 75BRF Sicot 730 CSX8521BRF CSX2027BRF

1988 1992 1993 1994 1996 1998 2002 2004 2006 2006 2005 2007 2008 2008 2010 2010 2010 2011 2012 Breeding line Breeding line

Yes Yes

Yes Yes Yes Yes

Yes Yes Yes

Yes

Reference

Reid (1992a) Reid (1992b) Reid (1995) Reid (1996) Reid (1998) Reid (2003) Stiller and Reid (2005) Stiller (2007a) Stiller (2007b) Stiller (2007c) Stiller (2008b) Stiller (2008a) Stiller (2010c) Stiller Stiller Stiller Stiller

(2010a) (2010b) (2011) and Reid (2013)

Field Crops Research xxx (xxxx) xxx–xxx

W.C. Conaty et al.

validation studies. 2.2.3. Field measurements Soil water was measured to a depth of 1.8 m via neutron attenuation in all plots of the control genotype (Sicot 71BRF). Access tubes were installed in the plant line after establishment. The soil profile was measured at 0.1 m intervals in the top 0.6 m of soil and at 0.2 m intervals below 0.6 m using a neutron attenuation metre (NAM) (503DR Hydroprobe®, CPN Corporation, Martinez CA, USA). Measurements were conducted between 30 and 150 DAS, on approximately a weekly schedule. Where possible, soil water was measured before predicted rainfall events, and was measured 48 h after rainfall events > 10 mm. Soil water was measured on the day of or the day before an irrigation event in the MSE treatment, and again 48 h after the irrigation. A calibration developed for the same field for the NAM (data not shown), was used to monitor the soil water during the season. The NAM was calibrated using the methodology of Hodgson and Chan (1987). Plant available water (PAW, %) was calculated on the basis of a soil moisture holding capacity of 521 mm, and a soil moisture at permanent wilting point of 357 mm, equating to a plant available water of 164 mm (Tennakoon and Hulugalle, 2006). One row of seed cotton for each plot was mechanically harvested with a spindle picker (modified Case International 1822) and weighed. The gin turn-out (% lint of seed cotton) was determined from a 300 g sub-sample of the seed cotton that was ginned in a 20 saw gin with a pre-cleaner (Continental Eagle, Prattville, AL U.S.A.), and was subsequently used to calculate lint yield (kg ha−1). Lint samples were collected and analysed for fibre quality using a spinlab High Volume Instrument (HVI) model 1000 (Uster Technologies AG, Uster, Switzerland).

Fig. 5. The average simulated lint yield for the 27 seasons where the rainfed simulated lint yield was below 550 kg ha−1 for each soil water content (% plant available water, PAW) and irrigation date (days after sowing, DAS) combination.

reached in all 27 dry years in the study (Table 3). The 51% and 39% PAW simulations at 107 DAS called for an irrigation in all the dry seasons of the study, and was able to recover simulated yields above 550 kg lint ha−1 in all but two seasons. This represented 1% of all the seasons in the 151 year dataset. It is of note that the simulated yield in these two particular seasons was not increased above 550 kg lint ha−1 in any date-by-PAW combination. The drier soil water contents (27% and 15% PAW) did not schedule an irrigation in all the dry seasons. Thus, of the soil water contents and irrigation dates studied, the 39% at 107 DAS was determined to be the most suitable combination. In reality, irrigating on a fixed soil water content and date combination may not always be feasible. Therefore, based on the results shown in Fig. 5 and Table 3 we concluded that the target soil water content should be 45-39% PAW, with a target irrigation date of 100–110 DAS.

2.2.4. Statistical analysis Initially, residual maximum likelihood (REML) was conducted independently on each season’s rainfed and MSE dataset, determining if genotype differences could be resolved within an experiment. REML on individual experiments was conducted using GenStat 16th ed. (VSN International, Hemel Hempstead, UK). Data from within a season was then pooled to assess within season GxE interactions. All available data was subsequently pooled to investigate genotype, season and water treatment main effects and interactions. These subsequent analyses were conducted in ASReml-R (Butler et al., 2009). As the MSE treatment was not applied in the 2015 season, the pooled dataset assessed main effects and interactions in the orthogonal two-year dataset (2013/ 14 and 2014/15 seasons). All analyses accounted for the spatial effects of the field layout (row and column) and differences were assessed at the 95% confidence level.

3.2. Field validation of rainfed breeding protocol 3.2.1. Soil water content Soil water content, expressed as PAW, was measured via neutron attenuation from the seedling stage of crop development to crop maturity (ca. 30–150 DAS) (Fig. 6). Soil profiles were characterised by gradual drying as the crop utilised stored soil moisture. Decreases in soil water content following rainfall were marginal, except for the 2015/16 season where soil water content was maintained above 75% PAW due to substantial in crop rainfall until after 100 DAS (Table 1 and Fig. 6c). Thus, the MSE treatment was only irrigated in the 2013/14 and 2014/15 seasons (Fig. 6a and b). These irrigations occurred at a soil water contents of 45% PAW at 96 DAS in the 2013/14 season, and 44% PAW at 99 DAS in the 2014/15 season. The soil water returned to near field capacity following the irrigations. High rates of water extraction in the MSE following the irrigation resulted in differences between the rainfed and MSE treatments for only 26 and 20 days in the 2013/14 and 2015/16 seasons, respectively. Treatment differences were not observed in the 2015/16 season, with the exception of at 152 DAS when the crop had reached maturity.

3. Results 3.1. Crop simulation modelling to develop a rainfed breeding protocol The results of the validation of the OZCOT simulation model, based on observed rainfed experiment means at ACRI between 1994 and 2011 are shown in Fig. 4. The simulation results correlated with observed actual experimental data. The regression of the observed and simulated yield data did not differ from the 1:1 line (p = 0.426), and approximately 60% of the variation in observed yield data reflected in the simulated yield data. The results of the simulations used to determine the optimal time for a single furrow irrigation in the rainfed experiments that yielded below 550 kg lint ha−1, with respect to timing and the soil water content at irrigation, is shown in Fig. 5 and Table 3. There was no increase in average yield at any soil water content at early and late irrigation datesbefore 92 DAS and after 107 DAS. The 51% PAW at 92 DAS irrigation threshold produced the highest average simulated yield. However, this PAW-date combination is unsuitable as the trigger point was not

3.2.2. Lint yield and quality With 127 mm of rainfall, the 2013/14 season was the driest of the three season studied. Under this environment average rainfed yield was 5

Field Crops Research xxx (xxxx) xxx–xxx

W.C. Conaty et al.

Table 3 Results of the 15th Oct. sowing date OZCOT simulations designed to determine the most effective irrigation date, with respect to soil water content. Simulations were conducted in the 27 seasons where simulated crop yield was less than 550 kg lint ha−1. The table outlines the number of seasons an irrigation is scheduled for each irrigation date-by-soil water content combination, as well as the average yield increase, the number of failed crops, and the number of crops that failed following an irrigation. Irrigation Date (DAS)

Plant available water (%)

Seasons with irrigation (maximum 27)

Avg. yield increase (%)

Seasons with failed crop (< 550 kg ha−1)

Seasons with failed crop when irrigated

Avg. simulated yield (kg ha−1)

61 76 92 107 123

51 51 51 51 51

0 3 26 27 12

– 155 212 186 118

27 24 1 2 21

0 1 1 2 6

458 485 957 841 500

61 76 92 107 123

39 39 39 39 39

0 0 22 27 12

– – 213 186 124

27 27 5 2 21

0 0 0 2 7

458 458 884 841 500

61 76 92 107 123

27 27 27 27 27

0 0 13 26 12

– – 214 189 115

27 27 14 2 21

0 0 0 1 6

458 458 714 840 500

61 76 92 107 123

15 15 15 15 15

0 0 2 24 12

– – 189 188 87

27 27 25 4 21

0 0 0 1 6

458 458 489 805 500

The 39% PAW at 107 DAS (shown in bold) was the optimal combination as it had the highest recovery rate of the combinations that called for an irrigation in all dry seasons.

546 kg ha−1 (Table 4). This value was below the minimum yield for our breeding target environment and is associated with a high level of statistical variation as a percent of the experimental mean (Fig. 3). Rainfed yields were significantly lower than that of the MSE treatment, which yielded 1120 kg ha−1 (Table 4). The coefficient of variation in the rainfed experiment (16.4%) was much higher than under MSE conditions (9.9%). Despite this observation, spatial analysis resolved genotype differences in both rainfed and MSE treatments (rainfed and MSE p < 0.001). Importantly, GxE interactions within a year were observed when treatments were pooled (p = 0.029; Table 4 and Fig. 7). This interaction was predominantly driven by the relatively better performance of Sicot 80BRF and Sicala 340BRF under rainfed conditions and to a lesser extent the relatively better performance of Sicot 75 under MSE conditions. The 2014/15 season received 199 mm of in-crop rainfall. Average rainfed yield was 785 kg ha−1, compared to MSE yield of 1324 kg ha−1 (Table 4). Again, statistical resolution of genotype was possible under both treatments (rainfed p = 0.001; MSE p = 0.009) and the coefficient of variation was reduced under MSE conditions (7.8%) when compared with rainfed conditions (11.0%). Genotype rankings under rainfed and MSE conditions differed (GxE interaction present; p = 0.003; Table 4 and Fig. 7). These differences were observed in a larger number of genotypes than in 2013/14. Although to a lesser extent, Sicot 80BRF’s pattern of performance under rainfed and MSE conditions was similar to the 2013/14 season. However, the pattern was not observed in Sioct 75 and Sicala 340BRF. Additionally, the cultivars Siokra 24BRF, Sicot 71BRF, Sicala V-2 and Sicot 71 performed relatively better under rainfed when compared to MSE conditions. Conversely, Siokra L23, Sicot 189 and Sicot 75BRF performed relatively better under MSE conditions over rainfed conditions. The 2015/16 season received the highest volume of in-crop rainfall. The vast majority of the 350 mm received occurred up to peak flowering, where soil water content did not fall below 75% PAW until after 100 DAS (Fig. 6). As a result, the MSE treatment was not irrigated. Average rainfed coefficient of variation was similar to that observed in 2013/14 and 2014/15 (14.1%), and lint yield was the highest across all three seasons in the study, 823 kg ha−1. Statistical resolution between genotypes was again observed (p = 0.004) (Table 4). As expected, since no treatment was imposed, there was no difference in genotype

rankings between the rainfed and unwatered MSE treatments (p = 0.309, data not shown). When data was pooled across the two orthogonal seasons (2013/14 and 2014/15) large differences in season (p < 0.001), water treatment (p < 0.001) and genotype yield were observed (p < 0.001). Genotype average lint yield ranged from 833 kg ha−1 in the cultivar Sicala 340BRF to 1009 kg ha−1 in Sicot 75BRF (Fig. 8c). An interaction between season-by-genotype (p < 0.001) and genotype and water treatment (GxE, p = 0.013) was observed (Table 5). The GxE interaction observed over the two year dataset was primarily driven by the relatively better performance of Sicot 80BRF under rainfed conditions, as well as Siokra L23 performing relatively better under MSE conditions over rainfed conditions (Fig. 7). Although interactions with genotype were present in the data, these contributed to significantly less of the variance in the data than genotype, season or water treatment main effects (Table 5). Despite the presence of the third order interaction of genotype-by-environment-by-season (p = 0.007), the vast majority (> 96%) of the variation in the data was explained by the main effects of season and water treatment (Table 5, Fig. 8). Genotype (p < 0.001) was the only main effect that altered gin turnout (Table 5 and Fig. 9a–c). Average gin turnout was lowest in CSX2027 at 37.3% and highest in Sicot 730 at 42.0%. GxE (p < 0.001), genotype-by-season (p < 0.001) and water-by-season (p = 0.017) interactions were also observed. Despite the occurrence of these interactions, the variance ratio associated with these interactions was only larger than the main effect of genotype in the water-by-season interaction (Table 5) where compared with the MSE treatment, the rainfed had a lower gin turnout in 2013/14, and a higher gin turnout in 2014/15. Season (p = 0.001), water treatment (p = 0.039) and genotype (p < 0.001) all had a significant effect on fibre length (Fig. 9d–f). The shortest fibre length was observed in Sicot 71 (27.5 mm) and longest in Sicala 340BRF (30.3 mm). GxE (p = 0.03), genotype-by-season (p < 0.001) and water-by-season (p = 0.019) interactions were observed. The largest portion (79%) of the variance ratio was attributed to the water-by-season interaction; which was 73 times larger than the portion attributed to genotype (Table 5). This water-by-season interaction was driven by a shorter fibre length in the rainfed treatment in 2013/14, while the rainfed treatment had a higher fibre length in 6

Field Crops Research xxx (xxxx) xxx–xxx

W.C. Conaty et al.

(p < 0.001) (Table 5). This interaction was characterised by stronger fibres in the MSE in 2013/14, while stronger fibres were observed in the rainfed treatment in 2014/15. Significant effects of season (p < 0.001), water treatment (p = 0.007) and genotype (p < 0.001) were observed in micronaire. The interaction between season and genotype (p < 0.001) as well as water and genotype (p = 0.033) were significant. However they only explained a small portion of the variance in the data. The majority of the variance in the data was explained by the effect of season (Table 5). 4. Discussion 4.1. Crop simulation modelling to develop a rainfed breeding protocol The OZCOT simulation model (Hearn, 1994) accurately reflected observed lint yield data from the CSIRO rainfed breeding program. Similar to the results of previous studies (Milroy et al., 2004; Bange et al., 2005), the observed and simulated yield data did not differ significantly from the 1:1 line (p = 0.426) (Fig. 4). Simulations where rainfed yield was below 550 kg lint ha−1 were used to determine the optimal timing of a single furrow irrigation in the rainfed experiments, with respect to crop phenology and the soil water content at irrigation. Based on these simulations, an irrigation trigger point of 45–39% PAW by 100-110 DAS was identified (Fig. 5 and Table 3). This irrigation trigger point was deemed the most favourable as it was the latest irrigation date that corresponded with: Firstly, an irrigation call in all 27 seasons where simulated yield < 550 kg ha−1; secondly, the lowest number of crops that following an irrigation yielded < 550 kg ha−1. A later irrigation date is more desirable as this increases the possibility of capturing more in-season rainfall before an irrigation is required. Although earlier irrigation trigger points resulted in higher simulated yields, irrigation thresholds were not reached in all of the seasons where simulated yield was below the threshold for acceptable within experiment variability (550 kg ha−1). Likewise, later irrigation trigger points were not reached in all of the study seasons. If a late season irrigation trigger point was reached it had a reduced impact on crop performance as season length tended to limit the ability of the crop to utilise the irrigation water. The impact of water stress on cotton yield is most pronounced during flowering because boll size and survival are reduced as well as vegetative compensation (Grimes et al., 1970; Hearn and Constable, 1984). Therefore, it is expected that the largest impact of the single irrigation was observed when crops were entering severe water stress during peak flowering. 4.1.1. Simulations for all seasons Although the developed MSE protocol was successful at recovering almost all low-yielding crops (Table 3), it is also important to assess the protocol in all the available seasons of weather data. An irrigation to rescue a crop was only required in 18% of seasons, but the MSE protocol schedules an irrigation in 65% of seasons. As such, in 46% of seasons the protocol advises an irrigation when there is sufficient subsequent rainfall to increase crop yield above 550 kg ha−1. This is noteworthy as a MSE crop in these seasons may result in altered genotype rankings when compared with rainfed crops (Table 4). There are numerous ways to mitigate this risk. Firstly, conduct rainfed, MSE and irrigated germplasm evaluations at all potential MSE sites. In seasons where a MSE irrigation was not necessary the number of rainfed replicate samples would double, presumably reducing the inherently high coefficient of variation observed in rainfed experiments. In seasons where a MSE irrigation was necessary, post-season analysis in the context of seasonal rainfall could be used to determine the value of both the rainfed and MSE yield results. Such an approach could substantially increase resources required by the breeding program. Secondly, a sensible approach to applying a MSE irrigation with respect to short and long-term rainfall forecasts may reduce the number of seasons that a MSE irrigation is unnecessarily applied. This approach is obviously

Fig. 6. Soil water content expressed as plant available water (%) measured via neutron attenuation in the control genotype (Sicot 71BRF) grown under both rainfed and managed stress environment (MSE) water treatments in the (a) 2013/14 season, (b) 2014/15 season, and (c) 2015/16 season. An * represents a date where differences between treatments was measured.

2014/15. This may be the result of incomplete boll development in a portion of the additional fruit produced in response to the MSE irrigation, skewing treatment means. It is speculated that this may have occurred in the 2014 season as the rate of crop water use following the MSE irrigation was more rapid in 2014 (Fig. 6). Regardless, the variance ratio of the GxE and genotype-by-season interactions was lower than all main effects, and insignificant compared to the season and water-by-season effects (Table 5). Fibre strength was affected by season (p = 0.015), water treatment (p = 0.006) and genotype (p < 0.001). The weakest fibres were observed in Sicot 71 (285.7 KN Tex−1) and strongest in Sicala 340BRF (308.8 KN Tex−1) (Fig. 9i). Although a season-by-genotype interaction was observed (p < 0.001), a large proportion in the variance of the data (82%) was explained by the water-by-season interaction 7

Field Crops Research xxx (xxxx) xxx–xxx

W.C. Conaty et al.

note that this date represents the same calendar date as the base scenario (30th Jan.). Results of this additional simulation are presented in Sup. Table 1 which shows that delaying sowing by just two weeks will increase yield and reduce the probability of experimental failure (i.e. mean yield < 550 kg ha−1), from 18% to just over 6% of seasons. This result is supported by agronomic studies (conducted at the same site as our study) which have shown that lint yield and water-use efficiency are not altered when sowing date is extended more than 30 d from the nominal target sowing date of 15 October (Braunack et al., 2012; Bange, 2016). However, while this approach may reduce the risk of experimental failure the potential effects of limited expression of genetic potential, with respect to both yield and fibre quality in a truncated season length, may alter genotype selections. Future studies should investigate the use of delayed sowing to mitigate the risk of experimental failure, and importantly the potential effects on genotype selection. However, these studies must be conducted in the context that commercial sowing decisions are largely dictated by stored soil moisture and rainfall.

Table 4 Summary of lint yield results for the 2013/14, 2014/15 and 2015/16 seasons, outlining in-crop rainfall, average lint yield, genotype p-value and the coefficient of variation (CV, %) for each rainfed and managed stress environment (MSE) experiment, as well as genotype-by-water treatment (GxE) interaction probabilities. Season

Water treatment

In-crop rainfall (mm)

Predicted lint yield (kg ha−1)

Genotype p-value

CV (%)

GxE pvalue

2013/14

Rainfed MSE

127

546 1120

< 0.001 < 0.001

16.4 9.9

0.029

2014/15

Rainfed MSE

199

783 1323

< 0.001 0.009

11.0 7.8

0.003

2015/16

Rainfed MSE

350

823 n/a

0.004 n/a

14.1 n/a

n/a

associated with more risk as rainfall forecasts are not always reliable. Finally, GxE interactions can also be managed and assessed with greater confidence by increasing the number of germplasm evaluation sites.

4.2. Field validation of rainfed breeding protocol

4.1.2. Sowing date As sowing was delayed in the 2015/16 field validation experiment until 31st Oct., OZCOT simulations were repeated with an adjusted sowing date of 31st Oct. Under this scenario simulated yield was below 550 kg lint ha−1 in only 10 seasons. Again multiple simulations, where an irrigation was applied when soil water contents reached 51, 39, 27, 15% PAW by 45, 60, 76, 91 and 107 DAS, were conducted to determine the most effective irrigation date. The simulation indicated an irrigation should also be applied when soil water content reached 38% PAW, but with a target irrigation date approximately two weeks earlier. It is of

4.2.1. Lint yield Crop yield increased while experimental error decreased with water availability; yield was consistently higher and reduced coefficient of variation (CV) under MSE conditions were observed (Table 4). While it is outside of the scope of this study to determine why this phenomenon occurs, it is possible that random cracking of the heavy clay soils was responsible for variable soil moisture availability within the relatively small 3 m × 13 m plots. Fig. 7. Rainfed and MSE lint yield GxE interactions: Scatter plots showing relationship between genotype lint yield performance under rainfed and MSE conditions in 2013 (a), 2014 (b) and (c) two-year genotype average. Deviation from mean performance under rainfed conditions compared with performance under MSE conditions in 2013 (d), 2014 (e) and, (f) two-year genotype average, where values greater than zero indicate relatively better MSE performance. Yield deviation is the difference between actual and fitted relationship as show in figure panes a, b or c. The order of genotypes in figure panes d, e and f follows the ranking of the two-season mean MSE lint yields.

8

Field Crops Research xxx (xxxx) xxx–xxx

W.C. Conaty et al.

selections may be more reflective of yield potential, rather than performance under extreme water deficit conditions. However, with a range of sites within a season, one site managed with MSE will ensure at least one site provides reliable data. Without one MSE site, the worst case scenario would be all sites expressing low lint yield outside our breeding target environment and/or no statistically significant genotypic resolution in yield data. Future studies, which include side-by-side comparison of rainfed, MSE and fully irrigated environments, need to be conducted to further dissect the possible implications of a MSE on genotype selection. However, as rainfed crops tend to be sown in seasons with favourable conditions, and grower income is greater under seasons with increased rainfall, we suggest that genotype performance under MSE may actually identify better and more realistic long term options for growers. Future studies are required to test this hypothesis. The use of a MSE significantly increased lint yield above our target breeding environment threshold, but also resulted in GxE or genotypeby-environment-by-season interactions (Fig. 7c and Table 5). Altered genotype performance across water treatment and seasons is presumably due to treatment and seasonal differences in fruit setting patterns in response to the water dynamics. In the two-year pooled dataset, the interaction is largely driven by the cultivars Sicot 80BRF and Siokra L23 (Reid, 1992b). Sicot 80BRF performed relatively better in terms of genotype rankings under rainfed conditions (Fig. 7b). Sicot 80BRF’s performance under rainfed conditions is probably a result of its selection for performance under our variable rainfall environment. Siokra L23’s MSE lint yield significantly outranked its rainfed lint yield; suggesting this genotype was better able to utilise the irrigation water applied at peak flowering (approximately 100 DAS). Long-season genotypes such as Siokra L23 have a later fruit setting pattern, are more indeterminate and thus are better able to utilise in-crop rainfall during the flowering period (Stiller et al., 2004). These conditions would be similar to those artificially created under a MSE system and complement the finding that later-maturing cultivars with phenological plasticity are desirable under Australian rainfed conditions (Stiller et al., 2004). Of note is the consistency of rankings in the performance of conventional cultivars present in both the Stiller et al. (2004) and the current study. Furthermore, despite the presence of a GxE interaction involving these two genotypes, neither would be used commercially in preference to the modern cultivars. Future research should determine a fruit setting ideotype which can best utilise additional in-crop rainfall in more favourable seasons, and determine if the pattern of fruit setting is reflective of yield performance and fibre quality under MSE conditions. This research should also investigate the effect of higher fruit retention and increased genotype determinacy (e.g. cultivars with transgenic insect resistance) on yield under MSE. 4.2.2. Fibre quality Gin turnout and fibre quality (length, strength and micronaire) were mostly affected by the main effects of season, water treatment and genotype. Interactions between season-by-water, genotype-by-season and GxE were also observed. However, except for the season-by-water interaction observed in gin turnout, fibre length and strength data, the variance ratio associated with the interactions were generally much lower than the main effects (Table 5). For example, much of the variation in data was explained by the season-by-water interaction, where in the case of fibre strength, up to 82% of the variance in the data is explained by this interaction. The observation of widespread and large main effects and interactions in fibre quality data are related to the low coefficient of variation in the datasets. This has been observed in previous studies investigating fibre quality GxE interactions, which outline that despite the presence of numerous interactions, main effects, such as genotype, explain the majority of the variance in the data (Meredith, 1984). When data was pooled across experiments, the mean genotype fibre quality packages observed did not result in commercially significant discounts. Notably,

Fig. 8. Predicted lint yield (kg ha−1) for the main effects of (a) season, (b) water treatment and (c) genotype. Vertical bar represents standard error. 2015 data was excluded from the analysis as the MSE protocol did not call for an irrigation.

Field validation results show that in dry seasons (2013/14) a MSE was necessary to increase yields to our target breeding environment, as well as reducing statistical variation expressed as a percent of the experimental mean, which has been shown to be associated with low trial mean yields (< 550 kg ha−1). However, once rainfed yield levels increase due to an increased in-crop rainfall (2014/15 and 2015/16), irrigation was no longer necessary for experiments to better reflect our target breeding environment. As a GxE interaction was observed in all years a MSE irrigation was applied, it can be concluded that a MSE is necessary in very dry years to ensure selections will favour those lines with better yield potential under our target environment. These 9

Field Crops Research xxx (xxxx) xxx–xxx

W.C. Conaty et al.

Table 5 Lint yield wald statistics and fibre quality variance ratios (v.r.) and p-values for the main effects of season (S), water treatment (W, i.e. rainfed or MSE) and genotype (G), as well as all interactions. Data is from the orthogonal 2 year (2013/14 and 2014/15) dataset. * = p ≤ 0.05, ** = p ≤ 0.01, *** = p ≤ 0.001, n.s.=non-significant. Source of variation

Lint yield (kg ha−1)

Gin turnout (%)

Fibre length (mm)

Fibre strength (KN Tex−1)

Micronaire

Wald stat

p-value

v.r.

p-value

v.r.

p-value

v.r.

p-value

v.r.

p-value

Season (S)

7534

***

0.4

n.s.

165.6

**

4.7

n.s.

5318.5

***

Water (W) Season.Water

459 0.2

*** n.s.

3.2 22.0

n.s. **

7.0 736.7

* ***

17.7 159.6

** ***

16.3 4.7

** n.s.

Genotype (G) Season.Genotype Water.Genotype S.W.G

156 70 37 39

*** ** * **

5.8 3.3 1.4 1.9

*** *** n.s. **

10.1 6.1 1.7 1.8

*** *** * *

3.8 2.7 1.1 0.9

*** *** n.s. n.s.

9.0 6.5 1.7 2.1

*** *** * **

modern cultivars had improved fibre quality packages than older cultivars (Table 2). The results of the present study suggest that while GxE and season-by-genotype interactions have been observed in the fibre quality data set, these interactions are minimal when compared to main effects and season-by-water interaction. Thus, the use of a MSE would have limited impact on fibre quality selections made on the basis of genotype fibre quality performance under rainfed or MSE scenarios. Future detailed studies investigating the within plant distribution to fibre quality on a node and boll position basis may further dissect the response of fibre quality to MSE conditions.

presumably also due to improvements in traits that are mutually beneficial under wet and dry environments (e.g. altered harvest indices, disease and pest resistance, nutrient use efficiency and stay green). However, if these traits are optimised and/or are no longer the major limitation crop performance, yield gains under water-limited scenarios may decline if selections are made purely on the basis of yield potential and yield reliability. A further caveat to producing drought adapted cultivars by breeding for increased yield potential and reliability under well-watered conditions is that the approach should be limited to scenarios where drought stress occurs irregularly and results in maximum yield reductions of 50% of an irrigated control. Outside of this scenario, it has been concluded that a specialised breeding program for dry environments may be necessary (multiple crops reviewed by Blum (2005), rye (Haffke et al., 2015) and rice (Pantuwan et al., 2002)). Notably, the Australian rainfed cotton industry can be characterised by its highly variable rainfall environment (Fig. 1); and historically, one-in-five seasons the CSIRO rainfed germplasm evaluations resulted in very low yields (< 550 kg ha−1). Furthermore, the reduction in yield between rainfed and MSE water treatments in these experiments was 42% (Fig. 8), and a reduction of about 50% and 75% was observed when MSE and rainfed yield were contrasted to comparable irrigated experiments, respectively (irrigated lint yield = 2608 kg ha−1, data not shown).

4.2.3. Breeding for water-limited environments Although there is limited literature available in cotton, in other crop species it is widely accepted that yield under water-limited scenarios is polygenic and has low heritability; hence it is difficult to select for superior germplasm (Turner et al., 2014). Studies undertaken in numerous crop species (cereals, sugar beet, maize and soybean) conclude that yield in water-limited environments is mainly determined by inherent yield potential (reviewed by Cattivelli et al. (2008)), thus maximising productivity in the absence of stress can result in yield improvements under moderate and mild drought conditions (Slafer et al., 2005; Tambussi et al., 2005; Cattivelli et al., 2008). This concurrent increase in yield under optimal and sub-optimal conditions is

Fig. 9. Main effects of season, water treatment (rainfed and managed stress environment, MSE) and genotype on gin turnout (a, b and c), fibre length (d, e and f), fibre strength (g, h and i) and micronaire (j, k and l). Vertical bar represents l.s.d., 2015 data was excluded from the analysis as the MSE protocol did not call for an irrigation.

10

Field Crops Research xxx (xxxx) xxx–xxx

W.C. Conaty et al.

performance across all sites, not just MSE.

Improvements in water-limited yield through selection in limited environments have been reported in maize (Bolanos and Edmeades, 1992; Byrne et al., 1995; Bänziger et al., 1999; Cooper et al., 2014), wheat (Morgan, 2000) and rice (Pinheiro et al., 2000; Lafitte et al., 2006; Ouk et al., 2006; Venuprasad et al., 2007; Kumar et al., 2008; Venuprasad et al., 2008; Verulkar et al., 2010). Broadly, these studies outline that it is possible to make long-term genetic gain for yield under favourable environmental conditions, where yield potential is expressed, and under water-limited scenarios, where traits associated with drought tolerance are expressed. These studies emphasise the importance of side-by-side irrigated and drought experiments, as well as careful evaluation of parents and testing of advanced lines under specific managed drought trials (Pinheiro et al., 2000; Verulkar et al., 2010; Rebetzke et al., 2013; Cooper et al., 2014). These studies used multiple sowing dates and delayed sowing to increase the chances of drought stress at the reproductive stage. Two studies in the literature are of particular note. Firstly, the similarity between the results of the current study and Byrne et al.’s (1995) conclusion from a tropical maize breeding program- that selection under managed levels of drought stress at one location, together with multi-location testing, may be desirable for breeding programs in drought-prone areas. Secondly, the differentiation between the approach of the current study to manage water stress during flowering and that of Verulkar et al. (2010), where experimental data was excluded when yield under managed drought conditions was < 85% or > 30% of irrigated yield. While the reasoning (that high levels of stress limits expression of genetic variability preventing reliable differentiation between drought-tolerant and susceptible breeding lines) is similar, our study actively aimed to avoid severe water stress during flowering and ensure the collection of data better matched to the breeding target environment. This differentiation may be vital for the progression of material through a breeding program because a breeding experiment which does not reflect its target environment or provide significant differences between genotypes is essentially wasted effort.

Acknowledgements This study was financially supported by Cotton Breeding Australia, a joint venture between CSIRO and Cotton Seed Distributors. The authors would like to thank technical staff of CSIRO cotton breeding group for their invaluable contribution to this work, particularly Mark Laird, Mick Price, Adam Suckling, Megan Cameron, Deon Cameron, Jo Price and Kellie Cooper. The authors appreciate the comments provided by Susan Jaconis and Richard Richards, as well as the anonymous reviewers and editor, which improved the manuscript. Appendix A. Supplementary data Supplementary data associated with this article can be found, in the online version, at http://dx.doi.org/10.1016/j.fcr.2017.10.012. References Allen, R.G., Walter, I.A., Elliot, R., Itenfisu, D., Brown, P., Jensen, M.E., Mecham, B., Howell, T.A., Snyder, R., Eching, S., Spofford, T., Hattendorf, M., Martin, D., Cuenca, R.H., Wright, J.L., 2005. The ASCE standardized reference evapotranspiration equation. In: Allen, R.G., Walter, I.A., Elliot, R., Howell, T.A., Itenfisu, D., Jensen, M.E., Mecham, B. (Eds.), ASCE-EWRI Task Comimittee Report, January, 2005. Aust. BOM, 2015. Australian Bureau of Meteorology: Climate Data Online. Commonwealth of Australia Bureau of Meteorology Web. Bänziger, M., Edmeades, G.O., Lafitte, H.R., 1999. Selection for drought tolerance increases maize yields across a range of nitrogen levels. Crop Sci. 39, 1035–1040. Bange, M., Marshall, J., Stiller, W., 2002. Agronomy. In: Schulze, R. (Ed.), Australian Dryland Cotton Production Guide. Cotton Research and Development Corporation, Narrabri, NSW, pp. 41–52. Bange, M.P., Carberry, P.S., Marshall, J., Milroy, S.P., 2005. Row configuration as a tool for managing rain-fed cotton systems: review and simulation analysis. Aust. J. Exp. Agric. 45, 65–77. Bange, M., 2016. Raingrown (dryland) Cotton. Australian Cotton Production Manual, 2016. Cotton Research and Development Coroporation, Narrabri, NSW, pp. 18–20. Blum, A., 2005. Drought resistance, water-use efficiency, and yield potential − are they compatible, dissonant, or mutually exclusive? Australian J. Agric. Res. 56, 1159–1168. Bolanos, J., Edmeades, G.O., 1992. Eight cycles of selection for drought tolerance in lowland tropical maize: I. Responses in grain yield, biomass, and radiation utilization. Field Crop. Res. 31, 233–252. Braunack, M.V., Bange, M.P., Johnston, D.B., 2012. Can planting date and cultivar selection improve resource use efficiency of cotton systems? Field Crop. Res. 137, 1–11. Butler, D.G., Cullis, B., Gilmour, A., Gogel, B., 2009. ASReml-R Reference Manual, Training Series No. QE 02001. The State of Queensland, Department of Primary Industries & Fisheries, Brisbane, Australia. Byrne, P.F., Bolanos, J., Edmeades, G.O., Eaton, D.L., 1995. Gains from selection under drought versus multilocation testing in related tropical maize populations. Crop Sci. 35, 63–69. Cattivelli, L., Rizza, F., Badeck, F.-W., Mazzucotelli, E., Mastrangelo, A.M., Francia, E., Marè, C., Tondelli, A., Stanca, A.M., 2008. Drought tolerance improvement in crop plants: an integrated view from breeding to genomics. Field Crop. Res. 105, 1–14. Cooper, M., Gho, C., Leafgren, R., Tang, T., Messina, C., 2014. Breeding drought-tolerant maize hybrids for the US corn-belt: discovery to product. J. Exp. Bot. 65, 6191–6204. Cotton Research and Development Corporation,, 2016. Australian Cotton Production Manual, 2016. Cotton Research and Development Corporation, Narrabri, NSW. Dowling, D., 2015. Yield records broken… again. Cotton Yearbook 2015 36, 6. Ford, B., Forrester, N., 2002. Impact of rainfall variability. In: Schulze, R. (Ed.), Australian Dryland Cotton Production Guide. Cotton Research and Development Corporation, Narrabri, NSW, pp. 13–16. Grimes, D.W., Miller, R.J., Dickens, L., 1970. Water stress during flowering of cotton. California Agric. 24, 4–6. Haffke, S., Wilde, P., Schmiedchen, B., Hackauf, B., Roux, S., Gottwald, M., Miedaner, T., 2015. Toward a selection of broadly adapted germplasm for yield stability of hybrid rye under normal and managed drought stress conditions. Crop Sci. 55, 1026–1034. Hearn, A.B., Constable, G.A., 1984. Irrigation for crops in a sub-humid environment: VII. Evaluation of irrigation strategies for cotton. Irrig. Sci. 5, 75–94. Hearn, A.B., Da Roza, G.D., 1985. A simple-model for crop management applications for cotton (Gossypium hirsutum L.). Field Crop. Res. 12, 49–69. Hearn, A.B., 1994. OZCOT − A simulation model for cotton crop management. Agric. Syst. 44, 257–299. Hearn, A.B., 1995. High prices and low rainfall: a recipe for frustraion or an opportunity for a calculated risk? Aust. Cottongrower 16, 20–28. Hodgson, A.S., Chan, K.Y., 1987. Field calibration of a neutron moisture metre in a cracking grey clay. Irrig. Sci. 8, 233–244. Jackson, B.S., Arkin, G.F., Hearn, A.B., 1988. The cotton simulation model COTTAM: fruiting model calibration and testing. Trans. ASAE 31, 846–854. Kumar, A., Bernier, J., Verulkar, S., Lafitte, H.R., Atlin, G.N., 2008. Breeding for drought

5. Conclusion Ensuring material is evaluated in the target breeding environment as well as experimentation resulting in reliable genotype resolution are of the upmost importance in a breeding program. Selection outside of the target breeding environment can result in the progression of inferior material, and experimental failure can limit the selection of superior material in a breeding program. Thus, breeding for rainfed cotton production in a variable rainfall environment can result in challenging experimental data. This study developed and explored the use of a managed stress environment (MSE). The MSE produced ‘rainfed’ experiments that ensure expression of genetic variability better matched to our breeding target environment during very dry seasons. We conclude that breeding should continue to focus on yield potential and yield reliability under non-stressed conditions, as well as evaluating germplasm for yield performance under rainfed conditions. However, in very dry seasons the addition of a MSE at one site will reduce the occurrence of experimental data that is not matched to our breeding target environment. This strategy should result in the development of germplasm suited to the Australian rainfed production, regardless of variable rainfall environments during germplasm testing and selection. As MSE in very dry seasons better matches the breeding target environment, genotype performance in these seasons under MSE may actually identify better and more realistic long term options for Australian rainfed cotton producers. Potential GxE interactions can be managed through a combination of a sensible approach to the application of MSE irrigations (based on rainfall outlooks), as well as continuing to evaluate germplasm at multiple sites under true rainfed conditions. Best performing genotypes under rainfed conditions in dry seasons can be captured through the evaluations conducted at alternative locations, and genotype selection should take into account 11

Field Crops Research xxx (xxxx) xxx–xxx

W.C. Conaty et al.

Ritchie, J.T., 1972. Model for predicting evaporation from a row crop with incomplete cover. Water Resour. Res. 8 (1204- &). Slafer, G.A., Araus, J.L., Royo, C., Del Moral, L.F.G., 2005. Promising eco-physiological traits for genetic improvement of cereal yields in Mediterranean environments. Ann. Appl. Biol. 146, 61–70. Stiller, W., Reid, P., 2005. Siokra 24. Plant Varieties J. 18, 89–92. Stiller, W., Reid, P., 2013. Sicot 730. Plant Varieties J. 26, 89–92. Stiller, W.N., Reid, P.E., Constable, G.A., 2004. Maturity and leaf shape as traits influencing cotton cultivar adaptation to dryland conditions. Agron. J. 96, 656–664. Stiller, W., 2007a. Sicala 60BRF. Plant Varieties J. 20, 121–124. Stiller, W., 2007b. Sicot 80BRF. Plant Varieties J. 20, 125–128. Stiller, W., 2007c. Sicot 81. Plant Varieties J. 20, 133–136. Stiller, W., 2008a. Sicot 71BRF. Plant Varieties J. 21, 194–197. Stiller, W., 2008b. Sicot 75. Plant Varieties J. 21, 190–193. Stiller, W., 2010a. Sicot 74BRF. Plant Varieties J. 23, 143–146. Stiller, W., 2010b. Siokra 24BRF. Plant Varieties J. 23, 135–138. Stiller, W., 2010c. Siokra V-18BRF. Plant Varieties J. 23, 135–138. Stiller, W., 2011. Sicot 75BRF. Plant Varieties J. 24, 148–151. Tambussi, E.A., Nogues, S., Ferrio, P., Voltas, J., Araus, J.L., 2005. Does higher yield potential improve barley performance in Mediterranean conditions? A case study. Field Crop. Res. 91, 149–160. Tennakoon, S.B., Hulugalle, N.R., 2006. Impact of crop rotation and minimum tillage on water use efficiency of irrigated cotton in a Vertisol. Irrigation Sci. 25, 45–52. Turner, N.C., Blum, A., Cakir, M., Steduto, P., Tuberosa, R., Young, N., 2014. Strategies to increase the yield and yield stability of crops under drought − are we making progress? Funct. Plant Biol 41, 1199–1206. Venuprasad, R., Lafitte, H.R., Atlin, G.N., 2007. Response to direct selection for grain yield under drought stress in rice. Crop Sci. 47, 285–293. Venuprasad, R., Sta Cruz, M.T., Amante, M., Magbanua, R., Kumar, A., Atlin, G.N., 2008. Response to two cycles of divergent selection for grain yield under drought stress in four rice breeding populations. Field Crop. Res. 107, 232–244. Verulkar, S.B., Mandal, N.P., Dwivedi, J.L., Singh, B.N., Sinha, P.K., Mahato, R.N., Dongre, P., Singh, O.N., Bose, L.K., Swain, P., Robin, S., Chandrababu, R., Senthil, S., Jain, A., Shashidhar, H.E., Hittalmani, S., Cruz, C.V., Paris, T., Raman, A., Haefele, S., Serraj, R., Atlin, G., Kumar, A., 2010. Breeding resilient and productive genotypes adapted to drought-prone rainfed ecosystem of India. Field Crop. Res. 117, 197–208. Whitaker, D., Williams, E.R., John, J.A., 2002. CycDesigN: a Package for the Computer Generation of Experimental Designs. CSIRO Forestry and Forest Products, Canberra, Australia.

tolerance direct selection for yield, response to selection and use of drought-tolerant donors in upland and lowland-adapted populations. Field Crop. Res. 107, 221–231. Lafitte, H.R., Li, Z.K., Vijayakumar, C.H.M., Gao, Y.M., Shi, Y., Xu, J.L., Fu, B.Y., Ali, A.J., Domingo, J., Maghirang, R., Torres, R., Mackill, D., 2006. Improvement of rice drought tolerance through backcross breeding: evaluation of donors and selection in drought nurseries. Field Crop. Res. 97, 77–86. Liu, S.M., Constable, G.A., Cullis, B.R., Stiller, W.N., Reid, P.E., 2015. Benefit of spatial analysis for furrow irrigated cotton breeding trials. Euphytica 201, 253–264. Meredith, W.R.J., 1984. Quantative genetics. In: Kohel, R.J., Lewis, C.F. (Eds.), Cotton. Americal Society of Agronomy, Inc, Crop Science Society of America, Inc., Soil Science Society of America, Inc Madison, WI, USA, pp. 131–150. Milroy, S.P., Bange, M.P., Hearn, A.B., 2004. Row configuration in rainfed cotton systems: modification of the OZCOT simulation model. Agric. Syst. 82, 1–16. Monsanto Australia Ltd, 2012a. Bollgard II® Cotton Technical Manual. Monsanto Australia Ltd. Web., Melbourne, Australia. Monsanto Australia Ltd, 2012b. Roundup Ready Flex® Cotton Technical Manual. Monsanto Australia Ltd. Web., Melbourne, Australia. Monsi, M., Saeki, T., 2005. On the factor light in plant communities and its importance for matter production. Ann. Bot. 95, 549–567. Morgan, J.M., 2000. Increases in grain yield of wheat by breeding for an osmoregulation gene: relationship to water supply and evaporative demand. Aust. J. Agric. Res. 51, 971–978. Ouk, M., Basnayake, J., Tsubo, M., Fukai, S., Fischer, K.S., Cooper, M., Nesbitt, H., 2006. Use of drought response index for identification of drought tolerant genotypes in rainfed lowland rice. Field Crop. Res. 99, 48–58. Pantuwan, G., Fukai, S., Cooper, M., Rajatasereekul, S., O’Toole, J.C., 2002. Yield response of rice (Oryza sativa L.) genotypes to different types of drought under rainfed lowlands: part 1. Grain yield and yield components. Field Crop. Res. 73, 153–168. Pinheiro, B.d.S., Austin, R.B., do Carmo, M.P., Hall, M.A., 2000. Carbon isotope discrimination and yield of upland rice as affected by drought at flowering. Pesqui. Agropecu. Bras. 35, 1939–1947. Rebetzke, G.J., Chenu, K., Biddulph, B., Moeller, C., Deery, D.M., Rattey, A.R., Bennett, D., Barrett-Lennard, E.G., Mayer, J.E., 2013. A multisite managed environment facility for targeted trait and germplasm phenotyping. Funct. Plant Biol. 40, 1–13. Reid, P., 1992a. CS 50. Plant Varieties J. 5, 12. Reid, P., 1992b. Siokra L23. Plant Varieties J. 5, 13–14. Reid, P., 1995. Sicala V-2. Plant Varieties J. 8, 12–13. Reid, P., 1996. Sicot 189. Plant Varieties J. 9, 18–19. Reid, P., 1998. Siokra V-16. Plant Varieties J. 11, 20. Reid, P.E., 2003. Sicot 71. Plant Varieties J. 16, 35–36.

12