Introduction

Amazon forests play a central role in regional and global climate regulation due to high fluxes of water (Nobre et al. 2009) and carbon (Aragão et al. 2014). Climate models predict an increase in the extremes of hydrological cycling in the region (Cox et al. 2004;Chiang et al. 2021), and indeed, both droughts and floods have been increasing in frequency since the 1970s (Marengo and Espinoza 2016), emphasizing the need to better understand Amazon forest responses in terms of key metrics of plant performance in terms of wateruse strategies (Nepstad et al. 2007;Santos et al. 2018;Smith et al. 2020;Esquivel-Muelbert et al. 2020).

Predicting organismal response to stress based on functional traits or ecological strategies has been a research priority in recent decades, especially for plant communities (McGill et al. 2006;Poorter and Bongers 2006). A major research focus has been on predicting vegetation response to drought conditions, by classifying plant water-use strategies along a spectrum from stringent to weak regulation of water status. Within this context, a widely used classification metric considers stringent regulation to be 'isohydric' (wherein leaf stomata tend to close with diel or seasonal drying, maintaining steady leaf water potentials, L , even under increases in atmospheric vapor pressure deficit (VPD), or reductions in soil moisture) and weak regulation to be 'anisohydric' (wherein stomata remain more open, allowing L to drop to very negative values) (Tardieu and Simonneau 1998). The idea has been that this stomatal regulation spectrum reflects a trade-off between inherent species-specific strategies of either accepting limits on photosynthetic carbon uptake in exchange for low risk of mortality from hydraulic failure (on the stringent isohydric end of the spectrum) or allowing expansive photosynthetic uptake that comes at the cost of greater mortality risk (at the weak stomatal regulation, anisohydric end) (McDowell et al. 2008;Rowland et al. 2015;Anderegg et al. 2018;Wang et al. 2020).

Although several efforts have been made to define a quantitative useful metric for this purpose (Martínez-Vilalta et al. 2014;Klein 2014;Fu and Meinzer 2018), recent studies have questioned the classification of water-use strategies within the isohydric-anisohydric spectrum framework, arguing that species responses may not reflect inherent conserved strategies or species-specific traits (Martínez-Vilalta and Garcia-Forner 2017;Hochberg et al. 2018). This critique arises from the observation that environmental factors strongly influence classification metrics, leading to the identification of different strategies for the same species (Hochberg et al. 2018). Consequently, this critique suggests that iso-anisohydry classification may be a poor proxy of plant species vulnerability to drought (Martínez-Vilalta and Garcia-Forner 2017).

Missing from the literature is an understanding of the interaction between iso-anisohydric water-use strategies and the variation in key environmental drivers (such as atmospheric water deficit and soil water availability) that shape hydrologic niches that in turn give rise to those strategies (Song et al. 2024). Here, to address this critique and fill the associated knowledge gap, we investigated whether a more holistic characterization of plant hydraulic strategies along the isohydric spectrum, one grounded in the environmental context that drives that spectrum (as suggested by Kannenberg et al. 2022), could preserve its usefulness for predicting tree response to drought in tropical forests.

We focus on tropical trees, with a synergistic approach that extends beyond measures solely based on water potential sensitivity (see Hochberg et al. 2018), to also incorporate paired physiological responses of g s and L as they are influenced by tree rooting depth. We emphasize the twin environmental factors of soil moisture and atmospheric vapor deficit because these are the primary means by which trees experience drought (Cai et al. 2022) and because they are critical interacting drivers of leaf and canopy-scale conductance for tropical tree communities (Barros et al. 2019;Fang et al. 2021). Additionally, atmospheric vapor and soil water limitations are intrinsically linked to tree size variation, a critical axis of hydraulic strategy that structures drought response in tropical forests: from drought avoidance (achieved by growing tall and developing deep roots to access deep soil water but greater vulnerability to xylem embolisms) to drought tolerance (achieved by strong resistance to xylem embolisms, a trait associated with smaller species with shallow roots) (Giardina et al. 2018;Brum et al. 2019;Chitra-Tarak et al. 2021).

Understanding how the diversity of water-use strategies exhibited by Amazonian tree species is structured by heightassociated environments is important as a hypothetical dominance of one strategy over others would have a strong impact on ecosystem processes (Roman et al. 2015). Given that vulnerability to hydraulic failure increases with tree height (Brum et al. 2019) and the higher evaporative demand in the upper canopy drives transpiration rates (Fang et al. 2021), these conditions can be buffered by deep roots (Martínez-Vilalta and Garcia-Forner 2017).

To address these issues, we pose the question: how can environmental gradients in exposure to the twin dimensions of drought-atmospheric drought (quantified by VPD) and soil drought (quantified by soil water content [SWC])-jointly structure the arrangement of plant water-use strategies (in terms of the degree of isohydricity) along the vertical profile of the forest? We hypothesize that species distributed along the vertical forest profile would display a continuum of water-use strategies, as follows (Fig. 1): upper canopy tree species are more isohydric, with higher diurnal and seasonal variation in stomatal conductance and less variation of leaf water potential than middle canopy and understory trees. Despite being less exposed to seasonal soil water stress due to their deeper roots, upper canopy trees are expected to experience higher VPD and higher light interception. Although more susceptible to seasonal soil water stress due to their shallow roots, middle canopy and understory tree species are buffered from changes in VPD, so they are expected to reach lower L than upper canopy tree species in order to maximize the extraction of water from a limited soil volume.

Materials and methods

Study area

This study was carried out in the Tapajós National Forest (54 • 58 W, 2 • 510 S), close to the K67 eddy covariance tower and research site established by the Large-Scale Biosphere-Atmosphere Experiment in Amazonia (LBA). The vegetation is a dense terra-firme rainforest (IBGE 2015) with a high diversity of tree species, about 265 cataloged at 20 ha (Vieira et al. 2004;Pyle et al. 2008). The forest displays a closed canopy with a maximum height of 40 to 45 m, which is vertically structured with a high density of small trees growing in the understory, and middle canopy trees attaining about 20 m (Rice et al. 2004;Vieira et al. 2004;Stark et al. 2012;Smith et al. 2019). The climate is strongly seasonal compared with other forests across the equatorial Amazon, with a rainfall regime (mean annual precipitation ∼2000 mm) defined by a wet season and up to ∼5 consecutive months with precipitation less than 100 mm month -1 and mean annual air temperature 25.3 • C (Restrepo-Coupe et al. 2013, 2017). The area has deep soil, characterized as Dystrophic Yellow Latosol with a high fraction of kaolinite clay minerals, welldrained, presenting a moderate surface horizon (Nepstad et al. 2002).

To test our hypothesis (Fig. 1), we selected trees of species occupying different canopy positions within the vertical structure of the forest, and we estimated their effective functional rooting depth (EFRD) as a proxy for their ability to access water at different depths within the soil. We measured stomatal conductance and leaf water potential to assess wateruse strategies according to metrics of the iso-anisohydric Polygon widths illustrate how the range of variability of above and belowground environments depends on vertical canopy position, while color within each polygon indicates relative water availability (upper canopy atmospheres are more variable, and tend to be drier, than understory, while deep soil water is more stable and wetter, than shallow). The general hypothesis is that upper canopy tree species are exposed to greater VPD variability but can access less variable abundant water supplies via deep roots and hence tend to adopt more isohydric strategies, with higher daily and seasonal stomatal regulation; understory tree species, in contrast, are exposed to lower and more stable atmospheric VPD, but to greater water variability and seasonal drought in the soil, and hence tend to adopt more anisohydric strategies with less stomatal regulation across the seasons. Mid-canopy trees, exposed to intermediate environments, are expected to adopt intermediate strategies along the isohydric to anisohydric spectrum.

framework. We also measured the soil and atmospheric environmental metrics to which these species were subjected to understand how the access to soil water and the variations of environmental conditions on forest vertical profile can be related to water-use strategies.

Canopy vertical structure, species selection, rooting depth and canopy access

We determined the vertical forest structure based on the light environment and tree height (Stark et al. 2015;Smith et al. 2019). Three forest strata were defined as: (i) upper canopy, where trees are under high direct sunlight exposure, ranging from 23 to 38 m tall; (ii) middle canopy, where trees have their canopy exposed to light that crossed through the upper layer, ranging from 11 to 14 m tall; and (iii) understory, where tree canopy exposure is constrained by canopy and middle canopy trees, occurring under lower light availability, ranging from 1.5 to 3 m tall (Table 1).

Nine tree species occurring along the vertical forest strata were selected among the most abundant and dominant tree species within a 4-ha plot, representing 38% of the total basal area, except the two understory species (Table 1; Pyle et al. 2008;Longo 2014). All individuals sampled were mature trees with their respective light niche attained (e.g. all individuals of middle canopy are not expected to achieve upper canopy status, and all individuals of the understory are not expected to achieve middle canopy status). We estimate the EFRD for each tree (Table 1) according to a model derived by Brum et al. (2019), which considers the diameter at breast height (DBH) as a predictor of root depth (details in Methods S1 available as Supplementary data at Tree Physiology Online).

We accessed the upper canopy trees (n = 5 species, one individual per species) using an elevated walkway installed at canopy trees about 30 m above the ground and through a 45-m high platform tower installed in the forest. We accessed the middle canopy leaves (n = 2 species, one individual per species) by climbing using arborism techniques. The understory trees (n = 2 species, one individual per species) were accessed from the ground. The intricate task of reaching canopy and mid-canopy trees hindered our ability to replicate species. While it was feasible to access understory trees from the ground, the endeavor to obtain diurnal curves throughout the day for multiple individuals of each species presented a formidable challenge. Furthermore, even during the dry season, intermittent rain events were frequent, which hindered a comprehensive sampling of stomatal conductance and leaf water potential measurements at hourly intervals across various seasons. Due to these constraints, we focused on a replicated study of forest canopy strata (with replication of n = 5, 2 and 2 tree species for the upper, middle and lower strata, respectively, as described above), making no statistical claims about interspecies differences since each of the nine species was represented by one individual. We note that when multiple leaves were measured per individual, these separate leaf measurements were pooled to one average value per replicate tree to avoid pseudoreplication.

Sampling period and climatological conditions

To evaluate the physiological responses under different precipitation conditions along the year, our sampling campaigns from 2016 to 2017 (see Fig. S1a and b available as Supplementary data at Tree Physiology Online) encompassed: the peak of the rainy season or wet season (May 2017; cumulative water deficit, CWD = 0 mm), the transition from wet to dry season or end-wet season (July 2016; CWD = 178.9 mm), peak of the dry season or dry season (August and September 2016; CWD = 220.8-243.9 mm) and the transition to the rainy season or end-dry season (December 2016; CWD = 33.0 mm, see Fig. S1b available as Supplementary data at Tree Physiology Online).

Ecosystem evapotranspiration and microclimate variables were obtained through the eddy covariance system installed at the K67 LBA flux-tower since 2001 (see Fig. S1c-f available as Supplementary data at Tree Physiology Online). Meteorological records were averaged to daily values to match our 2016 to 2017 field sampling (see Fig. S2a-d available as Supplementary data at Tree Physiology Online). Monthly precipitation obtained from the Tropical Rainfall Measuring Mission (TRMM 3B43-v7; 1998 to 2018) (Huffman et al. 2007) was used to obtain monthly values of CWD (details in Methods S2 available as Supplementary data at Tree Physiology Online; Restrepo-Coupe et al. 2023).

To evaluate how atmospheric demand varies along the canopy vertical profile, we estimated VPD in the canopy air space around each measured tree. The VPD was calculated from relative humidity (RH; %) and air temperature (T; • C) measurements recorded every 5 min during field sampling by HOBO sensors (HOBO Pro v2, U23-001 Onset, USA) installed on branches adjacent to measured leaves. We 1 Average canopy light exposure expressed as individual crown position index categories as adapted by Clark and Clark (1992) from Dawkins and Field (1978) , where: '1: No direct light (crown is not lit directly vertically or laterally); 1.5: Low lateral light; 2: Medium lateral light; 2.5: High lateral light; 3: Some vertical light (10% to 90% of the vertical projection of the crown is exposed to vertical light); 4: Crown completely exposed to vertical light, lateral light blocked within some, or all of the 90 • inverted cone encompassing the crown; 5: Crown completely exposed to vertical and lateral light within the 90 • inverted cone encompassing the crown.' 2 Average values for abundance were calculated by extracting individual species data (DBH, cm for all live stems ≥10 cm; year 2012) from a long-term forest tree inventory database (Pyle et al. 2008 ;Longo 2014 ); n = 4, 5 ha plots replicate for all species, except for understory species (n = 5, 0.05 ha plots; all live individuals > 0.30 m). To assess the variation in the VWC across different profiles throughout the season, we used data from horizontally installed sensors at depths of 0.50, 1, 2, 3, 4, 5, 7 and 10 m in two different soil pits. These sensors use a timedomain reflectometer (TDR) system produced by Campbell Scientific (TDR 100), logged to multiplexers (SDMX50), and various lengths of low-loss RG8 coaxial cables. We used data collection across seasons from 2008 to 2016 to describe the overall seasonality effect in the soil profile (Restrepo-Coupe et al. 2023).

Leaf water potential ( L ) and stomatal conductance (g s )

We measured paired diurnal curves of L and g s under natural field conditions roughly hourly between 5 AM and 7 PM (n = 15 time points per measurement day). Due to the high temporal resolution of the data collection and the logistical challenge of rapid access to multiple canopy trees (which precluded completion of curves for all individuals in each strata), we selected one individual of each species per measurement day, for days representative of each of four seasons across the year, as previously explained. For each individual and each time point, we selected five mature leaves from the upper position of the crown, whose measurements were pooled to arrive at one average value for that individual at that time point. We consider mature leaves located on either the third or fourth nodes, from the apex of the branch, with color, expansion and consistency characteristic of full development.

Furthermore, we conducted a comparative analysis of our L results with the measurements of L obtained during the severe drought of 2015 attributed to El Niño, as previously reported by Brum et al. 2019. This comparison allowed us to assess our dataset to the ENSO minimum L ( min ) values, which were averaged for each forest and represented in some of our figures.

We measured L using a pressure chamber (PMS Instrument Company, Corvallis, OR, USA, model 100) (Scholander et al. 1965). The leaves were collected and inserted in black plastic bags with moistened paper to prevent transpiration, totaling 15 hourly measurements on five replicate leaves = 75 leaves/day/individual (except during very rainy days). Within 5 min of collection, L was measured in the field. In each season, the maximum L of the day for each tree was measured between 5 and 7 AM (named as predawn water potential, pd ), while the min was considered as the lowest value of the day for each tree measured between 11 AM and 4 PM in that season.

We measured g s using a porometer (Decagon Devices, Pullman, WA, USA). The porometer chamber assumes the environmental radiation, wind speed and air relative humidity at the local condition within the canopy in the day of measurement, in automatic mode (standard measurement time for all leaves) applied to leaves nearly adjacent to the leaves being measured for L . Four measurements were taken on the abaxial face of each leaf (two on the right, at the base and at the apex, and two on the left at the base and at the apex of the leaf blade), and then averaged to a single value for that time. The g s measurements were made in situ, while leaves remained connected to the branches; the leaves used in the first g s measurement were monitored throughout the day. In each season, leaves with the same criteria (position along the branch and visual characteristics) were selected to perform the curves.

To compare stomatal regulation strategies among species with different absolute values of g s , we normalized conductance measurements by dividing each species' timeseries of mean hourly g s (each hour being an average of replicate measurements taken during that hour) by the maximum of all hourly g s values for that species (g s max ), i.e. g s norm = g s /g s max .

The result was a normalized g s norm time series with values between 0 (conductance zero) and 1 (conductance maximum for that species) (Aksoy and Haralick 2001).

Interaction between g s and L and vertical profile effects on trees isohydricity

We used the diurnal curves of g s norm and L across seasons to classify the water-use strategies along a spectrum of isohydric to anisohydric, according to different metrics and an extensive literature review (see Table 2 for references and criteria). Accordingly, we computed a parameter aimed at condensing the diurnal variation of g s norm into a single metric. Here, we computed the absolute difference between the individual mean afternoon g s norm (12 to 5 PM) and mean morning g s norm (7 to 11 AM), which we refer as g s norm . This metric effectively describes the degree of daily stomatal regulation (Scholz et al. 2012), where low or negative values indicate a more stomatal resistance in the afternoon in relation to early morning. Additionally, we conducted an analysis to investigate how the difference between pd and min in a particular individual (i.e. daily changes in leaf water potential or pd-min ; as described by Klein 2014 and Tardieu and Simonneau 1998), influences the diurnal course of g s norm , denoted as the absolute difference between afternoon g s norm and morning g s norm ( g s norm ), across the seasons (for more information, see Fig. S4 available as Supplementary data at Tree Physiology Online). Finally, we calculated the slope value by examining the relationship between daily pd and min at the individual level across various seasons, indicated by the sigma values. This approach follows the methodology outlined by Martínez-Vilalta et al. (2014) (for more information, see Fig. S5 available as Supplementary data at Tree Physiology Online). Due to the structure of experimental data, which involves measurement over time on the same individual, we conducted measurements on multiple individual leaves per tree, which were then averaged to obtain an average value for each replicate tree at each measurement time. Statistics were based on measurements across replicate trees within canopy strata to ensure the precision and validation of the results, and appropriate statistical analyses were conducted.

Statistical analysis

To understand how water-use strategies relate to forest structure and daily environmental conditions, we correlated g s norm and L as a function of canopy position (strata) and time (hour of the day) using a Linear Mixed Model (LMM). We set strata and hour of the day as fixed effects, species as random factors, and their interaction considering repeated measurements over time. In this analysis, the sampling unit was a tree of a given species at a given time, ∼n = 310 (nine trees, each measurement 15 times over hour-of-day for four times over year). The hour-of-day was the variable for repeated measurement and each species was considered a random effect of covariation between the observations of the same subject at different times (Littell et al. 1998;Blackwell et al. 2006).

We analyzed how daily changes in leaf water potential ( pd-min ) and min (predictors) affected the absolute difference between afternoon g s norm and morning g s norm ( g s norm ; response) as a function of canopy position (strata). We employed an LMM with measurements repeated over time. We treated the interaction between physiological variables and strata as fixed factors and species as random factors. In this analysis, the sampling unit was a tree of a given species at a given time, n = 33 (nine tree, each measurement four times over year-see details of the obtention of g s norm in the previous topic 'Interaction between g s and L and vertical profile effects on trees isohydricity').

Our analysis was conducted at the level of individual leaf data points, where each point corresponds to the mean value for a particular individual of a specific species according to your position in the vertical profile on each day of measurement. All analyses were performed using the SAS software (version 9.4), while graphics were generated in R (version 3.6.0). We used the PROC MIXED syntax in the LMM to account for repeated measures that implement the covariance structure of the data (Littell et al. 1998). See details in Methods S3 available as Supplementary data at Tree Physiology Online.

To test for differences among canopy positions along the vertical profile, we performed a t-test to identify mean differences in seasonal changes of VPD ( VPD) and min ( min seasonal ) by strata during 2016 and 2017, where VPD = VPD maximum, dry season -VPD maximum, wet season; and min seasonal = min wetmin end-dry . In order to increase the statistical power of these analyses, we classified the species from the understory and middle canopy into subcanopy groups. Therefore, the lower canopy corresponded to n = 8, while the upper canopy corresponded to n = 5.

To evaluate the relation between the position of individuals along the vertical profile and water-use strategy, we performed Pearson correlation tests between min seasonal and tree height (h; m), and between min seasonal and the estimated rooting depth for each tree according to tree DBH. To evaluate the relationship of physiological responses to environmental conditions, we fitted multiple generalized linear models using min seasonal as response variable, and the interaction between VPD and the average VWC measured during the dry season (a representation of the minimal VWC that tree roots experience in the vertical soil profile) as the predictive variables.

Results

Differences in atmospheric demand (VPD) across the forest vertical profile and water content (VWC) variability across the soil profile

The vertical forest profile in the Tapajos forest presents a gradient in VPD variation, ranging from pronounced fluctuation in the upper canopy to more stable conditions in the understory (Fig. 2a). This gradient becomes more accentuated as the dry-season advances, with VPD reaching mid-afternoon values of 3 kPa toward the end-dry period. In contrast, the middle canopy and understory layer experienced a smaller increase in dry-season VPD in the afternoon (∼0.5 kPa). During the wet season, VPD displays minimal diurnal variation across all forest strata ranging over 1.5, 0.7 and 0.15 kPa for upper canopy, middle canopy and understory, respectively ( Fig. 2a). There is a more seasonal variation in soil VWC in the shallow soil (above 2.5 m depth, there was a decrease of ∼0.04 cm 3 cm -3 between the wet and dry seasons), compared with more stable soil VWC condition in the deeper soil (below 3 m depth, there was minimal change across wet-dry season, 0.007 cm 3 cm -3 ) (Fig. 3). These observations suggest that upper canopy, middle canopy and understory trees experience different levels of seasonal variation in soil VWC. Upper canopy trees, with deeper roots reaching between 1.5 m and 12.6 m (Table 1), can access the less variable deep soil water reservoir, where annual average VWC was 0.32 cm 3 cm -3 . In contrast, shallow-rooted understory and middle canopy trees, with roots close to the surface, ranging between 0.30 and 1 m (Table 1), encounter higher seasonal variation and lower average water availability.

Diurnal and seasonal patterns of L and g s across the forest vertical profile

Leaf water potential ( L ). We found a statistical significant interaction between the hour of the measurement and the species' vertical stratum, and seasonal L variability (Fig. 2b, Table 3, Results S1 available as Supplementary data at Tree Physiology Online). Each stratum exhibited a distinct daily pattern in L (Fig. 2b). While the seasonal diurnal cycles of L remained relatively consistent for upper canopy trees, understory trees displayed greater seasonal variation (Tukey-Kramer post-hoc test, hereinafter PostHoc test P < 0.05). The intensity and duration of the L reduction varied among strata, primarily depending on the time of day, but it was only notable during the wet season (Fig. 2b). We observed significant differences between upper canopy and middle canopy, as well as between the upper canopy and understory, during the wet season (PostHoc test P < 0.005). Notably, the L of upper canopy trees was significantly more negative in the late morning (10 to 11 AM) and in the afternoon period (12 to 4 PM) when compared with trees from other strata (Fig. 2b).

The average min for the upper canopy trees was relatively stable (around -2.0 MPa across seasons; Fig. 2b, Fig. S3a available as Supplementary data at Tree Physiology Online). In contrast, there was a decrease in min during the driest season for middle canopy trees (-1.07 to -1.72 MPa) and a sharply decrease for understory trees (-0.83 to -1.94 MPa) (Fig. 2b). During an extreme drought event induced by 2015 ENSO, the average min of upper canopy trees was similar to Table 3. Result of linear mixed model results for L and g s norm as variable response and strata and hour of day as predictor. Model test for interactions between predictors for each season. Models included species identity as a random factor to account for repeated measures of the same species. Statistically supported effects are indicated in bold (P < 0.05).

Variable/ Season

Effect

DF F-value Pr > F L wet Strata 2 19.57 <0.0001 Hour 14 25.17 <0.0001 Strata × Hour 24 9.27 <0.0001 L end-wet Strata 2 1.38 0.25 Hour 13 35.97 <0.0001 Strata × Hour 24 4.71 <0.0001 L dry Strata 2 1.43 0.24 Hour 13 28.19 <0.0001 Strata × Hour 22 1.97 0.0069 L end-dry Strata 2 1.03 0.3585 Hour 12 91.49 <0.0001 Strata × Hour 21 6.08 <0.0001 g s norm wet Strata 2 3.24 0.0407 Hour 14 12.60 <0.0001 Strata × Hour 24 2.68 <0.0001 g s norm end-wet Strata 2 1.29 0.2781 Hour 13 7.54 <0.0001 Strata × Hour 24 3.80 <0.0001 g s norm dry Strata 2 1.34 0.2632 Hour 13 8.70 <0.0001 Strata × Hour 22 1.50 0.0726 g s norm end-dry Strata 2 1.23 0.2941 Hour 12 12.02 <0.0001 Strata × Hour 21 0.65 0.8757

our study ( -1.9 MPa). Conversely, the min of middle canopy and understory trees decreased compared with the preceding dry season (-2.5 and -3.0 MPa, respectively) (Fig. 2b).

Stomatal conductance (g s norm )

. We observed differences in g s norm variability across different forest strata exclusively during the wet season (LMM, F = 3.24; P < 0.04). There was an hourly effect determining the g s norm rates across all seasons under investigation (Fig. 2c, Table 3). Furthermore, during the wet season, the interaction effect between forest strata and hour was also significant (F = 2.68; P < 0.0001), as was the case during the transition from dry to wet season (F = 3.80; P < 0.0001).

Regarding the strata effect on g s norm , middle canopy and understory trees differed only in the wet season (PostHoc test P < 0.05). During this season, the diurnal g s norm of middle canopy trees reached 70% of the maximum g s norm , higher than the understory trees that were 44% of the maximum g s norm (Fig. 2c, Fig. S3b available as Supplementary data at Tree Physiology Online). The hourly effect, considering all strata together, was significant only in the wet season (Table 3). At 6 AM, g s norm was higher, with the mean g s norm for all strata being 72% of the maximum g s norm (PostHoc test P < 0.0001).

When considering the effect of the interaction between hour and strata on g s norm variation, we observed that forest strata explain the diurnal g s norm variation only in the wet and endwet season (PostHoc test P < 0.05). The upper canopy trees showed the greatest variation in the diurnal g s norm in both of these seasons, with the maximum g s norm rates occurring between 6 and 7 AM (PostHoc test P < 0.0001). The middle canopy trees presented relatively constant g s norm throughout the day (Fig. 2c). During the wet season, differences in g s norm occurred only between 7 AM to 2 PM, and in the late afternoon (6 PM), there was 45% reduction of maximum g s norm (PostHoc test P < 0.04). In the end-wet season, statistical differences were observed only between 7 AM and Table 4. Results of linear mixed models for gs norm as response variable (n = 33). Separate models were created, one using min as predictor (Model 1), and another one using pd-min as predictor (Model 2). Models tested for interactions between these predictors and forest stratum (ordinal variable: understory < middle canopy < upper canopy). Models included species identity as a random factor to account for repeated measures of the same species. However, in both models, this term was not statistically supported (Model 1: F = 0.19, P = 0.32; Model 2: F = 0.00, P = 0.55). Statistically supported effects are indicated in bold (P < 0.05).

Model

R adj 2 Model term Coefficient t P Model 1 0.26 Intercept 0.356 ND ND min 0.217 3.224 0.003 Stratum -0.200 -3.356 0.002 min × Stratum -0.102 -3.076 0.004 Model 2 0.19 Intercept 0.250 ND ND pd -min -0.207 -3.094 0.004 Stratum -0.155 -3.229 0.003 pd -min × Stratum 0.100 2.870 0.007

4 PM, when g s norm reduced 52% of maximum (PostHoc test P < 0.05). With respect to the understory, the g s norm was higher between 6 and 7 AM relative to the other periods of the day during the wet season ( Fig. 2c). However, in the endwet season, the diurnal curve of g s norm showed a parabolic pattern: there was a gradual increase in g s norm from 6 AM until reaching the maximum g s norm at 1 PM, followed by a gradual reduction in late afternoon (Fig. 2c).

Interaction between g s and L and vertical profile effects on trees isohydricity

We observed that the range of stomatal regulation through the day (stringency of g s norm to the afternoon in relation to morning) in response to the variation of the leaf water potential is explained by canopy position on the forest vertical profile (Table 4). The g s norm decreased with diurnal leaf pd-min and min in upper canopy trees (Fig. 4a and b). In contrast, this trend was the opposite for understory trees; the decrease in leaf water potential was related to a less stomatal regulation during the day across season. The g s norm increased with the increase in pd-min ; similarly, the decrease in min was correlated to a less stomatal regulation (less change in g s norm ) (Fig. 4a and b). We did not observe any tendency of increase or reduction considering the relationship between g s norm and pd-min , or g s norm and min for middle canopy trees (Fig. 4).

water-use strategies across the forest vertical profile and environmental drivers

The range of variation in VPD values between the dry and wet season ( VPD dry-wet ) was not uniform across the vertical forest profile, with values ranging from 1.74 kPa (SE ± 0.36) in the upper canopy (2.5 kPa considering the contrast with late dry season), to around 0.5 kPa in the middle canopy and understory layer. The VPD dry-wet values were significantly different between canopy environments, upper canopy and subcanopy layers (encompassing understory and middle-canopy layers) (T test; t = -3.56, df = 4.95; P < 0.01; Fig. 5a). Interestingly, the min wet-dry was lower for upper canopy trees, higher valuer for middle canopy trees (0.5 MPa) and greatest for understory trees (1.0 MPa). This trend was particularly pronounced during the ENSO year of 2015 to 2016 (Fig. 5c). In fact, the min (wet-dry) was negatively correlated with the tree height (Fig. 5d) and rooting depth (Fig. 5e). Additionally, subcanopy trees (encompassing understory and middle-canopy) presented higher average min (wet-dry) than upper canopy trees (T test; t = -3.48, df = 4.6; P < 0.05).

The seasonal average of pd-min of upper canopy trees was 1.44 MPa, while middle canopy trees varied from 0.57 MPa in wet season to 1.01 MPa during the enddry. In contrast, understory trees showed higher seasonal variations, spanning from 0.35 to 1.34 MPa (see Fig. S4a available as Supplementary data at Tree Physiology Online). The ( pd-min ) seasonal was strongly related with the tree height (R = -0.82; P = 0.006; see Fig. S4c available as Supplementary data at Tree Physiology Online). Furthermore, when examining the relationship between pd and min for each stratum, differences in their control became apparent (Fig. 6). The slope for upper and middle canopy trees (σ = 0.5 MPa/MPa) was significantly lower than for understory trees (σ = 1.5 MPa/MPa, Fig. 6).

According to generalized linear model, we found that VPD did not have an effect on the min wet-dry (Analysis of Deviance-Type II test; Chisq = 0.35; P = 0.55; intercept = 0.922; β = 4.057). However, the long-term average VWC measured during the dry season for each rooting depth demonstrates a significant impact on the min wet-dry (Chisq = 7.17; P = 0.007; β = -3.467). Additionally, there was a marginally significant interaction between VPD and long-term average VWC measured during the dry season at each rooting depth (Chisq = 3.796; P = 0.051; β = -13.077; Fig. 5f).

Discussion

The key finding of this study is in line with our hypotheses (Fig. 1), showing that water-use strategy diverges across tree canopy profiles at Tapajós National Forest. Upper canopy trees experiencing greater and more variable atmospheric demand as represented by seasonal VPD (Fig. 5a) but less variation in soil water availability across seasons, due to deeper roots that can access larger and more stable water resources deeper in the soil (Figs 4 and 5b). In contrast, middle canopy and understory trees experience less atmospheric demand but more variability in available soil moisture (Fig. 5a and b). These contrasting environments appear to select for divergent water-use strategies along the forest profile (Fig. 5e and f). Upper canopy trees follow an isohydric strategy by exercising greater stomatal control to achieve more stable L (Table 1, Fig. 4), while understory trees follow an anisohydric strategy, allowing large dry season declines in L , which declines even more during anomalous droughts ( Fig. 5c). Middle canopy species exhibited a third, intermediate strategy, tending neither isohydric nor anisohydric, with daily changes in g s uncoupled from changes in L (Fig. 4). The divergence in water-use patterns across tree size structures within canopy profiles (large tree-upper canopy to small tree-understory) unveils a crucial insight into forest function, carrying implications for both conceptual understanding and practical applications. Conceptually, this work provides additional evidence to support that the classification along the isohydric-anysohidric spectrum arises from trait-environments interactions, once tree hydraulic strategies can be selected and structured with respect to the environmental variability which specific tree occurs (Feng et al. 2018;Kannenberg et al. 2022). Practically, our work should be important for modeling studies, because the finding that different parts of the forest will respond differently to climatic changes is an important corrective to 'big leaf' models that treat the forest homogeneously. It also should help inform development and parameterization of the newer generation of trait-based models (Fisher et al. 2018;Meunier et al. 2021), by characterizing how plant stomatal and water potential functions are jointly constrained under VPD and soil water variation across different strata in the critical tropical forest biome.

In addition to that, it is of utmost importance to acknowledge certain limitations in our study of the divergent water-use strategies among canopy tree species. One significant factor to consider is the potential influence of different environmental conditions of the VPD in our site research between lower and upper canopies. The determination of stomatal response of understory species only was possible when VPD was below 1 kPa. The microclimate conditions could potentially affect the overall dynamics of stomatal conductance within the distinct vertical profiles of forests in Tapajós. As a result, it is challenging to accurately estimate understory gas exchange and regulation of leaf water potential regulation in the face of large variations in VPD. Therefore, further investigations are strongly recommended to gain a more comprehensive understanding of the intricate interplay of these environmental variables and their effects on understory tree responses in the seasonal Amazon forest.

The continuum of iso/anisohydric strategies observed in our study showed coordination between above and belowground functionality in vertical forest profiles. Upper canopy trees engaged in more stomatal regulation as long as pd-min was reduced (Fig. 4). Stronger stomatal control in upper canopy trees is a transient regulatory response that decreases photosynthesis during extreme drought (such as that induced by 2015 ENSO, as also observed in a central Amazon forest; Santos et al. 2018) and is a mechanism important for coping with higher seasonal variation in maximum VPD and hydraulic path length (McDowell and Allen 2015;Olson et al. 2018;Bittencourt et al. 2020;Garcia et al. 2021). Upper canopy trees are usually exposed to higher VPD variability and their isohydric behavior is likely driven by sustained access to more stable soil moisture (Fig. 5e). Therefore, the larger diurnal variation of VPD across seasons could explain the higher pd-min experienced for the upper strata, while the maintenance of low variation of minimum leaf water potential across seasons, min (wet-dry) , was possible due to the access by roots to stable sources of soil moisture. Our data also demonstrated that understory trees, which are subject to low VPD variation, but high soil water variation (Fig. 5a and e), had lax (anisohydric) stomatal regulation but greater drought tolerance with little variation of minimum L across seasons (Fig. 4), especially when under extreme drought, as during the 2015 to 2016 ENSO (Figs 2b and 5b, c).

Regardless of the metric used in this study, the results consistently support the hypothesis (Fig. 1) of divergent wateruse strategies as a function of the ability to access deep soil water and of the atmospheric environment (VPD): trees occupying the upper canopy are isohydric, while those occupying the understory are anisohydric (Figs 2, 4 and 5c). Although some previous studies (e.g. Klein 2014;Hochberg et al. 2018) would classify significant daily changes in leaf water potential ( pd-min ) as anisohydric, this concept only works well as a comparative measure among trees of similar stature. In this study, where adult trees species with significantly different maximum heights are compared, it is important to consider that the required water potential gradient for water flow is different for a tall tree (∼30 m) compared with a small tree (∼2 m). Estimates indicate that to overcome all resistances in hydraulic transport in tall trees, a pressure difference between the root and the leaf of about 3 MPa is necessary (Pimenta 2012). This inevitably requires lower midday leaf water potentials in large trees relative to small ones, independent of wateruse strategy, and so using the daily metric of pd-min as a classification for isohydricity would lead to a bias toward misclassification of large trees as anisohydric. A higher pd-min in the dry season found in upper canopy trees (daily scale, see Fig. S4a available as Supplementary data at Tree Physiology Online) therefore does not reflect an isohydric strategies. This conclusion was possible only because the pd-min in the wet season followed the same pattern as those in the dry season. The variation of pd-min across the wet and dry seasons showed how much the pd-min increases (anysohydric) or decreases (isohydric) from the wet season to the dry season, reflecting the accuracy of the degree of isohydricity; therefore, this metric is recommended only for the seasonal scale. In addition, we also propose to use the metric of isohydricity that takes into account the changes in the minimum leaf water potential (Martínez-Vilalta and Garcia-Forner 2017) for the seasonal scale (between wet and dry seasons)-that is, whether min seasonal (= min wetmin dry ) is small (isohydric) or large (anisohydric). This metric is straightforward to implement and shows consistency as a response trait across trees of different heights (Figs 2b and 5c,d,Fig. S4c available as Supplementary data at Tree Physiology Online).

Our results may help elucidate recent observations from Draper et al. (2021) indicating that hyperdominance is a widespread phenomenon across tropical forest strata, with distinct species dominating the forest understory, middle canopy, and upper canopy. The inherent intra-and interspecific variation in hydraulic traits and water-use strategies contribute to the specificity of niche occupation across vertical forest strata, thereby influencing dominance patterns in each forest level (Olson et al. 2018;Bittencourt et al. 2020;Garcia et al. 2022;Brum et al. 2023). The mechanisms regulating g s (which depends on VPD and hence, vertical canopy position and rooting depth) should be amenable to inclusion in height-or age-structured dynamic vegetation models (e.g. Levine et al. 2016;Longo et al. 2019). In the seasonal Amazon forest, taller tree species with deeper roots possess the capability to alleviate soil stress by tapping into water reserves at greater depths, where soil water depletion is less severe (Brum et al. 2019). However, this advantage comes with trade-offs, including coping with a longer water-transport path from the soil to their leaves and being susceptible to xylem embolisms (Chitra-Tarak et al. 2021). Furthermore, the pronounced variability in VPD within the upper canopy layer adds complexity to the regulation of stomatal conductance, affecting the water-useefficiency (Santos et al. 2018;Ding et al. 2021;Brum et al. 2023). Our findings align with recent evidence that stomatalinduced reductions in leaf-level photosynthesis in the central Amazon are highly sensitive to microclimate conditions in the canopy layer, especially during extreme drought events (Santos et al. 2018). This highlights the intricate interplay between VPD, rooting depth and physiological responses, revealing the potential effects of distinct canopy strata on the sensitivity to rainfall patterns and the seasonal variations of ecosystem gas exchange in these forest ecosystems, as suggested by modeling studies (Ivanov et al. 2012). Yet, we provide empirical evidence about maximization of gas exchange at individual level by understory strata in the forest vertical profile, which might affect large-scale ecosystem processes such as dry season productivity in this diverse forest (Restrepo-Coupe et al. 2013) and the linking of decrease in L tolerance and hydraulic safety margins associated with canopy conductance for a large scale precipitation gradient in Amazon (Barros et al. 2019).

The anisohydric performance of understory species with shallow roots might be related to a strategy to maximize soil water extraction rates from shallow soils with continued increase in g s across the dry season (Fig. 5). Understory tree species tolerate higher declines in their water potential during extreme drought events in seasonal Amazon forest (Brum et al. 2019) and have greater capacity to adjust carbon metabolism through stomatal conductance than large upper canopy tree species, as evidenced by the less pronounced decrease in photosynthesis rates (Nepstad et al. 2007;Ivanov et al. 2012;Santos et al. 2018;Bittencourt et al. 2020). The lower stomatal control and the higher drought tolerance in understory species are important mechanisms for persistence in lowlight environments as they maintain carbon uptake under intermittent understory light conditions. A higher average light availability during the dry season represents a window of opportunity for these trees to maximize productivity (higher stomatal conductance; Fig. 4) especially due to higher light availability in understory gap openings (Restrepo-Coupe et al. 2013;Smith et al. 2019). Our results may help further explain the higher drought tolerance observed at this same site during extreme drought induced by the 2015 to 2016 ENSO (Brum et al. 2019;Fig. 5). The variability in soil water conditions resulting from seasonal precipitation selects shallowrooted species with higher resistance to xylem embolism (more negative P 50, Brum et al. 2019), buffered by low variability VPD (Fig. 5f). In addition, these results help explain the low mortality rates observed for small trees in throughfall exclusion experiments in the Amazon (Nepstad et al. 2007;da Costa et al. 2010;Bartholomew et al. 2020), which have a larger capacity to acclimate their hydraulic systems (Giles et al. 2022).

Conclusion

Here, we used highly resolved diurnal measurements of stomatal conductance paired with leaf water potential, in wet and dry seasons, to identify different water-use strategies of tree species occupying different canopy positions in a seasonal Amazon forest. Our results support the hypothesis that there is a continuum of hydraulic strategies (high leaf stomatal regulation in the canopy versus less regulation in the understory) along the vertical profile of the forest that are structured by shifting hydrologic stressors, ranging from high VPD variance in the canopy to high soil moisture variance in the understory. Despite the high variation in VPD for the upper canopy tree species, the main driver of the isohydric behavior exhibited is, as well as for the behavior anisohydric for understory plants, the degree of seasonal variation of the SWC. Our work shows how tropical forest functional diversity is importantly structured by vertical gradients, and how the concept of the isohydric-anisohydric trait spectrum can be meaningfully preserved by reference to environmental context, in this case with tree canopy position mediating which environmental factors exert the strongest selection on hydraulic traits. This vertical structure implies-in contrast to prevalent 'big leaf' modelsthat distinct forest strata will respond differently to future climate changes and that accurate predictions of tropical forest response to such changes will require integrating a spectrum of traits coordinated by within-canopy environmental gradients.