•Night-time stomatal conductance (gnight) occurs in many ecosystems, but the gnight response to environmental drivers is relatively unknown, especially in deserts.
•Here, we conducted a Bayesian analysis of stomatal conductance (g) (N = 5013) from 16 species in the Sonoran, Chihuahuan, Mojave and Great Basin Deserts (North America). We partitioned daytime g (gday) and gnight responses by describing g as a mixture of two extreme (dark vs high light) behaviors.
•Significant gnight was observed across 15 species, and the gnight and gday behavior differed according to species, functional type and desert. The transition between extreme behaviors was determined by light environment, with the transition behavior differing between functional types and deserts. Sonoran and Chihuahuan C4 grasses were more sensitive to vapor pressure difference (D) at night and soil water potential (Ψsoil) during the day, Great Basin C3 shrubs were highly sensitive to D and Ψsoil during the day, and Mojave C3 shrubs were equally sensitive to D and Ψsoil during the day and night.
•Species were split between the exhibition of isohydric or anisohydric behavior during the day. Three species switched from anisohydric to isohydric behavior at night. Such behavior, combined with differential D, Ψsoil and light responses, suggests that different mechanisms underlie gday and gnight regulation.
Night-time stomatal conductance (gnight) and transpiration (Enight) have been observed in several plant functional groups from diverse ecosystems (Barbour et al., 2005; Daley & Phillips, 2006; Caird et al., 2007; Marks & Lechowicz, 2007), including semi-arid settings (Snyder et al., 2003; Dawson et al., 2007). Enight is generally < 15% of daytime transpiration (Caird et al., 2007), but may range up to 30% in deserts, partially reflecting low daytime transpiration as a result of low daytime stomatal conductance (gday) associated with stomatal closure or constrained maximum conductance (Snyder et al., 2003). Given that gnight varies in time and is substantially higher than cuticular conductance (Caird et al., 2007; Zeppel et al., 2010), this implies the occurrence of guard cell regulation (e.g. Roelfsema & Hedrich, 2005). Indeed, Enight is often positively correlated with the night-time leaf-to-air vapor pressure difference (D), such that Enight is greatest on high-temperature, low-humidity nights, and is less influenced by soil water potential (Ψsoil) (Phillips et al., 2010; Zeppel et al., 2010). However, Dawson et al. (2007) found that a variety of woody plants maintained low gnight on nights with high D or low Ψsoil (dry soil), and gnight was enhanced by improved soil water status. Such conflicting patterns make it unclear how D, Ψsoil and other environmental variables affect gnight and Enight.
Stomatal conductance (g) during the day can be highly variable between co-occurring species (e.g. Tardieu & Simonneau, 1998; Brodribb & Jordan, 2008; Quero et al., 2011), and gday of individual species can be characterized along a continuum between extreme isohydric and anisohydric behaviors (e.g. Buckley, 2005; Collins et al., 2010). Isohydric plants regulate gday to maintain relatively constant leaf water potentials (Ψleaf) (Lambers et al., 1998) and may operate at Ψleaf near xylem cavitation thresholds (e.g. Holtta et al., 2009). This strategy may increase susceptibility to mortality during prolonged drought (West et al., 2008). Conversely, anisohydric plants exhibit variable diurnal Ψleaf, together with decreasing Ψleaf and gday, in response to decreasing Ψsoil, allowing photosynthesis at very low Ψsoil (Lambers et al., 1998). Anisohydric and isohydric strategies are infrequently used to describe gnight, because these strategies have rarely been evaluated at night (but, see Rogiers et al., 2009). The characterization of anisohydry and isohydry during both daytime and night-time periods within and between species may lend mechanistic insight into the stomatal control of Enight.
We assessed diel g and transpiration (E) in the Chihuahuan, Great Basin, Mojave and Sonoran Deserts. In these deserts, we measured day and night patterns of g and E for 16 functionally and physiologically diverse species (C3 and C4 grasses, C3 shrubs and a C3 rosette). We synthesized observed g (N = 5013) within a hierarchical Bayesian (HB) framework to: assess gnight and Enight variation by species, functional type and desert; partition gnight and gday as a mixture of the g behavior expressed during dark vs high light; evaluate gnight and gday responsiveness to light, D and Ψsoil; and explore the variation in gday and gnight along the isohydric to anisohydric continuum.
Materials and Methods
We focused on 16 common plant species (Table 1; Supporting Information Fig. S1). Artemisia tridentata and Purshia tridentata were measured at the Valentine Eastern Sierra University of California Natural Reserve near Mammoth Lakes, CA, USA (Great Basin Desert). Achnatherum hymenoides, Krameria parvifolia, Larrea tridentata, Lycium andersonii, Lycium pallidum and Pleuraphis rigida were measured at the Mojave Global Change Facility, located on the Nevada Test Site, NV, USA (Mojave Desert). Heteropogon contortus and Eragrostis lehmanniana were measured at the Santa Rita Experimental Range near Tucson, AZ, USA (Sonoran Desert). Artemisia ludoviciana, Bouteloua curtipendula, Bouteloua hirsuta, Dasylirion leiophyllum, Gutierrezia microcephala, L. tridentata and Nolina texana were measured at Big Bend National Park, TX, USA (Chihuahuan Desert). Larrea tridentata was measured at two sites, resulting in 17 species–desert combinations (hereafter, just ‘species’). Although we used data from all species in the analysis, we focus on the 12 ‘data-rich’ species that had at least five night-time and 20 total observations (see Table 1). Study site details are given elsewhere (Gillespie & Loik, 2004; Barker et al., 2006; Ignace et al., 2007; Patrick et al., 2009).
Table 1. Species information and summaries of measured daytime (gday) and night-time (gnight) stomatal conductance based on field data collected in the four North American deserts
gday (mol m−2 s−1)
gnight (mol m−2 s−1)
Night : day
PFT, plant functional type; Moj., Mojave; GB, Great Basin; Son., Sonoran; Chi., Chihuahuan; ACHY, Achnatherum hymenoides; ARLU, Artemisia ludoviciana; ARTR, A. tridentata; BOCU, Bouteloua curtipendula; BOHI, B. hirsuta; DALE, Dasylirion leiophyllum; ERLE, Eragrostis lehmanniana; GUMI, Gutierrezia microcephala; HECO, Heteropogon contortus; KRPA, Krameria parvifolia; LATR, Larrea tridentata; LYAN, Lycium andersonii; LYPA, L. pallidum; NOTE, Nolina texana; PLRI, Pleuraphis rigida; PUTR, Purshia tridentata. Monte Carlo (MC) simulations were conducted to propagate uncertainty in the reported conductance values according to the observation error model for SE2 described with Eqn 2. For each MC simulation, the sample median and 2.5th and 97.5th percentiles were computed based on the sample size (n), and the median values of these statistics (across the 3000 MC simulations) are reported; values in parentheses below the median are the 2.5th and 97.5th percentiles for the median (again, based on the MC simulations). Within each MC simulation, the median gnight divided by the median gday was computed for data-rich species, and the median, 2.5th and 97.5th percentiles across the 3000 MC simulations are reported (ratio night : day). Data rich = Y (yes) if a species is associated with > 5 night-time observations and > 20 total observations, N (no) otherwise.
0.1936 (0.187, 0.198)
0.1417 (0.127, 0.151)
0.7322 (0.656, 0.787)
0.0868 (0.085, 0.089)
0.0544 (0.049, 0.059)
0.6268 (0.568, 0.678)
0.0589 (0.058, 0.060)
0.0330 (0.031, 0.034)
0.5607 (0.530, 0.576)
0.0847 (0.082, 0.087)
0.3778 (0.331, 0.395)
4.4533 (3.894, 4.674)
0.0836 (0.082, 0.087)
0.0906 (0.090, 0.091)
0.3493 (0.323, 0.375)
3.8546 (3.562, 4.141)
0.0663 (0.065, 0.067)
0.0473 (0.046, 0.049)
0.7127 (0.687, 0.737)
0.1356 (0.135, 0.137)
0.0946 (0.093, 0.097)
0.6978 (0.685, 0.714)
0.0881 (0.087, 0.089)
0.0431 (0.042, 0.045)
0.4903 (0.477, 0.508)
0.0798 (0.079, 0.080)
0.0284 (0.028, 0.029)
0.3566 (0.349, 0.362)
0.0151 (0.0147, 0.0153)
0.1178 (0.117, 0.119)
0.0543 (0.054, 0.055)
0.4604 (0.455, 0.464)
0.0434 (0.043, 0.044)
C3 monocot rosette
0.0883 (0.087, 0.089)
0.0022 (0.0021, 0.0024)
0.0255 (0.024, 0.027)
0.0617 (0.061, 0.063)
0.0214 (0.021, 0.022)
0.0093 (0.0091, 0.0094)
0.4337 (0.426, 0.442)
0.0021 (0.0020, 0.0021)
Field measurements of leaf and environmental variables
Leaf-level stomatal conductance (g), transpiration (E), ambient leaf-to-air vapor pressure difference (D), photosynthetically active radiation (PAR) and leaf temperature (T) were measured with portable, open-flow gas-exchange systems (Model LI-6400; LI-COR Inc., Lincoln, NE, USA) that were cross-calibrated when multiple systems were used concurrently. Data collection varied slightly between sites as a result of site-specific research objectives underlying the original measurements. The D and air temperature of incoming air approximated external ambient values, the flow rate was maintained near 300 μmol s−1 (27% of observations) or 500 μmol s−1 (68%) and reference CO2 was held constant at 380 μmol mol−1 for each measurement. Most measurements were logged multiple times within 2–3 min for a given leaf (or leaf cluster); values were averaged to obtain a single observation for that leaf and sampling interval. Most often, a single leaf or leaf cluster was repeatedly measured during a particular 24–96-h period. Measurements were leaf area corrected, and only positive-valued g observations were retained. Most data were collected during sampling periods beginning in the late afternoon and concluding 24 h later, except for an L. tridentata dataset collected over a 96-h period in the Chihuahuan Desert.
At each site, volumetric soil water content (SWC) was recorded at daily to biweekly to monthly intervals, depending on desert and year, using soil moisture probes placed horizontally to a depth of 15 cm. For automated measurements, mean daily SWC was determined by averaging values recorded over 24-h periods; details of the data-logging frequency varied by site (Barker et al., 2006; Ignace et al., 2007; Patrick et al., 2009). Soil water potential (Ψsoil), averaged over soil depths from 5 to 15 cm, was estimated using the measured SWC, estimated soil hydraulic properties and measured daily precipitation in a physical-based simulation model, HYDRUS-1D (Šimunek et al., 2005). HYDRUS-1D uses the Richards’ equation for saturated and unsaturated water flow to predict a continuous time series of SWC and Ψsoil at multiple depths that were used in subsequent analyses.
Estimates of integrated Enight
We explored the implications of gnight for water loss by calculating the total water transpired during the night (Enight) for each observed 24-h period; the night-time period was defined as 21:30 h to 06:30 h local time and further constrained to PAR = 0 μmol m−2 s−1. We compared the estimated Enight with the total amount of water transpired over 24 h. Total transpiration was estimated by summing the observed transpiration for individual plants over time.
Data synthesis methods
Disentangling night-time and daytime behavior Stomatal conductance (g) and covariates (D, PAR and Ψsoil) were analyzed within an HB framework (Clark, 2005; Ogle & Barber, 2008) to evaluate gday and gnight patterns. An HB approach was used because it: is easily structured to the nested and unbalanced sampling designs associated with this large dataset; can directly propagate uncertainty in the reported g values and accommodate other sources of uncertainty contributing to the observed g patterns; allows for the computation of sensitivity indices and accurate estimates of their uncertainty; and can accommodate a nonlinear model that treats g as a mixture of the behaviors predicted in dark vs high light. The HB model implemented is analogous to a nonlinear, mixed-effects model that would be challenging to implement in a classical statistical framework.
Our data model describes the likelihood of the observed g (gobs). Preliminary analyses suggested that gobs is normally distributed such that, for observation i (i = 1, 2, …, 5013):
where is the predicted (or mean) conductance; the variance is decomposed into a term that quantifies the gobs – error variance () plus a term that quantifies the uncertainty in the reported gobs values (); preliminary analyses indicated that the first term depends on D, such that:
where ν1 and ν2 are parameters we estimated; greater variation in gobs is expected under low D (e.g. Ewers & Oren, 2000), and Eqn 2 allows σ2 to increase with decreasing D (e.g. if v2 < 0). The expected standard error of the multiple logged values used to compute is represented by SEi, and we propagate uncertainty in SEi by sampling log(CVi), where CVi = , from a normal distribution with a mean and variance that differed by species and the time index (day vs night; see Table S1).
The gday and gnight behaviors were evaluated by modeling at PAR = 0 μmol m−2 s−1 () and at PAR ≥ 2000 μmol m−2 s−1 (). Thus, can be thought of as extreme ‘end-members’, whereby is the behavior that emerges in the dark (gnight is only based on ), and is the behavior under high light. We assumed that g at intermediate light levels is a mixture of these end-members:
where δi (0 ≤ δi ≤ 1) describes the relative contribution of to the overall stomatal response (, and ɛtp,sp, and represent time period (within day and night), day and plot random effects. The random effects describe the uncertainty associated with our inability to account for all temporal and plot-level factors affecting gobs (see Notes S1 for definitions). The notations tp(i), sp(i), DOY(i), d(i) and p(i) denote the time period (tp), species (sp), day of year (DOY), desert (d) and plot (p), respectively, associated with observation i.
We define the mixture weight δi as a function of PAR, where δi = 1 when PAR = 0 and δi = 0 when PAR ≥ 2000 μmol m−2 s−1, and for 0 ≤ PAR ≤ 2000:
The species-specific parameter θsp describes how quickly the contribution ‘disappears’ as PAR increases.
where t is the end-member index (t = 2000 for ). The term inside the brackets is the predicted end-member at Ψsoil = 0 MPa, where b is the ‘reference g’ at D = Dref, and, following standard approaches, we set Dref = 1 kPa, and thus −m is the ‘stomatal sensitivity’ of g to changes in ln(D) (Oren et al., 1999; Ewers et al., 2007). The sensitivity of g to changes in Ψsoil is captured by ρ; b, m and ρ are allowed to depend on the species (sp) and t to reflect the possibility that D and Ψsoil may differentially affect species-specific gday and gnight. Importantly, these parameters can be interpreted with respect to potential anisohydric and isohydric behavior.
Eqns 1–5 define a nonlinear, mixed-effects regression; unique to HB is the specification of priors for the parameters, thereby obeying basic probability results to obtain the posterior distribution of the parameters (Gelman et al., 2004; Ogle & Barber, 2008). We specified a hierarchical model for the species-specific parameters in Eqns 4, 5, such that they vary around the associated desert-level parameters; hence, for θ and α = b, m or ρ:
d(sp) indicates desert d associated with species sp, t is the end-member index, and and denote the desert-level parameters. We assigned relatively noninformative, standard priors to all remaining parameters (see Notes S1).
Exploring isohydric and anisohydric behavior during the day and night We evaluated differences between the parameters to understand the maximum potential difference between gday and gnight behavior. For parameters that vary by species (αt,sp = bt,sp, mt,sp or ρt,sp) and desert (), we obtained the posterior distributions for the differences .
We evaluated the reference g and sensitivity of g under different Ψsoil by computing (see Eqn 5), illustrating how stomatal behavior differs in dark vs high light. For each posterior sample of b, m and ρ, we obtained a posterior value of b′ and −m′ associated with different Ψsoil values, and we focus on the results for Ψsoil = − 0.1 and − 1.0 MPa (2.5th and 50th percentiles of observed Ψsoil). We also obtained the standardized sensitivity terms by computing S = −m/b, which represents the slope defining the relationship between −m and b (Oren et al., 1999; Katul et al., 2009). We obtained the posterior distribution of S to determine whether desert species behave similarly to species from more mesic regions (expected S =0.6), and to evaluate daytime and night-time isohydric (S generally ≥ 0.6) vs anisohydric (S generally < 0.5) behavior (Oren et al., 1999).
Implementation of the HB model The HB model was implemented in OpenBUGS (Spiegelhalter et al., 2003; Lunn et al., 2009), and Notes S2 provides the code. Three parallel Markov chain Monte Carlo (MCMC) chains were assigned relatively dispersed starting values and run for a sufficiently long period to achieve convergence and obtain a posterior sample size effectively equivalent to >3000 independent samples (for details on MCMC procedures, see Gelman et al., 2004; Gamerman & Lopes, 2006).
Variation in gnight and Enight across species, functional types and deserts
We observed substantial gnight in 11 of the 12 ‘data-rich’ species (Fig. 1, Table 1). The magnitude of gnight varied considerably across species and was not related to functional type. For example, Lycium andersonii (Mojave C3 shrub) and Purshia tridentata (nitrogen-fixing Great Basin C3 shrub) exhibited the highest observed gnight values (97.5th percentile = 0.561 and 0.512 mol m−2 s−1, respectively; Table 1, Fig. 1); the lowest gnight values occurred for Dasylirion leiophyllum (Chihuahuan C3 monocot rosette) and Larrea tridentata (C3 evergreen shrub) in the Chihuahuan and Mojave Deserts (2.5th percentile = 0.0002 and 0.002 mol m−2 s−1, respectively; Table 1). For most species, median gnight ranged from 36% (H. contortus) to 73% (A. hymenoides) of median gday (Table 1). The three exceptions were D. leiophyllum, L. pallidum and P. rigida; D. leiophyllum had a median gnight value that was an order of magnitude lower than that of most species; L. pallidum and P. rigida exhibited median gnight values that were c. 4–4.5 times larger than their median gday values (Table 1).
Relatively large gnight resulted in night-time water loss (Enight) estimates that varied from 17% to 26% of total daily water loss for all species, except D. leiophyllum, whose nocturnal water loss was only 6% of its total daily water loss (Fig. 2). For the remaining species, the night-time water loss percentage was similar across all four deserts and functional groups.
HB model goodness-of-fit
The HB model used to explore gnight and gday behavior fit the field data well (Fig. S2; r2 = 0.74 for all data). The goodness-of-fit was slightly better for gnight vs gday (r2 = 0.76 vs 0.70) and differed across deserts (r2 = 0.44, 0.80, 0.81 and 0.83 for the Sonoran, Great Basin, Chihuahuan and Mojave Deserts, respectively) and species (lowest r2 = 0.18 for A. ludoviciana in the Chihuahuan Desert (N = 6); second lowest r2 = 0.37 for E. lehmanniana in the Sonoran Desert (N =972); highest r2 = 0.93 for L. andersonii in the Mojave Desert (N =42); median r2 across all 17 species = 0.73).
Sources of unexplained variation
The uncertainty in the observed g values decreased with increasing D, but the σ (standard deviation describing the residual errors; Eqn 2) vs D relationship differed across deserts (Fig. 3a–d). The model fit the field data better at high D relative to low D; the predicted values (posterior median) for σ at D =0 kPa were 7.4, 3.7, 1.2 and 9.1 times higher than the predicted values at D =10 kPa for the Mojave, Great Basin, Sonoran and Chihuahuan Deserts, respectively (Fig. 3a–d). The variation explained by time period and day of year was generally of similar magnitude to this residual variation, but the plot-to-plot variation was comparatively small (Fig. 3e).
Disentangling night-time and daytime behavior
The relative importance of dark () and high-light () end-member behavior, as quantified by δ in Eqn 4, differed across species and deserts. Four patterns emerged with respect to how the behavior decayed with increasing PAR (Fig. 4). The vast majority of species (A. tridentata, K. parvifolia, L. andersonii, P. rigida, P. tridentata and L. tridentata in the Mojave Desert) exhibited a clear distinction between gday and gnight, such that δ ≅ 0 for PAR ≥ 200 μmol m−2 s−1 (Fig. 4a). For three of the four C4 grasses (B. curtipendula, E. lehmannianna, H. contortus), the behavior persisted under a range of PAR values, whereby the transition to 100% occurred gradually with increasing PAR (Fig. 4b). One species (D. leiophyllum) exhibited intermediate behavior, where the contribution of disappeared when PAR ≅ 800–1200 μmol m−2 s−1 (Fig. 4c). Across these three groups, the decay in the relative contribution of with increasing PAR was tightly constrained (see relatively narrow 95% credible intervals (CI) for δ; Fig. 4a–c). By contrast, δ was not well resolved for A.hymenoides or L. tridentata in the Chihuahuan Desert, and three ‘data-poor’ species (Fig. 4d).
Environmental controls on gnight and gday
The sensitivity of g to D (−m) was > 0 for 11 of the 12 species under high light (for ), with L. tridentata in the Chihuahuan Desert being the exception (Fig. 5; see Table S2-B for posterior estimates). During the night-time , −m was only statistically > 0 for seven of the 12 species, and −m was < 0 (i.e. stomata open in response to increasing D) for L. tridentata in the Chihuahuan Desert (Table S2-B). Moreover, −m did not differ significantly between dark and high-light conditions for eight of the species, but −m was higher in the dark for the two Great Basin shrubs (A. tridentata and P. tridentata) and for two C4 grasses (H. contortus and B. curtipendula). Uncertainty in the −m estimates was greater during the night-time for seven of the species (wider CIs; Fig. 5; Table S2-B). Integrating across species, the sensitivity to D appears to differ across deserts (see Table S3-B for posterior estimates of −). Mojave plants tend to have − > 0 for both day and night (sensitive to D), Sonoran and Chihuahuan plants appear to be relatively insensitive to D during both day and night (− ≅ 0) and Great Basin plants appear to be sensitive to D under high light but not at night (Table S3-B).
Reference g (b) was statistically > 0 for all 12 species under high light (for ), and for 10 species during the night-time (), with the exceptions being two Mojave C3 shrubs (K. parvifolia and L. pallidum) (Fig. 5; see Table S2-A for posterior estimates of b). The b estimates associated with and were similar for four Mojave species. Conversely, high-light b was statistically greater than night-time b for six species, whereas night-time b was greater than high-light b for D. leiophyllum. The uncertainty in b was greater for for six species, but was greater for for five other species (Fig. 5, Table S2-A). Integrating across species, the desert-level reference g () was statistically > 0 for all deserts, during both daytime and night-time. However, Great Basin plants tended to have a higher under high light relative to night-time, but was similar between day and night for plants in the other deserts (Table S3-A).
Soil water availability affected g and the response to D (Fig. 5), such that the effect of Ψsoil on and was statistically significant for eight and seven, respectively, of the 12 species (see Table S2-D for posterior estimates of ρ). With respect to the high-light response, the estimates of ρ were relatively constrained (fairly narrow CIs; Table S2-D). With the exception of three C4 grasses (B. curtipendula, E. lehmannianna, H. contortus), more negative values of Ψsoil reduced g (ρ > 0). With respect to the night-time response, the ρ estimates were characterized by greater uncertainty, but, in general, ρ > 0, indicating that more negative Ψsoil also led to stomatal closure at night. The degree to which Ψsoil affected g differed between for seven of the 12 species; was more sensitive than to Ψsoil for the two Great Basin shrubs and L. andersonii (Mojave Desert), but was more sensitive to Ψsoil for the two Sonoran C4 grasses, L. tridentata in the Mojave Desert and D. leiophyllum. Across species, the desert-level Ψsoil effects (; see Table S3-D for posterior estimates) differed across deserts. That is, g appeared to be insensitive to Ψsoil for Sonoran and Chihuahuan plants, and , but not , was reduced significantly by lower Ψsoil in Mojave and Great Basin plants.
Exploring isohydric and anisohydric behavior during the day and night
We used the normalized sensitivity index (S = −m/b) to explore isohydric vs anisohydric behavior. With respect to the high-light () behavior, S values were relatively constrained (see narrow 95% CIs; Fig. 6, Table S2-C), and the posterior median for daytime S ranged from 0.27 to 0.93 for 11 species, for which S was statistically different from zero (Table S2-C). Across all 12 species, S was indistinguishable from or > 0.6 for five of the species, and S was significantly < 0.6 for the other seven (Fig. 6). Thus, nearly half of the species (A. tridentata and all Mojave species, except L. tridentata) were hypersensitive to D or exhibited apparent isohydric behavior (S ≥ 0.6) under high light. The other species (L. tridentata, P. tridentata and all Sonoran and Chihuahuan species) were relatively insensitive to D or exhibited apparent anisohydric behavior (S < 0.5).
In contrast with high-light S, night-time S (for ) was associated with greater uncertainty (Fig. 6; Table S2-C). S was only statistically different from zero for six of the species, for which the posterior median ranged from 0.20 to 1.55 (Fig. 6). The two Great Basin shrubs and E. lehmannianna appeared to be hypersensitive to D at night (S > 0.6), and they were significantly more sensitive to D during the dark relative to high light. Three other species (L. tridentata in the Mojave, H. contortus and D. leiophyllum) appeared to be relatively insensitive to D, exhibiting apparent anisoydric behavior (0 < S < 0.6) during the daytime and night-time.
In general, high-light S differed markedly across species; S differed between species for 65% of the 66 species pair-wise comparisons. Conversely, because of the large uncertainty in night-time S (Fig. 6), only 41% of the pairs yielded significant differences. The differences in high-light S across species resulted in desert-level indices ( that were unique to each desert (Table S3-C). Conversely, large uncertainties in species-specific night-time S resulted in large uncertainties in night-time , yielding night-time estimates that were indistinguishable between deserts (Table S3-C). Ultimately, g of Mojave plants appeared to be hypersensitive to D under high light (narrow 95% CI for that contains 0.6), but uncoupled to D during the night-time (wide 95% CI for that spans zero). Great Basin plants appeared to be hypersensitive to D, especially at night ( > 0.6). Sonoran plants appeared to approach isohydric-type behavior at night ( = 0.6), but were relatively insensitive to D under high light ( < 0.5). A general pattern did not emerge in Chihuahuan plants because of divergent or highly uncertain species-specific S values.
Variation in gnight and Enight across species, functional types and deserts
In our study, gnight was c. 40–75% of gday, with the exception of the C3 monocot (D. leiophyllum), suggesting that substantial night-time stomatal opening may be important in deserts. Most notably, gnight exceeded 60% of gday for half of the species studied here, and the occurrence of significant gnight appeared to be independent of growth form, photosynthetic pathway or desert (Fig. 1, Table 1). Substantial gnight has been reported for a small number of species in semi-arid systems (Snyder et al., 2003) and several species from a variety of ecosystems (Caird et al., 2007; Dawson et al., 2007). Interestingly, we found that several desert grasses exhibited relatively high gnight, which has not been commonly observed (Caird et al., 2007). Moreover, reference g (i.e. g at D = 1 kPa) was predicted to be significantly > 0 for 10 species during the night-time (Table S2-A), implying that stomata may remain open during the dark when D is low and soils are relatively moist, as would occur during rainy seasons.
A significant implication of gnight is the potential for night-time transpiration (Enight) to represent a substantial proportion of total daily transpiration (E; Fig. 2). With the exception of D. leiophyllum, Enight was 17–26% of total daily E, similar to results in Snyder et al. (2003). Relative Enight in our study species was comparable with that of plants from wetter environments (e.g. coastal, temperate deciduous, tropical evergreen forests and crops), where Enight ranges from 10% to 30% of the maximum daytime E rates (Barbour et al., 2005; Caird et al., 2007; Cavender-Bares et al., 2007; Dawson et al., 2007). During periods of greater water stress and lower total daily E, as is typical of many deserts, Enight may become a greater proportion of total daily E (Loik et al., 2004; Cavender-Bares et al., 2007), although absolute values of Eday and Enight may be lower under these conditions (Cavender-Bares et al., 2007). Our calculations may overestimate Enight because of potential issues with scaling E measured in the gas exchange cuvette, but this study and others suggest the potential for significant Enight in deserts.
The broader implications of Enight include the potential to underestimate ecosystem water fluxes because models commonly assume that stomata are closed at night. This assumption has been perpetuated in most gap-filling algorithms of eddy covariance datasets that largely neglect Enight (Novick et al., 2009; Thomas et al., 2009), in large-scale climate or hydrologic models that assume zero canopy conductance at night (Liang et al., 1994; Friend et al., 2007) and in stable isotope mixing models used to partition ecosystem fluxes, which require an accurate representation of gnight to incorporate diel 18O discrimination (Cernusak et al., 2004; Barbour et al., 2005; Seibt et al., 2007).
Disentangling night-time and daytime behavior
Next, we asked whether gnight behavior is distinguishable from gday behavior. To address this, we employed a unique analytical framework for partitioning gnight and gday by assuming that g is a mixture of two extreme behaviors: and , the predicted g values at PAR = 0 and 2000 μmol m−2 s−1, respectively. The contribution of each end-member is assumed to depend on ambient light levels (PAR), representing the potential ability to adjust g in response to daylight conditions.
Three C4 grasses (B.curtipendula, E. lehmanniana and H.contortus) exhibited gday that was influenced by the dark behavior over a range of PAR levels (Fig. 4b). A potential explanation is that these species exhibit sluggish stomatal responses (e.g. Mott et al., 1999; Kaiser & Kappen, 2001), such that they slowly adjust guard cell turgor in response to increasing PAR (e.g. sunflecks or sunrise) or decreasing PAR (e.g. ephemeral cloud cover or sunset). Conversely, the high-light behavior dominated gday under nearly the full range of PAR levels for all of the ‘data-rich’ species in the Great Basin and Mojave Deserts (Fig. 4a), suggesting that their stomata can rapidly adjust to changes in light (e.g. Mott et al., 1999). Five of six of these species are C3 plants, including four shrubs. Differences in the responsiveness of g to PAR could be a result of transport times of hormones affecting stomatal aperture (e.g. abscisic acid), and the taller stature of the shrubs may expose their canopies to relatively high D (Rambo & North, 2009), potentially stimulating more rapid stomatal adjustment under changing PAR (Mott et al., 1999; Kaiser & Kappen, 2001). The shrubs may also have deeper roots and greater access to more stable soil water (e.g. Ogle et al., 2004), improving their ability to maintain turgid guard cells and to enable rapid stomatal responses (Saxe, 1979; Roelfsema & Hedrich, 2005). The dissimilar behavior of the C4 vs C3 species could also be attributed to the differential roles of direct (changes in g directly attributed to changes in PAR) and indirect (as affected by changes in leaf internal CO2 concentration) responses to PAR (Huxman & Monson, 2003).
Evaluating environmental controls on gnight and gday
As one might expect, g of nearly all species was affected by Ψsoil. The model in Eqn 5 assumes an overall, multiplicative effect of Ψsoil on both reference g (b) and stomatal sensitivity (−m). The assumption that both are similarly affected by Ψsoil is supported by the observed tight relationship between reference g and sensitivity (Oren et al., 1999; Katul et al., 2009), and a similar model was successfully applied to L. tridentata in the Chihuahuan Desert (Ogle & Reynolds, 2002). Across most species, lower Ψsoil was associated with reduced g, reference g and sensitivity for both gday and gnight, similar to the observation that soil water deficits led to stomatal closure at night in several woody plants from diverse ecosystems (Dawson et al., 2007). The negative effects of declining Ψsoil on g were expected because stomata should close under water stress to conserve water, regardless of the time of day (e.g. Caird et al., 2007). However, the magnitude of the Ψsoil effect often differed between time periods, species and deserts.
The C4 grasses (B.curtipendula, E. lehmanniana and H.contortus) universally exhibited dual stomatal behavior, such that their night-time responses to environmental drivers (D and Ψsoil) differed from their high-light responses, which facilitated the partitioning of the extreme end-member behaviors. However, the slow decay of the dark behavior contribution with increasing PAR suggests that, during most daytime conditions, these C4 plants experience a spectrum between the extreme g behaviors. Moreover, compared with the other species, these species are relatively unresponsive to changes in D, PAR and Ψsoil. Interestingly, the species with the more responsive stomata exhibited behavior specific to their geographic origin. The two Great Basin C3 shrubs exhibited dual stomatal behavior, whereby their dark and high-light responses differed with respect to their reference g and their sensitivities to D and Ψsoil. Although we could partition the two extreme behaviors for the Mojave Desert species, most of these species responded similarly to D and Ψsoil during the daytime and night-time.
Why do our study species generally exhibit different responsiveness to environmental drivers during the night-time relative to the daytime? With respect to the C4 grasses, differential responsiveness to D and Ψsoil may be associated with the relative control of photosynthetic feedbacks and the regulation of g to maintain leaf internal CO2 concentrations during the daytime (e.g. Wong et al., 1978; Huxman & Monson, 2003). As the optimization of carbon uptake is not relevant at night-time for C3 and C4 species, night-time stomatal regulation is expected to directly affect water balance, independent of carbon balance. The three C4 grasses were more responsive to environmental drivers during the day, suggesting that the optimization of carbon gain may be more important. By contrast, the Great Basin and Mojave species generally appear to be more responsive during the night-time, suggesting a comparatively greater importance of controlling water balance, potentially to optimize carbon balance over longer time-scales.
The aforementioned stomatal behavior characteristics suggest fundamental differences between C4 grasses and C3 shrubs, but such differences could also be attributed to their geographic associations. Plants in the Mojave Desert, the driest of the four deserts (Reynolds et al., 2004), exhibited g behavior that was most sensitive to D, perhaps indicating the need for the greater conservation of water. Plants in the Sonoran and Chihuahuan Deserts, which receive notable summer rains (Reynolds et al., 2004), were comparatively insensitive to D, indicating potential alleviation of growing season water stress. The greater diurnal swings in temperature and D in the high-elevation Great Basin may explain why plants in this desert were sensitive to D during the daytime (high D), but were relatively insensitive to D at night (low D) (Gillespie & Loik, 2004). In summary, geographic location, photosynthetic pathway and life form all appear to be potentially important for understanding gday and gnight behavior.
Exploring isohydric and anisohydric behavior during the day and night
Here, we used the normalized sensitivity (S = −m/b) as an index of the potential for a species to possess isohydric vs anisohydric behavior (e.g. Oren et al., 1999; Katul et al., 2009). Seven of the 12 study species exhibited apparent anisohydric behavior (S < 0.6) under high light, which is somewhat expected for desert plants (Oren et al., 1999; Ogle & Reynolds, 2002). Surprisingly, five species possessed apparent isohydric behavior (S > 0.6) under high light, similar to species from wetter environments (Oren et al., 1999), but the relatively high uncertainty associated with S for P. rigida and A. tridentata does not rule out anisohydry (Table S2). The two Great Basin shrubs and a Sonoran C4 grass were clearly associated with two temporally distinct behaviors analogous to an anisohydric strategy under high light and an isohydric strategy at night (Table S2, Fig. 6). Sap flux data from silver birch (Betula pendula) also indicate that nocturnal canopy conductance may be more sensitive to D at night than during the daytime (Sellin & Lubenets, 2010). Moreover, Barbour & Buckley (2007) evaluated the stomatal behavior of castor bean and observed that the sensitivity to D of gnight relative to gday depended on prior growth chamber D conditions. This could partially explain the dual stomatal behavior observed here, because D preceding the g measurements probably differed during the daytime vs night-time periods.
For those species that exhibited significant differences in night-time vs daytime S, we ask: why might stomata be hypersensitive to D at night? This could potentially reflect amplified ‘wrong-way’ responses and oscillatory behavior that have been observed in isohydric plants (Buckley, 2005). That is, stomata appeared to respond to changes in D, but they may overshoot their target aperture because other positive or negative feedback mechanisms (e.g. Jones & Sutherland, 1991; Buckley, 2005) may be lacking or muted during the night. The importance of different feedback mechanisms for gnight has not been rigorously evaluated (but, see Mott & Peak (2010)), and experiments addressing this would lend insight into the mechanisms underlying gnight behavior.
The shifting behavior reported here suggests that gnight and gday may be under different selection pressures (Christman et al., 2008). Katul et al. (2009) conducted a theoretical analysis, which predicted that S should vary between 0.5 and 0.7 to optimize daytime carbon gain and water use. Our study predicted notable differences between daytime and night-time S with respect to the median (three species) and the uncertainty (all species) in the predicted S, indicating that the daytime optimization strategy is not appropriate for describing gnight. Thus, we ask: what is the physiological strategy underlying gnight behavior? Although empirical studies provide some insight (e.g. Caird et al., 2007), theoretical analyses focus on gday (e.g. Medlyn et al., 2011), and the coupling of empirical and analytical work is needed to explore nocturnal strategies.
We implemented a novel analytical approach to partition gnight and gday behaviors, which allowed us to show that, within a given species and site, gnight may exhibit both isohydric and anisohydric behavior. We suggest that future studies should explore the mechanisms associated with differential gnight vs gday responses to environmental drivers, and determine the factors underlying exaggerated isohydric behavior at night. It was beyond the scope of this study to evaluate the direct physiological or genetic mechanisms underlying gnight, and it remains unclear whether gnight is an adaptive trait or under indirect selection (Caird et al., 2007; Christman et al., 2008). This study, however, demonstrated that stomata behave differently during the day relative to the night for some North American desert species, suggesting that gnight and gday may be under different selection pressures.
We thank C. Bentley, W. Cable, N. English, D. Potts, N. van Gestel, A. Eilts, H. Alpert and D. Charlet for their contributions to the field experiments, data collection or data processing. This work was funded by a Department of Energy (DOE) National Institute for Climate Change Research (NICCR) grant to K. O., T. E. H., M. E. L., S. D. S. and D. T. T.; K. O. and J. M. C. were supported by the National Science Foundation (EPS-0447681). L. P. B. was supported by the United States Environmental Protection Agency (EPA) under the Greater Research Opportunities (GRO) Graduate Program. D. T. T. was supported by a US National Park Service grant. The Philecology Foundation of Fort Worth, Texas provided additional support.