Experimental design and plot management
Cotton (Gossypium hirsutum, cv. 71BRF [Bollgard II® Roundup Ready Flex®], CSIRO Australia) was grown at the Australian Cotton Research Institute (ACRI), Narrabri, NSW during the 2011/12 growing season. Cotton was planted at a row spacing of 1 m with a sowing density of 14 plant· m−2 on three dates within the sowing window: early-season (S1) = 5th October 2011; mid-season (S2) = 9th November 2011; late-season (S3) = 30th November 2011. The three sowing times exposed developing cotton to different minimum and maximum temperatures and relative humidity, consequently resulting in different atmospheric VPD within the same cotton developmental state, and then the physiological responses to VPD was measured. Experiments were managed according to current Australian practices, except for imposed irrigation treatments as outlined. Daily weather conditions, including minimum and maximum air temperatures and rainfall events were obtained from the Myall Vale weather station (Fig. 1). The authors acknowledge that data for this study was from 2011. Weather conditions are usually highly variable between years, however, a separate study of climate change across Australian cotton regions shows that there has been no significant change in minimum temperature, temperatures above 40 °C, or rainfall distribution in Narrabri from 1997 to 2018 (Broughton et al., in press). Therefore, we are comfortable using data from 2011 for this study, which focuses more on the physiological response of cotton to differing environmental conditions rather than specific climates.
Soil water availability was manipulated using three irrigation treatments: (a) fully watered—non-stressed (NS); (b) limited water—early stress (ES); and (c) limited water—late stress (LS). Similar to the sowing time, the purpose of the irrigation treatments was to generate different VPD environments within each cotton developmental stage rather than assessing the effect of water stress on cotton physiology. The first irrigation event was skipped for the ES. One irrigation event was skipped during the early boll-fill stage for the LS. Plastic was used to cover the ground during the water stress treatments to reduce the risk of rainfall prematurely alleviating the stress. However, heavy rainfall and flooding during the season led to the exclusion of measurements in some of the treatments (i.e., the first sowing and late season water stress). Furrow irrigation and rainfall events resulted in lower VPD environments and skipped irrigation treatments resulted in higher VPD environments during periods of measurement.
Each plot consisted of two rows of furrow-irrigated cotton with an additional two-row buffer around (a total of four rows per plot). Plants were irrigated down the centre three rows to minimise the lateral movement of water from the fully irrigated to the water-stressed plots. Each row was 66 m long, except for the three control irrigation treatments, which were each 22 m long. The field replication is not orthogonal in design due to logistical difficulties in applying irrigation treatments and laying the plastic, however, the purpose was to generate differences in VPD. Plant response to the variable environment was determined through measuring gas exchange of 2∼3 different plants per day in each sampling period.
Leaf gas exchange
Net photosynthesis (Asat) and stomatal conductance (gs-sat) under saturating light were measured on recently fully expanded leaves using a portable open gas exchange system (LI-6400XT, LI-COR, Lincoln, USA). Leaf gas exchange measurements were taken at saturating light (photosynthetic photon flux density of 2 000 µmol·m−2·s−1), 400 µL·L−1 [CO2], and block temperature was set to the anticipated mid-day temperature. VPD response curves were achieved by controlling VPD (temperature × relative humidity) at the leaf surface within the infrared gas analyzer (IRGA) chamber. Measurements began on full-bypass of air, representing natural ambient humidity conditions in the field; these initial measurements were referred to as the ‘field VPD’ dataset. Subsequently, water from the incoming air was slowly absorbed using the desiccant Drierite (W. A. Hammond Drierite Co., USA), but block temperature remained the same, thereby increasing leaf vapour pressure deficit (VPDL) by approximately 0.5 kPa for each gas exchange measurement. All gas exchange measurements, including these VPD response curves, were referred to as the ‘variable VPD’ dataset. The number of measurements captured for each response curve varied because of differences in the range of VPD generated by the IRGA at each time and day.
Measurements were made between the hours of 10 am to 3:30 pm (Australian Eastern Daylight Time, AEDT), over 3 or 4 consecutive days at different times during the soil water stress treatment. Measurements during the same period of consecutive days were made on the same recently mature leaf (3rd leaf from the top of the plant on the first day of measurement), which had been tagged. Each leaf was allowed approximately 2 min, or until parameters stabilised, to equilibrate before the reading was recorded. Leaf gas exchange measurements of the equivalent control (i.e., non water-stressed treatment) plants were taken on the same day as the water-stressed plants. Two or three plants from each plot were measured per day.
Soil water status
A neutron probe was used to monitor soil water content at 0.2 m intervals to a depth of 1.2 m. These measurements were taken every 10 days throughout the season, and each week during water stress treatments. Volumetric soil water content (VSWC, %) was calculated using a formula, which was calibrated for soils in an adjacent field, with the same soil classification at ACRI (Ward et al. 1999). VSWC(% )= 0.000 6x + 24.225; where x is the count measurement at each depth (Warren Conaty, pers comm.). VSWC(%) was averaged across all depths (i.e., 0∼120 cm).
Canopy temperature
Wireless, battery-operated SmartCrop infrared thermometers (Smartfield Inc., Lubbock, TX, USA) were used to monitor canopy temperature in each plot. Sensors were periodically repositioned to maintain them at 20∼30 cm above the canopy pointing south at an angle of 45° to the horizontal during the measurement. Where possible, two sensors were placed towards the centre of each plot; however, some plots only had one sensor due to a limited number of sensors. Stress hours were recorded by the SmartCrop sensors as hours that the canopy temperature was above 28 °C, which is considered the optimal thermal temperature for cotton (Conaty 2011), while regardless of actual canopy temperature. These data were used to calculate accumulated temperature stress hours (ASH) between irrigation events, similar to the calculation of heat units by Mahan et al. (2014).
Statistical analysis
Testing treatment effects
Summary statistics were used to explore the variation generated by the sowing dates and water treatment effects across the variable VPD dataset.
Testing environmental effects
Generalised linear models were used to link the responses of field-grown cotton to the treatment effects with the overall responses of the plants to the biological and environmental responses. To test model effects, data were analysed using Genstat version 16 (VSN International, Hemel Hempstead, UK) by forward stepwise regression. Variables were sequentially added to the model using a significance level of 0.05 to determine if the variable remained in the model. This method was used both for the field VPD dataset and the variable VPD dataset for both stomatal conductance and photosynthesis. A number of variables were tested, but the best maximal model for both stomatal conductance and photosynthesis was found to be: VPDL × Plant × ASH, where: VPDL is leaf-level vapour pressure deficit, Plant is the individual plant, and ASH is accumulated temperature stress hours. Stomatal conductance data were transformed logarithmically, which improved R2 over a linear regression (from 0.387 to 0.428). Note that while the individual plant was a random effect rather than an environmental variable, variation attributed to the plant needed to be taken into account as part of the regression analysis.
Testing the Asat/E (ITE) model
Duursma et al. (2013) showed that for cotton grown in a controlled environment glasshouse (in ambient and elevated CO2 and temperature treatments), the following equation gave satisfactory fits to measured ITE when VPD was varied independently of temperature and other environmental drivers:
$$ITE=\frac{A}{E}=\frac{{C}_{a}{P}_{a}}{{g}_{1}{Ds}^{k}+Ds}$$
(1)
where ITE is the ratio of photosynthesis to transpiration (µmol·mmol−1), Ca is atmospheric [CO2] (%), Pa is the atmospheric pressure (kPa), g1 is the ‘slope parameter’ which is related to the marginal cost of water (see Medlyn et al. 2011 for details), Ds is the leaf-to-air vapour pressure deficit (kPa), and k is an empirical parameter. Based on the assumption that stomata respond optimally to changes in VPD, k would equal to 0.5.
To test whether this model, and the parameters estimated by Duursma et al. (2013) are suitable to estimate Asat/E in field conditions, Eq. (1) was fitted to a “well-watered” subset of the VPD measurements of field grown cotton using R (version 3.1.0). R was the preferred software for this analysis because of easier application of the model. The “well-watered” subset based on VPD data within the first five days of gas exchange measurements for each treatment was used because it is known that the model is not appropriate for water-stressed conditions (Remko Duursma, pers. comm). Predicted (modelled) Asat/E was compared with measured Asat/E using parameters estimated from (a) the fit to field data and (b) glasshouse prediction data. The root mean square error (RMSE) was used to indicate the goodness-of-fit for model predictions of the measured Asat/E values. The mean absolute difference (MAD) was used as a measure of the difference between modelled estimates and measured values.