The experiment was performed in the Macrocosms platform of the CNRS Montpellier European Ecotron. This platform houses 12 identical and independent experimental units. Each unit is composed of a dome under natural light covering a lysimeter inserted in a technical room. The linear series of 12 domes is oriented east-west with two additional domes added at each extremity to eliminate any self-shading edge effects. The 30 m3 transparent domes allow for the confinement and control of the atmosphere. Below each dome a lysimeter/ technical room hosts: the soil monolith contained in a lysimeter (2 m2 area, 2 m depth), the lysimeter’s weighing strain gauges and various soil-related sensors, the canopy air temperature and relative humidity conditioning units and the air CO2 regulation. Each dome has a circular base area of 25 m2, of which 20 m2 is covered by concrete and 5 m2 central area allocated for the model ecosystems (area 2, 4 or 5 m2), height in the centre of the dome is 3.5 m. The airflow from the dome area is prevented to leak into the lysimeter room by the means of fitting metal plates and rubber seals. Such airflow (from the cooling system) is of two volumes per minute (=70 m3 min−1) creating a turbulent environment, where wind speed varies between 0.7–2.5 m s−1 in a fraction of a second and with averaged (during a few seconds) anemometer readings (Almemo 2890-9, Coalville, UK) of 0.9–1 m s−1. This led to a well-coupled canopy where no significant differences between leaf (MS LT, Optris GmbH, Berlin, Germany) and air temperature (PC33, Mitchell Instrument SAS, Lyon, France) existed (intercept = −4.3 ± 4.5 [mean ±95%CI]; slope = 1.15 ± 0.17; R2 = 0.89). The concrete is covered with epoxy-resin to prevent its CO2 absorption.
Each macrocosm was designed as an open flow gas exchange system. A multiplexer allowed for the CO2 concentrations at the inlet and outlet of each dome to be measured every 12 min (LI-7000 CO2/H2O analysers, LI-COR Biosciences, Lincoln, NE, USA). These data combined with the measurement of the air mass flow through each dome allowed for the calculation of canopy carbon assimilation (Ac). Transpiration (mass loss of the lysimeter) was monitored continuously by four CMI-C3 shear beam load cells (Precia-Molen, Privas, France) providing 3 measurements per minute. We ensured only canopy carbon (Ac) and water (Ec) balances were measured by covering the ground with a dark plastic cover that prevented flux mixing. This plastic cover was sealed to the fitting metal plates and not to the lysimeter upper ring. There was a slight over-pressure (+5 Pa) in the dome and a small proportion of the well mixed air canopy could be passing around the plant stems, therefore flushing the soil respiration and evaporation below the plastic sheet and into the lysimeter room.
The dome was covered by a material highly transparent to light and UV radiation (tetrafluoroethylene film, Dupont USA, 250 μm thick, PAR transmission 0.9) and exposed to natural light except during the reduced radiation experiments. Here, an opaque fitted cover (PVC coated polyester sheet Ferrari 502, assembled by IASO, Lleida, Spain) was placed on each dome and a set of 5 dimmable plasma lamps with a sun-like spectrum (GAN 300 LEP with the Luxim STA 41.02 bulb, Gavita Netherlands), allowed to control radiation. The plasma lamps were then turned off to study dark circadian regulation of stomatal conductance. Our conditions may differ from a cloudy day in that radiation was direct, not diffuse. We were interested in testing how reductions in carbon assimilation affect nocturnal transpiration, therefore, avoiding diffuse radiation was considered advantageous because it increases carbon uptake31.
Bean and cotton were planted in rows, one month before the start of the measurements and thinned at densities of 10.5 and 9 individuals m−2 respectively. Six macrocosms were assigned to each species and each individual experiment measuring campaign lasted for 3–4 days. The experiments under constant darkness lasted for 30 hours and we used lysimeter weight readings from three macrocosms per species (six per species in all the other reported experiments). In the three other macrocosms researchers were entering every 4 hours to conduct manual leaf gas exchange measurements at three leaves per dome (LI-6400, LI-COR Biosciences, Lincoln, NE, USA). At the time of measurements, bean and cotton were both at the inflorescence emergence developmental growth stage (codes 51–59 in BBCH scale32).
The soil was regularly watered nearly to field capacity by drip irrigation, although irrigation was stopped during the few days of each measuring campaign in order not to interfere with water flux measurements. No significant differences (at P < 0.05, paired t-test, n = 3) in predawn leaf water potential occurred after a few days of withholding watering. This indicates that no effect of potential changes in soil moisture on plant water status over the course of the experiment.
Transpiration was calculated from the slope of the linear regression between lysimeter weight and time every 3 hours successive periods. Statistical analyses of temporal patterns were then conducted with Generalized Additive Mixed Model (GAMM) fitting with automated smoothness selection33 in the R software environment (mgcv library in R 3.0.2, The R Foundation for Statistical Computing, Vienna, Austria), including macrocosms as a random factor and without including outliers (values above 95% quantile during day or night). This approach was chosen because it makes no a priori assumption about the functional relationship between variables. We accounted for temporal autocorrelation in the residuals by adding a first-order autoregressive process structure (nlme library34). Significant temporal variation in the GAMM best-fit line was analysed after computation of the first derivative (the slope, or rate of change) with the finite differences method. We also computed standard errors and a 95% point-wise confidence interval for the first derivative. The trend was subsequently deemed significant when the derivative confidence interval was bounded away from zero at the 95% level (for full details on this method see35). Non-significant periods, reflecting lack of local statistically significant trending, are illustrated on the figures by the yellow line portions and significant differences occur elsewhere.
Differences in the magnitude of total transpiration and in canopy conductance for each species under the different radiation environments were calculated from mixed models that included radiation as a fixed factor (and hour also for canopy conductance) and macrocosms and day of measurement as random factors (each measuring campaign lasted 3–4 days).