Maximum water stress is decoupled from climate, traits, and growth in a xeric oak

Summary

• Our ability to explain and predict the spatial patterns of forest mortality remains limited. We tested the predictability of water stress across populations and explored how water availability and plant allocation mediate spatial variation in tree growth in a widespread oak species (Quercus douglasii Hook & Arn.).

• We conducted a range‐wide survey of drought stress, xylem water isotopes, growth, and allocation traits across 15 sites spanning almost 1000 m difference in annual precipitation.

• We found little relationship between end‐of‐season water availability (predawn leaf water potentials) or maximum stress (midday water potentials) and climate or soils. Instead, stem water stable isotopes indicated that water availability was predicted by access to deep water resources derived from large winter storms. We also found a three‐way decoupling of water stress, prior growth, and allocation to leaf tissue, such that simulated hydraulic risk from a mechanistic model was uncorrelated with observed performance.

• Our results reveal that deeply rooted trees can be hydrologically decoupled from aboveground climate, and that the seasonality of growth, trait development, and hydraulic risk are phenologically disconnected. The complicated relationship between carbon gain and hydraulic risk in seasonal environments and limited data on critical zone hydrology are key challenges to predicting drought vulnerability.

We sampled sites across the geographic distribution of blue oak (Quercus douglasii) in California, USA, which spans a large range of mean annual precipitation.

Introduction

Anthropogenic climate change drives more frequent and severe droughts, with widespread ecological consequences such as regional scale forest mortality (Allen et al., ²⁰¹⁰; Brodribb et al., ²⁰²⁰; Hammond et al., ²⁰²²). Two decades of ecophysiological research have revealed that failure of the hydraulic system, interacting with disruptions to carbon metabolism–particularly for defense against biotic agents–drives tree mortality during drought (Adams et al., ²⁰¹⁷; Brodribb et al., ²⁰²⁰). This insight has sparked a focus on plant hydraulics as the key to understanding and predicting spatial patterns of forest mortality (Tai et al., ²⁰¹⁶). However, our ability to hindcast which trees will die (in space and time) remains extremely limited (Anderegg et al., ²⁰¹⁶; Benito Garzón et al., ²⁰¹⁸; Venturas et al., ²⁰²⁰; Trugman et al., ²⁰²¹).

Our inability to predict mortality may be driven by: a poor understanding of water supply (i.e. how meteorological conditions translate into soil moisture limitation); by unappreciated variation in the biological factors mediating how supply limitation and atmospheric demand result in physiological stress; and/or an incomplete understanding of how hydraulic risk during periods of maximum water stress relates to performance at other timescales (e.g. growth, which can be offset by months or years from water stress in arid environments). The first issue is the persistent challenge of translating meteorological inputs into soil moisture supply over a tree's functional rooting depth (Anderegg et al., ²⁰¹³). Even as our gridded climate products improve and multiply, our understanding of tree‐level critical zone hydrology (from the top of unweathered bedrock to the top of the vegetation) remains quite limited (Rempe & Dietrich, 2018; Fan et al., ²⁰¹⁹; Dawson et al., ²⁰²⁰; Callahan et al., ²⁰²²; Hahm et al., ²⁰²²). As a consequence, translating meteorological drought into plant available water (PAW) remains challenging, particularly in geologically complex landscapes. If this edaphic uncertainty dwarfs climate uncertainty, improving the accuracy and resolution of climate products will not yield improved predictions of plant stress or mortality.

Climatic or meteorological predictors of drought stress often involve water balance metrics such as ‘climatic water deficit’ (CWD, seasonally integrated potential evapotranspiration minus modeled actual evapotranspiration), which estimate the mismatch between water supply and demand (Stephenson, 1990, 1998; Mitchell et al., ²⁰¹⁶; Berner et al., ²⁰¹⁷; Ledo et al., ²⁰¹⁸). These metrics in theory capture the critical interactions between precipitation, temperature, and humidity (Stephenson, 1998). However, water balance approaches require substantial assumptions about soil water holding capacity that drastically simplify complex critical zone hydrology. In the simplest water balance metrics such as Aridity Index (ratio of precipitation to potential evapotranspiration), all precipitation is considered to be plant available. More sophisticated metrics such as CWD involve a simplified soil bucket model based on large‐scale gridded soils data of 1–3 m depth, and a seasonal model of actual evapotranspiration constrained by this soil bucket (Flint et al., ²⁰¹³; Abatzoglou et al., ²⁰¹⁸). These soils data are often quite coarse and inaccurate outside of agricultural areas (Terribile et al., ²⁰¹¹) and typically do not address deeper soil, regolith, or fractured bedrock that can supply substantial water for many forests globally (McCormick et al., ²⁰²¹). It thus remains an open question whether water balance metrics capture supply limitations in deep‐rooted species growing in geomorphically complex systems.

Testing this question requires on‐the‐ground observations, such as measurements of plant water potentials and plant stable water isotopes across large spatial extents. Predawn water potential (Ψ_PD), when a tree has theoretically equilibrated with the soil (assuming no nighttime transpiration and that the plant is still hydraulically coupled with the soil), provides an empirical measurement of PAW and is the gold standard for measurement of plant water supply limitation. Meanwhile, xylem stable water isotopes can indicate the hydrological source of transpired water, with deuterium (D or ²H) and ¹⁸O being more depleted in precipitation falling in cold months and large storms (Ehleringer et al., ¹⁹⁹³). In addition, the relative proportion of DHO to H₂ ¹⁸O can reveal functional rooting depth (if soil water isotopic composition is known), seasonal water sourcing (isotopically depleted winter or enriched summer precipitation) or the extent to which plants use shallow soil water sources that are influenced by evaporation. Evaporation at less than 100% relative humidity enriches D faster than ¹⁸O, causing evaporatively enriched soil water to have a D ‘excess’ relative to the ‘Meteoric Water Line’ of D and ¹⁸O in precipitation. If site‐specific or even individual‐specific isotopic data can explain spatial variation in Ψ_PD but climatic water balance metrics cannot, that suggests that our climate or (more likely) soils data are insufficient representations of the critical zone to predict water supply limitation, much less mortality.

The second major issue for predicting drought mortality is how spatial variation in plant physiology complicates the translation of water limitation into maximum plant water stress. A tree's exposure to lethal hydraulic damage is determined by its xylem vulnerability to embolism and its minimum midday plant water potential (Ψ_MD). Within species (including our study species), xylem vulnerability has often proven fairly fixed (e.g. Pinus sylvestris (Lamy et al., ²⁰¹³) or Quercus douglasii (Skelton et al., ²⁰¹⁹)). However, plant hydraulic efficiency determines how a given level of soil water supply (Ψ_PD) and evaporative demand (vapor pressure deficit or VPD) translate into Ψ_MD, and hydraulic efficiency can vary substantially within species (Martinez‐Vilalta et al., ²⁰⁰⁹; Anderegg & HilleRisLambers, ²⁰¹⁶).

In particular, decreasing allocation to evaporative leaf area relative to conductive stem area (decreasing the A _l : A _s ratio, sometimes referred to by its inverse the ‘Huber value’) helps plants avoid extremely negative Ψ_MD by increasing leaf area‐specific hydraulic efficiency (Mencuccini & Grace, 1995; Martinez‐Vilalta et al., ²⁰⁰⁹; Trugman et al., ²⁰¹⁹; Mencuccini et al., ^2019b; Sanchez Martinez et al., ²⁰²⁰). While many within‐species trait–climate relationships are idiosyncratic (Anderegg, 2023), aridity‐driven intraspecific variation in A _l : A _s has emerged as a consistent pattern across studies and systems (Mencuccini & Bonosi, 2001; Martinez‐Vilalta et al., ²⁰⁰⁹; Anderegg & HilleRisLambers, ²⁰¹⁶; Rosas et al., ²⁰¹⁹; Anderegg et al., ²⁰²¹). Thus, variable allocation to leaf area may be a key hydraulic acclimatory mechanism driving within‐species variation in mortality risk (Trugman et al., ²⁰¹⁹; Mencuccini et al., ^2019a).

From a carbon economy perspective, biomass‐based traits such as the mass of leaves relative to the mass of stems (M _l : M _s) are an alternative basis for quantifying leaf : stem allocation, and also show relationships with water availability (Poorter et al., ²⁰¹²; Ledo et al., ²⁰¹⁸), though they have been less studied from a water stress perspective than A _l : A _s. Variation in allocation to leaf area should both moderate plant exposure to drought and strongly impact plant growth (Lambers & Poorter, 1992).

Finally, there is a critical question of timescale. Damaging Ψ_MD and even extreme soil moisture limitation (Ψ_PD) are short‐term states, lasting hours (for extreme, VPD‐driven Ψ_MD) to months. Meanwhile, particularly in Mediterranean‐type climates, the majority of carbon assimilation, growth, and trait adjustments occur months to years before the conditions that drive mortality risk. This may fundamentally decouple the things we can more easily survey across space (growth, allocation to leaf area) from the acute risk of mortality during drought. Within a species, mean growth (Klesse et al., ²⁰²⁰), growth sensitivity to drought (Canham et al., ²⁰¹⁸; Camarero et al., ²⁰²¹), drought‐related traits (Martinez‐Vilalta et al., ²⁰⁰⁹; Anderegg et al., ²⁰²¹), and drought‐induced mortality (Anderegg et al., ²⁰¹⁹) often correlate with site aridity or proximity to the dry range edge. Thus, we often assume that these processes are correlated across space despite the very different timescales they represent (e.g. short‐term Ψ_MD vs traits such as A _l : A _s that are developed every spring vs multi‐year growth). In studies of how traits influence growth and mortality risk, and particularly in biogeographic studies of how traits vary across a species' range, the implicit assumption is that a tree or site's relative condition (e.g. how its Ψ_PD or Ψ_MD at the time of sampling compares to other trees/sites) is highly correlated across seasons and years. Dry sites are usually dry compared to wet sites, within and across years. Yet this assumption is rarely tested. If, due to mismatches in timescale, maximum water stress (and risk of mortality) does not correlate spatially with drought‐related traits or the water stress constraint on growth, the task of predicting spatial patterns of drought mortality becomes substantially more complicated.

We explored these challenges in blue oak (Quercus douglasii Hook & Arn.), an ecologically and culturally important species that dominates low elevation woodlands and savannas across California, US. This deciduous, deeply rooted species is valued for dendroclimate reconstructions because of its high growth sensitivity to precipitation (Stahle et al., ²⁰¹³). Blue oak also experienced substantial mortality during the California 2012–2016 drought (Das et al., ²⁰¹⁹), though only c. 30% of the spatial variation in canopy dieback could be explained by climate, drought severity, or groundwater depletion (Brown et al., ²⁰¹⁸). Variation among populations in hydraulic damage thresholds (leaf and stem P50 or the xylem water potential that causes 50% embolism) appears limited in blue oaks (Skelton et al., ²⁰¹⁹), which suggests that spatial variation in hydrology or plant hydraulics (e.g. variation in trait such as A _l : A _s) are likely key to predicting blue oak mortality (McLaughlin et al., ²⁰²⁰).

We tested how predictable soil moisture limitation is across a c. 1000 mm precipitation gradient, how trait variation moderates drought exposure, and how linked peak drought stress, drought‐related traits, and growth are. To do this, we measured end of dry season soil moisture availability (Ψ_PD) and maximum midday water stress (Ψ_MD) during a drought, stable water isotope tracers of functional rooting depth, leaf and allocation traits, and growth rates across the geographic range of blue oak. We asked: (1) can we explain the spatial variation in water availability and maximum drought stress across the landscape using climate, soil, or stable water isotopes? (2) How do allocation traits, particularly A _l : A _s, vary as a function of climate or soil moisture, and do they feed back to mediate maximum water stress? (3) Do 1 or 5‐yr individual tree growth rates correspond to observed end‐of‐season soil moisture limitation or maximum water stress, site climate/meteorology, or variation in allocation traits? And (4) does mechanistically simulated end‐of‐season hydraulic risk or carbon gain correlate with season‐integrated or multi‐year tree growth?

Materials and Methods

Study System

We sampled 15 sites across the geographic and climatic distribution of blue oak (Quercus douglasii Hook & Arn., Fig. 1a), selecting a subset of the most climatically diverse yet logistically accessible/feasible from the 26 sites of McBride et al. (1997), Papper & Ackerly (2021). Blue oak inhabits a wide range of precipitation and potential evapotranspiration (Fig. 1b), typified by Mediterranean‐type hot, dry growing seasons (April to September) with most precipitation falling December to March.

Fig. 1.

Water stress, traits and growth were sampled at 15 site across the range of blue oak (Quercus douglasii) during a drought year. (a) The geographic distribution of blue oak in California, USA spans a large range of mean annual precipitation (MAP in mm, from TerraClimate (Abatzoglou et al., ²⁰¹⁸)). Black points show locations of herbarium specimens (Baldwin et al., ²⁰¹⁷), sampled locations are shown as orange points. (b) Sampled locations span the vast majority of the MAP and potential evapotranspiration (PET) space inhabited by blue oak. The 2018 water year (wy, crosses) when samples were collected was substantially hotter and drier than the 1981–2010 historical period (closed circles).

At the end of the 2018 growing season from 17 September to 3 October, when trees were experiencing the annual minimum soil moisture before leaf senescence, we sampled predawn and midday leaf water potentials, allocation traits, leaf gas exchange, and tree cores for water stable isotope and growth analysis. We prioritized visiting all sites as quickly as possible for comparability of water potentials, at the cost of ancillary measurements at some sites. We measured water potentials at all 15 sites, but allocation traits at only 12 sites, growth and isotopes at 11 sites, and leaf gas exchange on a subset of trees at 7 sites. Very little precipitation fell (< 15 mm or < 2% of total precipitation) at any site during the 4 months preceding our sampling. Thus, we achieved consistent antecedent drought conditions for all study sites. Finally, we used a mechanistic tree hydraulics model to synthesize multiple interacting factors that cannot be parsed statistically, namely interactions between meteorology and plant water use (e.g. VPD and the effects on stomatal closure), our observations of supply limitation and the effects of leaf area on the plant hydraulic balance.

Leaf water potential

We quantified moisture availability with predawn leaf water potential (Ψ_PD) sampled between 3 am and first light (local time). We measured 3–8 leaves from each of 5–7 trees per site (mean 5.5), severing the petiole with a sharp razor blade (for short trees) or collecting a branch > 50 cm long using pole pruners and immediately cutting 2–3 leaves using a razor blade. Leaves were immediately wrapped in tinfoil and a lightly damp paper towel to limit evaporation without rehydrating the leaf and measured in a Scholander‐type pressure chamber (PMS Instruments, Corvallis, OR, USA) within 1 min of cutting. A ‘modified petiole’ (Rodriguez‐Dominguez et al., ²⁰²²) was cut using a razor blade for leaves with short petioles.

We then measured leaf water potential at midday (Ψ_MD) between 12:30 h and 14:00 h to estimate maximum water stress. The sampling year (2018) was drier and hotter than average (Fig. 1b), so end‐of‐season Ψ_MD was a reasonable estimate of long‐term Ψ_min values. We calculated ΔΨ (the water potential gradient driven by transpiration from soil to leaf) as Ψ_PD–Ψ_MD.

For a subset of 2–4 trees at 6 sites, we also measured midday stem xylem potential of terminal stems (Ψ_{MD_xy}, for woody stems immediately adjacent to leaves, typically < 5 cm from branch meristem) by wrapping 4–6 leaves on a terminal stem in tinfoil, damp paper towels, and plastic bags to stop transpiration > 20 min before leaf collection (e.g. Naor, ¹⁹⁹⁸; Ortuño et al., ²⁰⁰⁶). Ψ_{MD_xy} was on average 7% (SE ±1.5%) less negative than Ψ_MD. We calculated Ψ_{MD_xy} for all trees assuming an 7% offset from Ψ_MD (i.e. that 7% of total resistance was between the branch xylem and leaf lamina). We used the species average P50 values for leaves (−3.88 SE ±0.09 MPa) and stems (−4.47 ± 0.1 MPa) from Skelton et al. (2019), who found < 3.5% CV in P50 among seven populations, to calculate Hydraulic Safety Margins (HSM_leaf = Ψ_MD–P50_leaf, HSM_stem = Ψ_{MD_xy}–P50_stem). Because HSM_leaf and HSM_stem are linear transforms of Ψ_MD we report statistical relationships only for Ψ_MD.

Trait measurements

At 12 of the 15 sample sites, we collected three terminal twigs from the sun exposed mid‐ to upper‐canopy of five trees using pole pruners. Terminal twigs, cut at the previous year bud scar, included current year (2018) stem growth and all subtending leaves, thus represent 1 yr's investment in branch and leaf tissue. Branches were rehydrated using the ‘partial rehydration’ method (Pérez‐Harguindeguy et al., ²⁰¹³) and stored in a cooler or refrigerator for at least 12 h before processing. In the lab, leaves were removed (with petiole), weighed for wet mass (M_{l_wet}), and scanned on a flatbed scanner. Images were analyzed with ImageJ (Schneider et al., ²⁰¹²) to calculate total twig leaf area (A _l), and total leaf area was divided by leaf number for average leaf size. Stem basal diameter underneath the bark was measured at 2–5 orientations using digital calipers and used to calculate stem cross‐sectional area (A _s). Total twig A _l was divided by stem A _s to calculate branch A _l : A _s ratio. Stem length was measured with calipers or a ruler. Leaves and stems were oven dried at 60°C for 72 h, and their dry mass measured (M _l, M _s). Leaf mass per area (LMA) was calculated as M _l/A _l, leaf mass to stem mass ratio as M _l/M _s, and leaf dry matter content (LDMC) as leaf dry mass by leaf wet mass (M _l/M _{l_wet}). Because samples were not processed until after the water potential campaign, we excluded LDMC values from samples that were stored longer than 1.5 wk from analysis, resulting in exclusion of LDMC values from four sites (n = 8 sites rather than n = 12 for LDMC).

Stable isotope analyses of water

Xylem water oxygen (δ¹⁸O) and hydrogen (δ²H or δD) stable isotope composition was measured at 11 of the study sites on 5 trees (7 trees at two sites, 3 trees at one site). Tree cores c. 4 cm long were extracted at breast height, rapidly placed in pop top scintillation vials (after removing bark and cambium), wrapped in parafilm, and stored in a cooler until frozen in the lab. Water was cryogenically extracted (Ehleringer et al., ²⁰⁰⁰). The oxygen isotope composition of extracted water was determined through mass spectrometry by CO₂ headspace equilibration using a Gas Bench II (GB; ThermoFinnigan, Bremen, Germany) connected to a Delta Plus XL mass spectrometer (ThermoFinnigan) at the Center for Stable Isotope Biogeochemistry (CSIB), University of California, Berkeley, CA. The hydrogen isotope composition of samples was determined via injection into an H/Device (HDEV; ThermoFinnigan) coupled to a Delta Plus mass spectrometer (ThermoFinnigan), also at CSIB. All isotopic compositions are reported in delta (δ) notation in parts per thousand (‰) relative to the V‐SMOW standard.

$${\mathrm{\delta }}^{18}\mathrm{O}\text{or}\mathrm{\delta D}=\left(\frac{R_{\text{sample}}}{R_{\text{standard}}}-1\right)\times 1000$$ (Eqn 1)

where R = ¹⁸O/¹⁶O or D/H.

Monthly meteoric water isotope values were estimated for each site using the Online Isotopes in Precipitation Calculator, version OIPC3.2 (Bowen et al., ²⁰⁰⁵; Bowen, ²⁰²⁴). The evaporative deuterium enrichment of each sample relative to each site's local meteoric water line was used to calculate ‘line conditioned’ excess (lc excess, Landwher & Coplen, ²⁰⁰⁴). Average monthly precipitation‐weighted meteoric water isotopic values for the sampling year (δD_precip, δ¹⁸O_precip) were calculated for each site. Raw xylem δD/δ¹⁸O were interpreted as reflecting the hydrological source of water (more depleted indicating water deriving from colder and larger storms) and lc excess was interpreted as indicative of deep vs shallow soil moisture (higher lc excess indicating shallow, evaporatively enriched water sources) (Ehleringer & Dawson, 1992; Landwher & Coplen, ²⁰⁰⁴). Because xylem δD shows a bias with tissue gravimetric water content (Sobota et al., ²⁰²⁴), we primarily present results using δ¹⁸O in the main text.

Growth

We measured 1 yr branch growth as the branch length of terminal branches collected for trait measurements (12 sites). We also measured 5 yr radial growth on the tree cores collected for water isotopes after cryogenic extraction by sanding them and measuring the most recent 5 yr of ring widths (2014–2018) using calipers and a dissecting scope. From radial growth increment, we calculated basal area increment (BAI) from the outside of the tree inward. Because both mean and variance of BAI were strong functions of tree diameter at breast height (DBH), we corrected for size‐related BAI trends by calculating each tree's growth relative to the maximum observed growth at that DBH across all sites (Supporting Information Fig. S1). We fitted a quantile regression to the 90^th percentile BAI–DBH relationship, and then calculated each tree's size‐standardized growth as the ‘percent of maximum BAI’ (% max BAI) or the observed BAI divided by the 90th quantile BAI for the tree's DBH.

Climate and meteorological data

A priori, we expected the water balance of the sampling year to be the most important predictor of water stress, traits and growth, however, we included a range of meteorological (2018 water year or ‘2018 wy’) and climatological (30 yr climate norms, 1981–2010) as potential predictors. Climate normals and 2018 wy (1 October 2017 to 30 September 2018) data were extracted for each site from TerraClimate (Abatzoglou et al., ²⁰¹⁸). TerraClimate has c. 4 km resolution monthly climate and water balance values for the global terrestrial surface from 1958 to 2020 (updated periodically). We included precipitation (PPT), evaporative demand (potential evapotranspiration, PET), and water balance metrics (actual evapotranspiration, AET; climate water deficit, CWD = PET–AET), plus maximum and minimum temperature for the historical period (T _max, T _min), plus 2018 wy annual and seasonal variables. See Methods S1 for full description of variables and selection philosophy. To avoid model overfitting, we only fit one climate/meteorological predictor in any model. When no climate variable was found to predict a quantity, we plotted it against 2018 wy CWD.

Soils data

To test whether site edaphic characteristics can predict water availability, traits, or growth, we extracted PAW, available water holding capacity (AWC), and minimum bedrock depth for each site from the Soil Survey Geographic Database (SSURGO). The data were collected at scales ranging from 1 : 20 000 to 1 : 24 000. Both AWC (0–100 cm) and minimum bedrock depth were taken from the soil map unit composition. In two sites, where minimum bedrock depth was not available, we used average data from the dominant soil series. PAW (cm) was pulled from the dominant soil series in the soil map unit.

Leaf gas exchange

For 1–4 trees at 7 sites (when time allowed while collecting water potentials), 16 total trees, we measured stomatal conductance (g _s) and photosynthesis (A _s) as a function of leaf water potential at midday. Using a LI‐6800 gas exchange system (LI‐COR Biosciences, Lincoln, NE, USA), we collected spot measurements of g _s and A _s, setting the chamber to ambient temperature, CO₂ concentration and relative humidity, and light to 1500 μmol m⁻² s⁻¹. Measurements were made after the conditions in the chamber stabilized (c. 1 min) but before stomata began to react to the chamber environment (typically 3–5 min) and corrected for the leaf area in the cuvette. Water potential was then measured on a leaf immediately adjacent to the measured leaf. At one site, repeated gas exchange and water potential measurements were conducted from 07:00 h to 12:00 h on one tree to determine how the temporal relationship between stomatal conductance and leaf water potential compared to the spatial relationship across sites.

Mechanistic model simulations

We used the HOTTER model (the Hydraulic Optimization Theory for Tree and Ecosystem Resilience model) (Trugman et al., ²⁰¹⁸; Quetin et al., ²⁰²³), a physiologically based single tree model run at a daily timestep, with a realistic representation of gas exchange (Eller et al., ²⁰¹⁸) and a detailed representation of plant hydraulics (Trugman et al., ²⁰¹⁸) to capture both the effects of supply limitation and VPD on Ψ_leaf, carbon balance, and hydraulic function (percent loss of conductivity, PLC). The goal was to integrate the nonlinear interactions between allocation, water availability and VPD to test whether daily tree performance at the time of sampling related to tree growth at longer timescales. We simulated each tree with tree‐specific allocation (A _l : A _s scaled to the whole tree) and water availability (Ψ_PD) and site‐specific meteorology for the sampling period (temperature, VPD from 17 to 27 September 2018) and assessed simulated Ψ_MD against observed Ψ_MD to determine model skill. See Methods S2 for specifics of model forcing and parameterization. We then compared simulated end‐of‐season hydraulic risk (PLC) and carbon gain (gross primary productivity, GPP, and net primary productivity, NPP = GPP – respiration) to observed growth to see whether performance under extreme stress (end of a very dry growing season) was correlated with performance at longer timescales.

Analysis

We averaged all measurements to individual for analysis (87 trees for water Ψ, 60 traits, 40 for LDMC, 57 for water isotopes, and 59 for growth). We used ANOVAs to calculate the among‐population effect size (η ²) and unbiased among‐population effect size (ω ²) for each trait. We then sought to explain spatial variation in plant water potentials (Ψ_PD, Ψ_MD, ΔΨ), plant traits (A _l : A _s, M _l : M _s, leaf size, LMA, LDMC), xylem stable water isotope composition (δ¹⁸O, δD, and lc excess), and growth (branch length, tree height, % max BAI) with climate or soils using Akaike's Information Criterion corrected (AICc for small sample sizes)‐based model selection in the R statistical environment (v.4.3.1, Team, ²⁰¹⁶). We fit linear mixed models relating each response variable to each of the 24 climate, meteorological or soil predictors individually (described above) with a random intercept for site, using the lmer() function in the lme4 package (Bates et al., ²⁰¹⁵). We selected the most parsimonious model of the 25 possible models (24 models with individual predictors plus a null model, Table S1) with AICc, ranking among predictors based on AICc score of their univariate models and considering predictors in single predictor models that reduced AICc by 2 or more units over the null to be significant (Zuur et al., ²⁰⁰⁹). We used t‐tests based on Satterthwaite's approximate degrees of freedom to determine the P‐value for the AICc‐selected predictors, using the R package (Kuznetsova et al., ²⁰¹⁶). We report marginal R ² (R ² _marg) values for attribute–climate relationships from these linear mixed models, which reflect the proportion of site‐to‐site variation explained. We then tested whether including A _l : A _s and an interaction between A _l : A _s and climate improved linear models for growth and Ψ_MD (see Methods S3).

We also used linear mixed effect models with a site random intercept to test whether: (1) xylem water stable isotopes explained tree‐to‐tree variation in plant water potentials; (2) water potential (either Ψ_PD, Ψ_MD, or ΔΨ) explained growth (either 5 yr BAI or 1 yr stem length); (3) traits predicted growth; (4) HOTTER output (PLC, GPP, or NPP) predicted growth. For these models, we report R ² _marg values as the proportion of tree‐to‐tree variation within site explained.

Functionally, mixed models with tree‐level predictors (1–4 above) tested whether, on average, there was a significant relationship among trees within a site. We also tested for significant relationships among sites rather than among trees using linear models of site‐level means, and report R ² values from these linear models as the proportion of among‐site variation explained.

Finally, we used linear mixed effect models (site random intercept) and AICc to test whether adding A _l : A _s as an interaction with climate improved the best climate models of Ψ_MD or growth.

Data are available in the Zotera repository DOI: 10.5281/zenodo.19365699. Analysis code is available on Github at https://github.com/leanderegg/BlueOakLandscape.

Results

Predicting drought stress

Despite sampling across large geographic differences in precipitation and potential evapotranspiration (Fig. 1), and despite massive differences between sites in leaf water potential (−0.82 to −3.78 MPa for Ψ_PD, −2.02 to −4.45 MPa for Ψ_MD), neither Ψ_PD nor Ψ_MD could be predicted by long‐term site climate, 2018 water year meteorology, or soil characteristics (Fig. 2). There was also no evidence that leaf area allocation (A _l : A _s) mediated the relationship between Ψ_MD and climate (null linear mixed model had lower AICc than all nested models of climate, A _l : A _s and their interaction, Fig. S2). The tree‐to‐tree variation within site was substantial (mean within site range of −1.3 MPa for both Ψ_PD and Ψ_MD) and was partially explained by tree size, with larger DBH trees having less negative Ψ_PD (Fig. S3). However, the between‐site variation still constituted 71% (Ψ_PD) and 51% (Ψ_MD) of the total water potential variation (Table S2). The ΔΨ (Ψ_PD–Ψ_MD, potential gradient caused by transpiration) increased with growing season VPD (linear mixed model P = 0.011). The average leaf hydraulic safety margin (HSM, Ψ_MD–P50_leaf) was 0 or even negative for a number of both climatically wet and dry sites, and some individual trees from both wet and dry sites had negative stem hydraulic safety margins (Ψ_{MD_xy}–P50_stem) (Fig. 2d). Despite these negative safety margins, no trees in four sites with smallest safety margins (HSM < 0.1 MPa) died following this study, based on high resolution satellite imagery from 2021 to 2025 (Google Earth Pro, v.7.3.6.10441).

Fig. 2.

At the end of a dry growing season (2018), neither predawn leaf water potential (a, Ψ_PD, root zone‐integrated soil moisture availability) nor midday leaf water potential (b, Ψ_MD, maximum water stress) of Quercus douglasii were predicted by site climate (1981–2010) or sampling year meteorology (2018 water year). (c) Transpirative water potential drop (ΔΨ) increased with site 2018 growing season vapor pressure deficit (VPD). (d) hydraulic safety margin (Ψ_MD–P50) for leaves (black) and stems (orange). Gray points indicate individual tree average leaf water potentials; black points indicate site average leaf water potentials. Response variables with no significant climate predictor (predawn and midday leaf water potentials and HSM) are plotted against the climatic water deficit (CWD) of sampling year (2017–2018 water year or ‘2018 wy’). AET, actual evapotranspiration; PPT, precipitation. P‐values from linear mixed effect models with site as random intercept. Gray and black dashed lines indicate leaf and stem P50, respectively, from Skelton et al. (2019). n = 15 sites for all variables.

Meanwhile, xylem stable isotope composition, but not evaporative enrichment, predicted leaf water potentials, both among sites and among trees within a site. Xylem water more enriched in δD and δ¹⁸O was marginally or significantly associated with more negative Ψ_PD, more negative Ψ_MD, and a smaller ΔΨ both among sites and among trees (Figs 3, S4; Table S3). Excluding one site (Pepperwood Preserve, ‘PWD’, gray in Fig. 3), which showed anomalously hydrated water potentials given its water isotopic signature, possibly due to a moderate severity wildfire 1 yr before sample collection, these relationships were even stronger. Excluding PWD, the among tree R ² _marg of δ¹⁸O was 0.2 (Ψ_PD) and 0.27 (Ψ_MD) and among site R ² was 0.57 for Ψ_PD, 0.35 for Ψ_MD (Table S3). Precipitation‐weighted average meteoric δ¹⁸O_precip/δD_precip of the site also predicted Ψ_PD (Fig. S5). However, xylem δD (though not δ¹⁸O) was still a significant predictor of Ψ_PD when meteoric composition was included as a covariate in the linear mixed model (P < 0.001 excluding PWD), and tree‐level residuals from the δD_xylem~δD_precip relationship were also significantly related to Ψ_PD within site (Fig. S5). The lc excess, which is designed to isolate the effect of evaporative enrichment on ecosystem water, was unrelated to plant water potentials, at either the site or individual level.

Fig. 3.

Leaf water potentials of Quercus douglasii at the end of 2018 were strongly related to 2018 xylem water isotopes within and among sites. Predawn leaf water potential (Ψ_PD, a), midday leaf water potential (Ψ_MD, b), and the difference between them (ΔΨ, c) decrease with increasing xylem water enrichment of δ¹⁸O. Points show tree average values, triangles show site average values, colors indicate different sites. Gray trend lines show statistically significant (P < 0.05) within‐site relationships from a mixed model with site random intercepts, with the outlier burned site (gray points, triangle) excluded, and black trend line and colored triangles indicate that the relationship is also significant (linear model P < 0.05) or marginally significant (dashed line, P < 0.1). n = 11 sites (10 excluding gray burned site).

Plant water potential also had profound effects on end of season leaf gas exchange. Across sites and trees within sites, midday stomatal conductance and assimilation steadily declined as Ψ_MD decreased toward −4 MPa (Fig. 4). The point of stomatal closure coincided quite closely with the previously reported P50 of leaves of −3.88 MPa, a threshold that was consistent across populations of blue oak (Skelton et al., ²⁰¹⁹). This point of stomatal closure was also apparent in the morning time course of a single tree as its leaf water potential approached −3.88 near midday (Fig. 4). Thus, the consequences of observed water potential variation across the landscape for late season carbon gain and water loss were substantial. At two of the sampled sites, mean Ψ_PD was similar to or more negative than leaf P50 (Fig. 2), implying sustained stomatal closure and limited carbon gain at these sites for the entirety of the late growing season.

Fig. 4.

Leaf gas exchange was strongly regulated by stomata as leaves neared Quercus douglasii's P50 (leaf water potential causing 50% embolism of leaf xylem). Stomatal conductance (a) and photosynthesis (b) of Q. douglasii as a function of leaf water potential in late 2018. Closed circles show midday gas exchange and Ψ_MD values (color indicates different sites). Gray x symbols indicate the morning time‐course of a single tree as it approached midday. The gray and black vertical lines indicate the leaf and stem P50 (respectively) of blue oak from Skelton et al. (2019). n = 7 sites.

Predicting traits

All leaf and stem traits demonstrated significant variation among sites, ranging from 17% (M _L : M _S) to 58% (LDMC) of total trait variation (Table S2). Site climate or meteorology was a significant predictor of only two traits (M _L : M _S and leaf size), while soil water holding capacity predicted A _L : A _S and LMA (Fig. 5). M _L : M _S decreased with increasing minimum temperature of the sampling year (−0.48 ± 0.16 standardized effect size), and leaf size increased with increasing historical precipitation (0.43 ± 0.17 standardized effect size). Meanwhile, A _L : A _S increased and LMA decreased with soil water holding capacity (0.48 ± 0.17 and −0.46 ± 0.16 standardized effect sizes, respectively) and were not predicted by any climate variables when excluding soil predictors. LDMC was not predicted by any environmental variable, but had a smaller sample size.

Fig. 5.

Observed traits of blue oak (Quercus douglasii) vary across the landscape as a function of site climate and soils. (a) Tree‐average terminal branch leaf area to stem area ratio (A _l : A _s), (b) leaf mass to stem mass (M _l : Ms), (c) mean leaf size, (d) leaf mass per area (LMA), and (e) leaf dry matter content (LDMC) measured in September–October 2018 as a function of the best site climate (30 yr climatology from 1981 to 2010, ‘30 yr’), sample year meteorology (2018 water year ‘2018 wy’), or soil predictor based on model selection. Black circles show site means, gray circles show tree means; solid trend lines show significant relationships (linear mixed effect model with random site intercept, P < 0.05). n = 12 sites (8 sites for LDMC). ‘1 m soil storage’ is soil water storage capacity in the top meter; ‘Plant Avail Water’ is the water available between field capacity and permanent wilting point (−1.5 MPa) in the top meter of soil. CWD, climatic water deficit.

End‐of‐season soil moisture and maximum water stress were almost completely unrelated to leaf and allocation traits. Ψ_PD was not correlated with any trait, and Ψ_MD was only correlated with LDMC (Fig. S6), with less negative Ψ_MD/larger HSMs being associated with higher LDMC (linear mixed effect models, P = 0.004). Moreover, traits were largely uncorrelated with each other. A _L : A _S was positively correlated with M _L : M _S, as well as leaf size (Fig. S6), but no other traits were significantly correlated.

Drought and trait relationships with prior growth

One year branch elongation and 5 yr basal area growth were significantly positively correlated, but the relationship was relatively weak (marginal R ² = 0.12, P = 0.01). Both basal area growth from the prior 5 yr and (log₁₀‐transformed) branch length of the current year's growth were almost completely unrelated to end of growing season soil moisture (Ψ_PD), maximum water stress (Ψ_MD) or ΔΨ, with the exception of current year branch length marginally significantly increasing with increasing ΔΨ (Fig. 6a,c). Moreover, no site water‐related climate variables outperformed the null model to explain either growth metric. Indeed, the only climate variables that had lower AICc than the null model were winter/spring temperature‐related, and the best climate variable for both 5 yr radial and 1 yr stem growth was 30 yr T _min (Fig. 6b,d), with sites with warmer winter temperatures growing faster.

Fig. 6.

Observed end of season water status of blue oak (Quercus douglasii) does not predict growth at either of two timescales, within or among sites, but site minimum temperature does predict growth among sites. (a) Predawn leaf water potential (Ψ_PD) does not predict 5 yr radial growth, calculated as the realized % of maximum basal area increment (max BAI) given a tree's diameter at breast height (DBH). (b) Radial growth increases strongly with increasing 30 yr average minimum temperatures of the coldest month. (c) Transpirative water potential drop (ΔΨ) is weakly related to average 1 yr stem length (but neither Ψ_PD nor Ψ_MD themselves are, data not shown). (d) Stem length is also positively related to site minimum temperature. n = 11 sites for % of max BAI and n = 12 sites for stem length. For (a, c) significant within‐site relationships were assessed with linear mixed effect models with a site random intercept and among site relationships with linear models and site means. For (b, d) significance of among‐site relationships was assessed via linear mixed effect models with a site random intercept.

Growth was also not related to leaf and stem traits in any expected ways, assuming that prior year traits moderated growth and the sampled values were good proxies for prior year trait values. Most traits were unrelated to either 1 yr stem or 5 yr basal area growth, but both within and among sites A _l : A _s and M _l : M _s were negatively related to percent of maximum BAI, and M _l : M _s was also extremely strongly negatively related to branch length (Fig. 7). Both A _l : A _s and M _l : M _s decreased strongly with branch length, suggesting a structural, rather than hydraulic or carbon economic driver of both branch allocation. If branch length was included as a covariate in models predicting % max BAI, M _l : M _s was no longer significantly related (P = 0.21) but A _l : A _s remained significantly negatively related to % max BAI (P = 0.01). Across sites, leaf size was positively related to branch length, but not among trees within a site and not to % max BAI. LDMC was marginally significantly related to branch length among trees within sites but not across sites. Including traits as interacting predictors with climate did not alter the growth–climate relationships, and winter variables remained the best climate predictors (see Methods S3).

Fig. 7.

Leaf and stem traits are poor predictors of tree growth. Relationships between leaf traits of Quercus douglasii, including leaf area : stem area ratio (A _l : A _s, a, f), leaf mass : stem mass ratio (M _l : M _s, b, g), average leaf size (c, h), leaf mass per area (LMA, d, i), and leaf dry matter content (LDMC, e, j), and two measures of growth: 5‐yr radial growth calculated as the percent of realized basal area increment growth (% of max BAI) relative to the 90th percentile growth for that tree size (a–e), and average branch length of the current year terminal branches (1 yr branch length, f–j). Points show tree average traits, triangles show site average traits. Gray lines show significant (solid, P < 0.05) and marginally significant (dashed, P < 0.1) trends among trees within sites from linear mixed effect models, with the associated P‐value and marginal R ² (R ² _marg) shown in gray. Solid/dashed black lines and colored triangles indicate significant/marginally significant trends among site means from linear models, with associated statistics shown in black. n = 11 sites for % of max BAI, n = 12 sites for Branch Length.

Mechanistic model synthesis

Daily HOTTER simulations of Ψ_MD, parameterized with observed, tree‐specific A _L : A _S and forced with tree‐specific end‐of‐season water availability (Ψ_PD) and site meteorology, were strongly correlated with observations (Fig. 8a). This was true both within‐site (R ² _marg = 0.52, RMSE = 0.33 MPa) and among sites (R ² = 0.73, RMSE = 0.3 MPa). Variation in VPD did not strongly influence simulated Ψ_MD for most trees, though it did strongly control carbon uptake (Fig. S7). Simulated 2018 end‐of‐season PLC was completely unrelated to either observed 5 yr basal area growth (Fig. 8b) or 1 yr branch growth (Fig. S8). Meanwhile, simulated GPP at the time of sampling was marginally negatively correlated with observed 5 yr growth and unrelated to 1 yr growth (Figs 8c, S8c). At the same time, daily NPP at mean VPD, or photosynthesis minus temperature‐ and allocation‐dependent respiration, was negative for about half of trees and unrelated to observed 5 yr or 1 yr growth (Figs 8d, S8d).

Fig. 8.

A tree hydraulic model captures observed water potentials but not growth. Comparison of variables simulated with the HOTTER mechanistic model based on trait observations from across the geographic range of Quercus douglasii and end‐of‐season meteorological forcings demonstrate reasonable model skill and highlight the disconnect between end‐of‐season hydraulic risk and observed growth. (a) Observed vs simulated midday leaf water potentials (Ψ_MD), with points showing simulations at the median vapor pressure deficit (VPD) during the sampling (error bars span the lowest VPD day to the highest VPD day during the sampling period). (b) Simulated percent loss of conductivity (PLC) was completely unrelated to observe growth rates (circles, trees; triangles, site means). (c) Simulated gross primary productivity (GPP) was negatively correlated with observed basal area growth within but not among sites. (d) While many trees were simulated to have a negative net primary productivity (NPP) (i.e. respiration higher than photosynthesis), simulated NPP was unrelated to observed growth (gray lines show significant trends among trees within sites from linear mixed effect models. n = 15 sites for (a)) and n = 11 sites for (b–d).

Discussion

Our range‐wide survey of water availability, water stress, traits and growth in blue oak revealed substantial site‐to‐site variation in all quantities but remarkable disconnects between them. While isotopic proxies for site hydrology were strongly related to end‐of‐season soil water availability (Ψ_PD), site water balance and soil characteristics rarely predicted plant‐available soil water, water stress, traits, or growth, suggesting that existing water balance products do not capture the hydrological reality of these trees. Moreover, end‐of‐season water stress, leaf and allocation traits, and growth were largely unrelated to each other, and statistically incorporating climate interactions with leaf allocation did not improve our ability to explain spatial variation in either Ψ_MD or growth. A mechanistic model that integrated soil moisture limitation, sampling period VPD and tree‐specific allocation did skillfully capture variation in Ψ_MD within and among sites; however, the resulting hydraulic risk and daily carbon gain during the sampled drought simulated by a plant hydraulics model was unrelated to observed prior growth (either 1 or 5 yr). These results raise questions for both how we predict supply limitation and its effects in deeply rooted tree species in geologically and geomorphically complex landscapes, and how we conceptually integrate hydraulic risk and carbon gain in a highly seasonal environment.

Deep water access decouples water availability and stress from aboveground climate, making drought stress hard to predict

We found a remarkable decoupling of plant water status from aboveground climate. No single metric of water supply, demand, or balance (nor their interaction with A _l : A _s) could explain spatial variation in root‐zone soil moisture (Ψ_PD) or maximum water stress (Ψ_MD). This could be partly driven by topographic variation not captured by 4 km climate data, though most sites were on shallow slopes and results were very similar using climate normals from higher resolution products (e.g. 290 m BCM, Flint et al., ²⁰¹³, data not shown). Moreover, a previous exhaustive study of blue oak Ψ_PD across a local elevation gradient found no relationship between topo‐edaphic factors (slope, aspect, elevation) (Knops & Koenig, 1994) even while year‐to‐year variation in precipitation drove interannual variation in Ψ_PD in the same trees (Knops & Koenig, 2000). Thus, we believe the issue is probably not climatic uncertainty, but rather the difficulty of capturing subsurface hydrology with water balance products.

The overriding importance of the subsurface hydrology, or the ‘weather underground’ (McLaughlin et al., ²⁰²⁰), is highlighted by the strong relationship between enriched xylem water stable isotope composition and negative plant water potential (both Ψ_PD and Ψ_MD). This impact of hydrology may be somewhat independent of rooting depth, as larger trees had less negative Ψ_PD presumably due to deeper roots yet did not have more depleted xylem isotopes indicative of stable water sources (Fig. S3). The relationship between Ψ_PD and xylem water isotopes partly reflects the influence of meteoric water (Fig. S5b), with precipitation isotopes proving a more successful predictor of Ψ than water balance metrics (Figs 2, 3, Table S2). In this system, spatial variation in precipitation‐weighted δD and δ¹⁸O is driven by the timing of precipitation and storm size, with more isotopically depleted sites receiving a larger fraction of rain as large, cold, winter storms. Coincidentally, these are the same types of storms likely to recharge shallow aquifers and fractured bedrock, providing deep, isotopically depleted water for blue oaks to use far into the growing season. Moisture balance metrics evidently did not capture the ecohydrological importance of recharge rate and timing in the same way.

Interestingly, meteoric water (δ¹⁸O_precip) was a better predictor of Ψ_PD than was the evaporative enrichment (lc excess) of xylem water. We could not characterize isotopic end members for each of 15 sites and the absolute functional rooting depth implied by the isotopic values we and others have obtained (see Weitz, ²⁰¹⁸) may vary based on site hydrology. However, water limitation was related to isotopic enrichment but not evaporative enrichment (e.g. lc excess), and all trees were similarly offset from the meteoric water line (Fig. S9). Most trees showed a similar amount of evaporative enrichment but the trees sourcing more depleted meteoric sources, presumably from large, winter storms held in deep, stable water reservoirs, remained the most hydrated by the end of the growing season. Collectively, these results highlight that improved representation of Critical Zone hydrology is key to improving spatial predictions of drought stress. Current soils data (either SSURGO or the coarser gridded soils used in water balance products) are insufficient to describe water supply limitation for deeply rooted species in geologically complex environments. However, the relationship between Ψ_PD and interpolated δ¹⁸O_precip, which are empirically derived from only latitude, altitude, and a spatial interpolation of observations, suggests that the oak‐relevant variation in Critical Zone hydrology is regionally coherent/predictable even without site‐specific edaphic information.

Growth and hydraulic risk are decoupled in a Mediterranean‐type climate

Water stress clearly limits assimilation in blue oak across much of its range by the end of the growing season as leaf water potentials approach the point of substantial leaf embolism (leaf P50, Skelton et al., ²⁰¹⁹). The decline in leaf gas exchange approaching leaf P50 was consistent in midday gas exchange rates across sites, within a single tree over the course of a morning (Fig. 4), and previously at a sight over the course of a growing season (see fig. 4 in Xu & Baldocchi, ²⁰⁰³). Such behavior is consistent with optimal stomatal behavior based on hydraulic cost (Anderegg et al., ²⁰¹⁹; Sperry et al., ²⁰¹⁶). Many sampled trees appeared quite close to the hydraulic edge by the end of 2018, with a number of trees surpassing leaf and stem P50 at midday and even predawn (Fig. 2), suggesting substantial stomatal limitation on carbon gain at the time of sampling.

If end‐of‐season water availability (and limitation to carbon gain) is correlated with conditions earlier in the growing season and growth is primarily water‐limited, 1 yr or even 5 yr tree growth should decline across sites with increasingly negative end‐of‐season Ψ_PD (and inferred carbon limitation). This was not the case (Fig. 6). Indeed, the simulated end‐of‐season hydraulic damage and gross photosynthesis from a plant hydraulics model, which captures our best understanding of the interacting effects of leaf allocation's impact on whole‐plant hydraulics, and stomatal balancing of carbon gain and hydraulic risk given VPD and supply limitation (Ψ_PD) at the time of sampling, had no or even a negative relationship with 5 yr observed radial growth and 1 yr branch growth (Figs 8, S8).

One explanation for this disconnect is that, across a c. 1000 mm gradient in annual precipitation, a site's relative end‐of‐season dryness is uncorrelated with its early season dryness. In Mediterranean‐type climates, this could be mediated by complex interactions between precipitation supply, transpiration rates, and critical zone water holding capacity, particularly in the poorly understood bedrock vadose zone (Dralle et al., ²⁰²⁰; Hahm et al., ²⁰²²). Related to this possibility, a phenological mismatch could decouple the times of year most critical for carbon gain/growth and those that determine hydraulic risk from drought stress. Due to California's Mediterranean‐type climate (cool wet winters and hot dry summers), photosynthesis in blue oak woodlands peaks in the spring and early summer, up to 50 d earlier than peak solar radiation and then declines precipitously (Ma et al., ²⁰¹¹; Xu & Baldocchi, ²⁰⁰³). Thus, the most productive growing season for blue oaks is a small fraction of their total leaf‐on period. Our results suggest that the water status of the later growing season, which determines blue oak's proximity to hydraulic damage thresholds, may be fundamentally disconnected from the early growing season factors that drive growth, even though growth itself is highly sensitive to water availability (Stahle et al., ^{2013, 2020}).

A second temporal disconnect that may decouple growth from end‐of‐season water stress is the use of prior year(s) carbon to fuel current year growth. Lagged growth effects, memory effects, or legacy effects are ubiquitous in tree rings (Ogle et al., ²⁰¹⁴; Anderegg et al., ²⁰¹⁵; Klesse et al., ²⁰²³). In blue oak at dry sites and during dry years, turgor‐driven limits to radial growth likely manifest early in the growing season (when predawn water potentials drop below −0.8 MPa; Körner, ²⁰¹⁵). Consequently, the carbon fixed in much of the leaf‐on period will support respiration and future growth rather than current growth. Thus, the end‐of‐season water limitation we document most likely influences future growth.

Finally, our HOTTER simulations may suggest a nontrivial role for respiration and possibly structural overshoot in determining late growing season carbon balance. At the maximum and mean daily VPDs during the sampling period, trees with a negative carbon balance (respiration exceeded GPP) showed a strong negative relationship between NPP and total tree leaf area (trees with more leaves had more negative carbon balance, Fig. S7). While HOTTER's respiration has not been as rigorously validated as its hydraulics, observations of Ψ_PD at or beyond the point of stomatal closure (Fig. 2a) suggest that a substantial number of trees never opened stomata during the day, supporting the importance of end‐of‐season respiration for tree carbon balance.

Our results also raise a subsidiary question: Why do the spatial controls on blue oak growth not mirror the interannual controls on growth? Blue oak is a canonical dendroclimatology species (Stahle et al., ²⁰¹³), often considered ‘better than rain gauges’ for reconstructing precipitation. Yet spatial variation in growth was not related to either tree‐level end‐of‐season water availability, water‐related climate/meteorological variables, or their interaction with A _l : A _s. Instead, spatial variation in growth was correlated with T _min, suggesting that processes such as the onset of leaf out may control site‐to‐site variation. Recently, contrasting spatial and temporal growth sensitivities to climate have been documented in multiple tree species and attributed to local adaptation or acclimation (Canham et al., ²⁰¹⁸; Perret et al., ²⁰²⁴). This appears to strongly be the case for blue oaks and is supported by the large among‐site variation in leaf and allocation traits (Table S1), hydraulic traits other than P50 (Anderegg et al., ²⁰²⁴), and ecotypic variation in seedling performance (McLaughlin et al., ²⁰²²; Poudel et al., ²⁰²⁵).

Traits have complicated cause‐vs‐effect relationships with growth and hydraulic risk

End of season leaf traits, with the exception of leaf size, were generally unrelated to end of season plant water potentials (Fig. S6), water balance‐related climate variables (Fig. 5), and 5 or 1 yr growth rates (Fig. 7). Leaf to stem allocation, either expressed as A _l : A _s or M _l : M _s, was strongly negatively related to growth rates. This directly contrasts with theory (Lambers & Poorter, 1992) and observations in herbs (Poorter et al., ²⁰¹²) and may suggest that branch‐level A _l : A _s is the consequence rather than cause of growth variation. At the level of individual branches, A _l : A _s independently increases with branch leaf size and decreases with branch length, potentially suggesting a joint allocation (leaf area) and structural (branch length) control on branch‐level A _l : A _s, with longer branches requiring more branch tissue (Fig. S10, Table S4). Alternatively, it could suggest that A _l : A _s negatively impacted growth through a structural overshoot, which HOTTER simulations suggest to occur under high VPDs (Fig. S7). Full simulations for the entire growing period, or better yet 5 yr growth period, would be needed to actually test for overshoot (hydraulic or carbon costs exceeding carbon returns). However, the 5 yr basal growth period included the end of massive 2012–2016 drought (Williams et al., ²⁰²²), and the 1 yr branch growth captured a dry year as well (Fig. 1b). It is possible that high allocation to leaf area was maladaptive during both of these periods.

Ultimately, the large site‐to‐site variation in all traits suggests that traits are responding to a complex combination of environmental cues and endogenous factors (e.g. growth rate) that we still do not fully understand. Functional ecology often focuses on the consequences of traits for fitness components such as growth or survival (Violle et al., ²⁰⁰⁷). However, our results highlight that, particularly within species, trait variation can be challenging to interpret, as it represents (presumably) adaptive responses to maximize sometimes orthogonal fitness components. For instance, in blue oak, leaf morphology and allocation to leaves vs stems may be simultaneously maximizing early growing season carbon gain and minimizing end of growing season mortality risk due to hydraulic damage, which we found to be largely unrelated. Add in the feedback effects of growth rate on traits themselves and the difficulty of disentangling cause vs effect becomes extremely challenging. The mechanistic modeling application presented here was a first attempt to synthesize the interactions between supply limitation, evaporative demand, and allocation traits and could provide a useful approach for exploring the cause of the disconnection between traits, growth, and end‐of‐season drought stress.

Conclusion

We conducted a range‐wide survey of blue oak supply limitation and drought stress at the end of a substantial drought year, visiting 15 sites over a mean annual precipitation gradient from 280 mm to 1242 mm. We found a remarkable disconnect between site water balance, end‐of‐season drought stress, growth, and drought avoidance‐related allocation traits. These results highlight our profound uncertainty about the edaphic environment of deeply rooted tree species and the importance of illuminating Critical Zone hydrology to understand drought‐induced mortality. Our results also illustrate how, particularly in Mediterranean‐type climates, the temporal mismatch between water stress and the early season conditions that determine growth and allocation traits challenges our simple assumptions about maximizing carbon gain and organismal performance under water limitation.

Competing interests

None declared.

Author contributions

LDLA, RPS, PP, DDA and TED conceived of the project ideas; LDLA, RPS, PP and ATT designed the methodology; LDLA, RPS, JD and PL collected the data; ATT performed model simulations; LDLA analyzed the data and led the writing of the manuscript. All authors contributed to editing the manuscript.

Disclaimer

The New Phytologist Foundation remains neutral with regard to jurisdictional claims in maps and in any institutional affiliations.

Supporting information

Fig. S1 Demonstration of the size‐standardization of basal area increment (BAI) using quantile regression to create % of max BAI.

Fig. S2 Example of how A _L : A _S does not interact with the Ψ_MD–Climatic Water Deficit relationship.

Fig. S3 Relationship between Ψ_MD/δ¹⁸O and tree DBH.

Fig. S4 Water potential relationships with δD (compare to Fig. 3 with δ¹⁸O in main text).

Fig. S5 Relationships between meteoric water isotopes and stem water isotopes and predawn water potentials, and the relationship between stem water isotopes and predawn water potentials after controlling for meteoric differences among sites.

Fig. S6 Pairplots of water potentials and leaf/stem traits.

Fig. S7 Sensitivity of HOTTER simulated Net Primary Productivity (NPP) to tree leaf area for the max, mean and min VPD simulations.

Fig. S8 HOTTER results compared to 1 yr branch growth (compare to Fig. 8 in the main text with 5 yr radial growth).

Fig. S9 ‘Dual isotope’ plot (δD vs δ¹⁸O) of tree stem water isotopes.

Fig. S10 Branch level A _L : A _S as a function of both leaf size and branch length.

Methods S1 Description of climate variable selection philosophy and procedure.

Methods S2 Full details of the HOTTER model simulations.

Methods S3 Full exploration of interactions between A _L : A _S and the growth– or Ψ_MD–climate relationships.

Table S1 Full description of climate and soil variables used as predictors.

Table S2 Summary of water status, traits and growth and their best environmental predictors.

Table S3 Statistical model results predicting leaf water potential from stem water stable isotopes, with and without the outlier site.

Table S4 Statistical model results from mixed effects models relating log (A _l : A _s) to log(leaf size) and log(branch length).

Please note: Wiley is not responsible for the content or functionality of any Supporting Information supplied by the authors. Any queries (other than missing material) should be directed to the New Phytologist Central Office.

Data availability

Data are available in the Zenodo repository DOI: 10.5281/zenodo.19365699. Analysis code is available on Github at https://github.com/leanderegg/BlueOakLandscape.