Observed forest trait velocities have not kept pace with hydraulic stress from climate change

The extent to which future climate change will increase forest stress and the amount to which species and forest ecosystems can acclimate or adapt to increased stress is a major unknown. We used high‐resolution maps of hydraulic traits representing the diversity in tree drought tolerance across the United States, a hydraulically enabled tree model, and forest inventory observations of demographic shifts to quantify the ability for within‐species acclimation and between‐species range shifts to mediate climate stress. We found that forests are likely to experience increases in both acute and chronic hydraulic stress with climate change. Based on current species distributions, regional hydraulic trait diversity was sufficient to buffer against increased stress in 88% of forested areas. However, observed trait velocities in 81% of forested areas are not keeping up with the rate required to ameliorate projected future stress without leaf area acclimation.


| INTRODUC TI ON
Biogeographic range boundaries are highly visible outcomes of ecosystem functionalities due to spatial variations in climate and other limiting resources. However, the fundamental mechanisms controlling tree species distributions and range boundaries are still poorly understood, despite over two centuries of study (Anderegg & HilleRisLambers, 2019;Colwell et al., 2008;Louthan et al., 2015;MacArthur, 1984;Von Humboldt & Bonpland, 2010). Understanding the mechanisms regulating forest range dynamics is urgent, given that climate change has already caused widespread shifts in tree species' geographic extents across multiple biomes (Chen et al., 2011;Colwell et al., 2008;Esquivel-Muelbert et al., 2019;Lenoir et al., 2008), and it is expected that hotter droughts with further climate change have the potential to cause widespread declines in forest productivity and increases in mortality (Park Williams et al., 2013). A continuation or escalation of the observed trends could cause drastic range shifts or biome collapses (Cano et al., 2022;Cox et al., 2000) and flip the terrestrial carbon sink to a carbon source (Adams et al., 2010;Kurz et al., 2008).
There are several mechanisms by which tree species and forested communities can compensate for intensifying water stress associated with climate change. At the species level, one such mechanism that can mitigate stress is the coordinated adjustments in plant biomass allocation, such as adjustments in root or leaf allocation (Barton & Montagu, 2006;Mackay et al., 2020). Here, we focus on widely documented adjustments of leaf area relative to sapwood area (referred to subsequently as "adjustment in leaf area") within the same species across climate gradients such that individuals within a species decrease water demand (leaf area) relative to supply (sapwood area) in drier locations or seasons Martínez-Vilalta et al., 2009;Mencuccini & Grace, 1995;Piñol & Sala, 2000;Rosas et al., 2019). Various other physiological adjustments can buffer against drought stress within a species (e.g., local adaptation or acclimation of more embolism-resistant xylem or leaf tissue, deeper roots, etc.), but we focus on adjustments in leaf area here because allocation traits generally show much larger withinspecies variation than most other drought-related traits (Anderegg, Griffith, et al., 2022;Trugman, Anderegg, Sperry, et al., 2019). At the community or stand level, there can be shifts in composition toward species with more drought-resistant plant hydraulic traits that can buffer a forest against water stress (Allen & Breshears, 1998;García-Valdés et al., 2021;Trugman et al., 2020).
Finally, it is possible for biogeographic shifts to occur where more drought-resistant species expand their ranges to new stands over longer time horizons, provided that sufficient hydraulic trait reservoirs are available from proximal seed sources (Peters et al., 2015).
These species shifts may take decades during which ecosystem services and forest carbon storage will decline (Seidl & Turner, 2022).
Trait-based vegetation models simulate the physiological processes hypothesized to be fundamental in governing terrestrial ecosystem responses to the variation of climate across space and climate change in the future, while being able to connect organ-level hydraulic traits measured in the field to organism-and ecosystemlevel function. Yet few vegetation models incorporate the appropriate processes to test the extent to which shifts in leaf allocation within a species (Trugman, Anderegg, Wolfe, et al., 2019) or biogeographic changes in species composition may ameliorate climateinduced changes in forest stress. Here, we tested the extent to which changes in species biogeographic ranges, shifts in species relative abundance, and adjustment in leaf area within a species can compensate for increases in water stress with climate change. We used an optimization-based model of tree gas exchange, hydraulic transport, and leaf carbon allocation ( Figure S1). The model was parameterized with current forest hydraulic trait distributions for the continental United States based on US Forest Inventory and Analysis data, tree height, and canopy efficiency Figures S2-S5) and forced with historical (1995)(1996)(1997)(1998)(1999)(2000)(2001)(2002)(2003)(2004)(2005)(2006)(2007)(2008)(2009)(2010)(2011)(2012)(2013)(2014)

| MATERIAL S AND ME THODS
We used the HOTTER model (the Hydraulic Optimization Theory for Tree and Ecosystem Resilience model; Mathias & Trugman, 2022;Trugman, Anderegg, Wolfe, et al., 2019; Figure S1), a physiologically based tree model with a realistic representation of gas exchange (Eller et al., 2018) and a detailed representation of plant hydraulics (Trugman et al., 2018) to quantify spatial variations in tree water status, hydraulic stress, and carbon gain across gradients in climate and plant traits in the continental United States. Our model experiments combined hydraulic trait maps based on high-resolution species distribution and abundance data derived from the US Forest Service Forest Inventory and Analysis Program with a large hydraulic trait database , tree height measured remotely by satellite, and daily historical and future climate forcing data (see Section 2). This combination of physiological traits and climate shaped the geographical distribution of hydraulic stress and carbon gain in trees for present-day conditions (1995)(1996)(1997)(1998)(1999)(2000)(2001)(2002)(2003)(2004)(2005)(2006)(2007)(2008)(2009)(2010)(2011)(2012)(2013)(2014) and their response to future conditions (2081-2100).

| HOTTER model description
HOTTER combines key physiological processes of photosynthesis, autotrophic respiration, and plant hydraulics to simulate plant water status, carbon gain, and hydraulic stress. Here, we updated previous versions of the model (Mathias & Trugman, 2022;Trugman et al., 2018), to include a new stomatal optimization, the Stomatal Optimization Based on Xylem Hydraulics (SOX) (Eller et al., 2018). In this new version, stomatal behavior is governed by a leaf scale tradeoff between the carbon gain of assimilation and a cost function based on the percent loss of conductivity in the tree xylem. The stomatal model was chosen to allow investigation of the response of the tree hydraulics to variations in traits and under future climate conditions. HOTTER model meteorological forcings include atmospheric CO 2 concentrations (CO 2 , ppm), temperature (T, C°), vapor pressure deficit (VPD, Pa), and soil water potential (ψ soil , MPa). In addition, the tree model requires inputs or allometric equations for the state of the tree, leaf area (al, m 2 ), height (h, m), and diameter at breast height (dbh, cm 2 ). In these experiments, we use an observed dataset of height as an input, determine dbh via allometry, and determine leaf area through an optimization that maximizes net primary productivity (NPP) over a given time period (Figures S4, S5, S8; see Section 2.1.3 for details). The hydraulic dynamics of the model are primarily controlled by three plant physiological traits: the water potential and 50% loss of conductivity (P50, MPa); a slope parameter (b2 in the model) which governs how plant conductivity decreases with increased water stress (e.g., more negative plant water potentials); and the conductivity of the roots, xylem, and petioles (K max , mmol H 2 O m −1 s −1 MPa −1 ). In our experiment, we set the P50 values based on the maps (Trugman et al., 2020, p. 20) (Figure S2), kept the slope parameter fixed for parsimony, and determine the K max using an empirical function representing a safety versus efficiency tradeoff where higher conductivity values are associated with less negative (i.e., more vulnerable to hydraulic stress) P50 values ( Figure S3). For the entire experiment, we used an observed relationship between P50 and K max derived from Liu, Ye, et al. (2021) for angiosperms, which included the most observations and was broadly similar to gymnosperms.
HOTTER represents a tree canopy as a single leaf layer without shading and assumes that under average growth conditions, the two limiting photosynthetic rates should be equal resulting in a "maximum canopy efficiency" ( Figure S5; Smith et al., 2019;Trugman, Anderegg, Wolfe, et al., 2019;Walker et al., 2014; see Section 2.1.2 for further details). The canopy efficiency is sensitive to temperature through a fixed Q 10 response to temperature (Mathias & Trugman, 2022).
Finally, each tree organ, including leaf area, xylem, phloem, and roots, includes respiration that is also sensitive to temperature through a fixed Q 10 response to temperature (Mathias & Trugman, 2022). The combination of these respiration terms, a 30% growth respiration, and carbon assimilation through photosynthesis allowed the model to predict gross primary productivity (GPP), autotrophic respiration (Rh), and the difference between the two-net primary productivity (NPP)-in different climates and for different traits and states of trees.
Additionally, HOTTER diagnostic outputs include tree-level transpiration (Tr), and the water potential (ψ) and percent loss of conductivity (PLC) for all of the tree's conductive elements: roots, xylem, and leaves.
Model outputs are daily level and assume 12 h of daylight for carbon assimilation during the growing season and 24 h of respiration. See Table S1 for a summary of HOTTER parameterizations and outputs.

| Incorporation of a cost-gain stomatal representation
The SOX stomatal model has shown skill in simulating stomatal responses to drought (Eller et al., 2018, p. 202) and cost-gain models have been developed to represent the hydraulic changes observed in the field Sabot et al., 2022;Sperry et al., 2016Sperry et al., , 2017Venturas et al., 2017Venturas et al., , 2018Wang et al., 2020). Our implementation of the SOX model uses the PLC of the xylem to represent the cost function (k xylem ) and the assimilation rate (A) in the tree model to represent the gain function. Then for each daily time step, the optimal internal CO 2 concentration, and thus stomatal conductance, to maximize the gain-cost function is determined (Equation 1). to capture the dynamics from "leaky stomata" (Machado et al., 2021), we have included a minimum stomatal conductance of 0.002 mol m −2 s −1 (Bonan, 2002, p. 246). Optimization was performed using the Scipy minimize function "fmin" (Virtanen et al., 2020).

| Canopy efficiency
As a simplification, a tree canopy in HOTTER is comprised of a single leaf layer and it is assumed that two limiting photosynthetic rates (i.e., the photosynthetic rate limited by the maximum rate of Rubisco carboxylation and the photosynthetic rate limited by the electron transport rate for the regeneration of ribulose-1,5,-bisphosphate) should be equal according to Smith et al. (2019), resulting in a "maximum canopy efficiency" under average growing conditions. For the calculation of canopy efficiency, we used a spatially varying climatology during the growing season of insolation created from daily data ( Figure S6d). Canopy efficiency is sensitive to temperature as in Mathias and Trugman (2022). Scaling productivity from the leaf to the canopy scale is accomplished by multiplying leaf scale productivity by total leaf area ( Figure S5). Note that this scaling includes the simplifying assumption of zero shading between leaves in the canopy.

| Leaf area optimized for maximizing carbon gain
In all model experiments, the leaf area of the model tree in each climate grid cell was determined through optimizing tree carbon gain (i.e., the NPP) to maximize integrated NPP over the 20-year simulation period (either 20-year historical or future climate period). As leaf area increased so did productivity (through the increased canopy size) but the carbon loss through respiration from leaves, stem, and roots (determined as proportional to canopy size) also increased.
In addition, greater leaf areas increase transpiration, which increases hydraulic stress for a given ψ soil , triggering stomatal closure. The balance of these opposing mechanisms led to a unique (i.e., determinable through optimization) leaf area for maximizing carbon gain. The optimization was performed using the SciPy optimization package in Python (Virtanen et al., 2020). The optimization method used was the "Nelder-Mead" with maximum iterations of 20 and a leaf area tolerance of 0.2.
The GLDAS-2 data are a combination of satellite and ground-based observations assimilated into land surface models. The data were primarily chosen because of a track record of use in modeling the land surface, relatively high spatial resolution (0.25°), and a daily time resolution which enabled model predictions for day-to-day variations in water stress. While the uncertainty due to meteorology is not explored here, previous work predicting carbon fluxes has shown that the choice of driving meteorology can lead to variations of 9% in global gross primary productivity (Wu et al., 2017). The data can be downloaded from https://disc.gsfc.nasa.gov/. (1) The vapor pressure deficit was calculated from daily mean values of temperature and specific humidity. Soil water potential was calculated from root zone soil water content based on Campbell (1974) as implemented in Mathias and Trugman (2022).

| Tree height
We determined tree height as a model input from remote-sensed observations of tree height from the Global Ecosystem Dynamics

Investigation (GEDI) instrument and processed by the Global Land
Analysis and Discover team at the University of Maryland (Potapov et al., 2021). The data are a snapshot of tree height from April to October 2019. The native resolution was 0.00025 × 0.00025 degree and we re-binned the data to the 0.25 × 0.25-degree grid of the meteorology. Re-binning was performed through an area-weighted mean of the higher resolution observations to the coarser grid ( Figure S4). Note that while the coarsened mean values are representative of each larger pixel, the native high resolution of the dataset shows there is also variability in height. This variability represents another dimension of uncertainty and potential acclimation not directly investigated in this paper ( Figure S4b). In addition, absolute uncertainty is likely to increase in tall dense forests such as the Pacific Northwest and Northeast (Liu, Cheng, & Chen, 2021 Figure S3). Here, we used the tradeoff fit to angiosperm observations determined in Liu, Ye, et al. (2021) which had the most observations and were similar to gymnosperm.

| SPEI drought index
The

| Climatological water deficit
Climatological water deficit was used to determine dry regions (greater than 100 mm average monthly deficit during the growing season (June, July, August)) and wet regions (<100 mm average monthly deficit during the growing season (June, July, August)).
We used the climatology of climatological water deficit from TerraClimate calculated between 1981 and 2010 and regridded through linear interpolation to the 0.25 degree HOTTER resolution (Abatzoglou et al., 2018). The data were downloaded as a netCDF  (Mathias & Trugman, 2022). In addition, we find that present hydraulic stress in our runs with HOTTER (as represented by the 90th percentile of PLC and the stressed seasons experienced from 1995 to 2014) show skill in separating regions with high tree mortality for both dry and wet regions (Figure 3). To test this separation, we performed the Mann-Whitney U-test (test for difference in distributions) (Mann & Whitney, 1947) and the t-test (difference in mean values) (Yuen, 1974) between the observed mortality in two bins of HOTTER hydraulic stress.
The SciPy package in python was used to perform both tests (Virtanen et al., 2020; see Table S2). This relationship between modeled hydraulic stress and mortality, along with previous model validation, supports the utility of HOTTER in exploring the present-day and future hydraulic stress of forests in the contiguous United States. In addition, HOTTER shows an increased ability to separate areas of mortality versus other points when compared with the purely climate-driven SPEI index (Figure 3).

| Change in percent loss of hydraulic conductivity
Where the percent loss in hydraulic conductivity (PLC) is reported, we use the 90th percentile calculated on the time series of days to highlight the days with higher stress as well as the difference between 90th percentiles in the present and future. To match this, we also report the 10th percentile (most stressed) days of the leaf water potential in Figure 1.
For the present-day histograms, we present only the days more negative (greater) than the 10th (90th) percentile for water potential (PLC) (Figure 1; Figure S9). For the difference with the future, we present the difference between days that are more negative (greater) than the 10th (90th) percentile for water potential (PLC) (Figure 1; Figure S9).

| Defining a hydraulically stressed season
Here, we defined a season with potentially damaging hydraulic stress as a season with any 1 day exceeding a PLC >50%. In the present-day runs, we found that points that had at least one stressed season had more mortality (as measured in (Anderegg, Chegwidden, et al., 2022)) than points with zero stressed seasons (Figure 3).
The distributions of mortality separated in this way were significantly different (p ≪ .001) by both a Mann-Whitney U test (Mann & Whitney, 1947) and t-test (Yuen, 1974) as implemented in the Scipy package in Python (Virtanen et al., 2020) (see Table S2).

| Uncertainty analysis of future differences
We performed a bootstrap analysis to estimate the uncertainty in the difference of the 90th percentile PLC values (and 10th percentile leaf water potential values), as well as their differences. To perform the bootstrap analysis, we randomly selected 50% of the modeled days from the present, future, and both when doing a difference. We did this 100 times with replacement of days and calculated the 90th percentile each time for the present and future, as well as 100 times before calculating the difference between them. Uncertainty is reported as the 5-95th range of this ensemble of estimates. The 5-95th range is reported in Figure S10. Stippling was provided in Figures 1e-g and 5a,b where a difference had more than 5% of the distribution less than zero.

| Present-day runs 1995-2014
HOTTER experiments for the present day were run at a daily time step over the growing season (June, July, August) for the 20 years F I G U R E 1 Climate change increases chronic hydraulic stress while geographic gradients in aridity and hydraulic traits mediate patterns of plant water status and stress. Model-predicted most stressed (10th percentile) daily leaf water potential (a) and most stressed (90th percentile) daily plant percent loss in conductivity (PLC) (c) calculated from 1995 to 2014 growing seasons. Change (future-historical) in the percentile of most stressed daily water potential (e) and the percentile of most stressed daily difference in PLC (g). Negative values in (e) indicate future decreases in plant water availability and positive values in (g) indicate future increases in stress. Maps show results for simulations parameterized with community mean hydraulic traits (CW mean). Histograms (b, d) include the 10% most stressed days, (f, h) include the difference between the 10% most stressed days in the present and those days in the future for each grid cell for hydraulic trait maps ranging from most drought vulnerable to least drought vulnerable species in a grid cell; "regional" corresponds with the most extreme trait values in a 0.25° grid cell, "plot" corresponds with average within-plot trait extremes, "CW" represents the min/mean/max communityweighted means across all forested plots in a grid cell.  Figure S2; Table S3).
For each experiment, tree leaf area is adaptively acclimated to maximize NPP for the entire 20-year simulation.
We ran sensitivity tests with different trait distributions used to parameterize the model, including community-weighted hydraulic traits weighted by species basal area within a forest plot, as well as the least and most vulnerable hydraulic traits co-located both within a plot and within a 0.25° climate grid cell ( Figures S2 and S3). These sensitivity tests represent the range of scenarios where there may be shifts in plot composition toward species with more droughtresistant plant hydraulic traits or expansion of drought-resistant species to new stands within a region (i.e., a 0.25° climate grid cell where seed dispersal may be possible).

| Future runs 2081-2100
For the future runs, daily meteorology for temperature and vapor pressure deficit were created using the median values of the change in climate predicted by six models participating in CMIP6 for the "medium" SSP3-7.0 scenario that we chose to best match expectations for the trajectory of CO 2 emissions (see Data Sources). In addition to changing temperature and vapor pressure deficit, we applied a mean value of the change in atmospheric CO 2 concentrations (753 ppm). For these future experiments, we left the soil water potential unchanged from the present day as there is large uncertainty in soil water content change across models and the general prediction is that there will be relatively small changes (Mathias & Trugman, 2022). For future runs, we ran a number of different scenarios: with and without leaf acclimation, and attributing change to each of temperature, vapor pressure deficit, and atmospheric concentrations of CO 2 . Collectively, these experiments enabled us to quantify the impact of individual future climate factors on forest stress and the extent to which adaptive acclimation in leaf area within a species may mitigate increased climate stress. See Table S3 for the experimental summary.

| Leaf acclimation
Future runs were performed using both the leaf area determined from the optimization in the present ("Fixed Leaf Area") and for leaf area that was optimized to maximize NPP over the future (2081-2100) period where temperature, vapor pressure deficit, and atmospheric concentrations of CO 2 change to expected future conditions consistent with the SSP 3-7.0 scenario in CMIP6 ("Acclimated Leaf Area") ( Figure 4; Figures S8 and S12).

| Factorial runs
We quantified the effect of individual future climate factors, the adjustment of leaf area, and hydraulic trait diversity with a set of factorial experiments using the tree hydraulic model. We systematically included individual future forcings in isolation for the following climate factors combined with historical climate variables for the rest of the model forcing: VPD, atmospheric CO 2 , and temperature (which impacted tree respiration and photosynthetic chemical reactions according to Mathias and Trugman (2022), but not VPD). We also tested the effect of adjustments in leaf area with experiments where leaf area remained optimized to historical conditions ("fixed leaf area") rather than being able to adjust to the future climate ("acclimated leaf area") ( Figure S8).

| Software
The numerical modeling and analysis were performed in Python with the support of Anaconda installation with particularly heavy use of the following open-source modules: NumPy, Scipy, Xarray, Matplotlib, Cartopy (Hoyer & Hamman, 2017;Hunter, 2007;Met Office, 2010;van der Walt et al., 2011;Virtanen et al., 2020).

| RE SULTS AND D ISCUSS I ON
We first generated continental-scale maps of plant water status for the historical period for two widely used diagnostics of plant water status and water stress (respectively), leaf water potential, and percent loss in hydraulic conductivity (PLC). PLC characterizes the extent to which tree xylem transport vessels or tracheids are blocked by air bubbles formed by hydraulic stress which can be induced either by low water availability, high evaporative demand, or both . Elevated PLC values have been linked to elevated tree mortality (Anderegg et al., 2015;Venturas et al., 2018).
To capture the period of maximum stress in trees, we focus on days when leaf water potential was in the 10th percentile (e.g., the most negative leaf water potentials) and PLC was in the 90th percentile (e.g., PLC was highest) in each forested grid cell (see Material and Methods). We found that historical leaf water potential varied substantially across space due both to variations in aridity and hydraulic traits. The drier areas in the western United States had much more negative leaf water potentials, both because of drier soils and higher vapor pressure deficits (VPDs), and also because of the biogeographic ranges of more drought-tolerant tree populations that have more resistant hydraulic traits which allow trees to continue functioning at more negative water potentials without risking hydraulic damage (Figure 1a). Regions already experiencing elevated hydraulic stress in historical simulations included dry regions such as the US Mountain West where many species are resistant to water stress but climate change has increased drought frequency and severity (Diffenbaugh et al., 2015;Marvel et al., 2021). Despite wetter baseline conditions, parts of the Midwest and eastern United States also saw elevated PLCs because of more hydraulically vulnerable trait compositions (Figure 1c). This may be due in part to microsite variations in water availability that are not captured by our climate forcing, resulting in overprediction of stress in some midwestern areas where forest cover is sparse and restricted to riparian areas. Also, note that parts of the Midwest that show large stress in the present day are also some of the most uncertain regions ( Figure S10).  Figure S11). Such novel climates make it particularly important to understand the fundamental mechanistic trade-offs that trees face with increasing climate stress. Interestingly, warmer future temperatures increased the respiratory cost of leaves relative to their carbon gain potential, reducing leaf area enough to ameliorate hydraulic stress (Materials and Methods). Depending on species' drought vulnerability, some increases in PLC were mediated by more drought-resistant traits either within local forest plots or the regional climate grid cell (Figure 1b,d,f,h). However, locations throughout the Mountain West, Midwest, and Southeast were found to be already experiencing potentially lethal levels of stress (Figures 1c and 2a). Climate stress in these regions is expected to accelerate under future climates (Figures 1e-g and 2b,c), a pattern that has been documented in numerous observational studies of widespread tree mortality (Anderegg et al., 2012;Anderegg, Wu, et al., 2022;Breshears et al., 2005, p. 20015;Hammond et al., 2022;Figure 3a-d).

With projected increases in temperatures and
Model-predicted hydraulic stress, either through elevated PLC or stressed seasons, corresponded well with forest inventoryobserved elevated mortality (Figure 3a-d). While tree mortality has proven difficult to predict, we find that high values of present-day PLC or stressed seasons in the model relate to higher levels of observed mortality and offered substantial improvements over standard drought metrics such as SPEI (Figure 3e,f; Kolb, 2015;Mathias & Trugman, 2022;Venturas et al., 2021). Specifically, we find significantly elevated rates of mortality at grid points where the 90th percentile of daily PLC is greater than 25% or where there has been F I G U R E 2 US forests are already experiencing growing seasons with potentially lethal levels of hydraulic stress, a trend which will likely continue with future climate change. The number of stressed historical growing seasons (a). Change in stressed growing seasons between historical and future climates with adaptive acclimation in tree leaf area (b) and difference (e.g., leaf acclimation effect) in the future with and without leaf area acclimation (c). Maps show results for simulations parameterized with community mean hydraulic traits. Gray coloration indicates no seasons with PLC >50%. For (a), redder colors indicate more seasons with PLC >50% with a maximum of 20 seasons over the 20-year historical simulation period. For (b, c), redder colors indicate an increase in the number of seasons with PLC >50% (e.g., an increase in potentially lethal stress), yellow indicates the same number of seasons with PLC >50% (e.g., no change in potentially lethal stress, but lethal stress predicted), blue indicates a decrease in seasons with PLC >50% or (b) between future climates with and without adaptive acclimation in leaf area. Map lines delineate study areas and do not necessarily depict accepted national boundaries.
at least one stressed season in the study period-this is true for both dry and wet regions (Figure 3a Model projections indicate that most of the stressed daily PLC values will increase by ~1-2% with future climate change. While this increase in PLC appears small in comparison to the measurement precision of current hydraulic techniques , this across-the-board increase in chronic stress is concerning for longterm forest health because it represents a shift in baseline stress conditions during times of already high stress (related to tree mortality (Figure 3)), which feedback to further decrease plant health through "cavitation fatigue" (Hacke et al., 2001), or legacy effects on growth (Anderegg et al., 2015). Our results show that the majority of US forests may face elevated chronic stress moving forward that will be compounded when trees experience expected increases in acute stress due to higher frequency droughts associated with novel climate extremes (Diffenbaugh et al., 2015). This means that even drought-resistant species may become more vulnerable to drought events over time due to chronic mechanical stress in hydraulic tissues that results in higher conductivity losses at less negative water potentials (e.g., at higher water availability) for the same tree due to cavitation fatigue (Hacke et al., 2001).  Table S2.
temperature and VPD ( Figure S7), the hot and dry Mountain west (MW), and the hot and wet South (Figure 4; Figure S12). Across all locations, all else equal, elevated CO 2 levels had an ameliorating effect on water stress due to increased water use efficiency (Walker et al., 2021). Projected increases in VPD increased stress across all regions, though the magnitude of this effect varied widely depending on the region. Temperature change did not have a large effect with fixed leaf area. We also found that both traits and adjustment in leaf area had the potential to alter future forest stress. In particular and depending on location, adjustment in leaf area to maximize car-  Figure S12. Map lines delineate study areas and do not necessarily depict accepted national boundaries. [Farrior et al., 2013]). The ability to avoid some future stress through acclimation highlights the need to incorporate leaf area allocation strategies in response to plant water stress in prognostic terrestrial biosphere models along with the demographic changes in the ecosystem (Fisher et al., 2015;Trugman, Anderegg, Wolfe, et al., 2019).
Predicting biogeographic transitions in forest species across climate gradients from first principles has been an unsolved problem in the ecological and vegetation modeling communities for over a decade (Fisher et al., 2015). Here, we use a hydraulically enabled vegetation model parameterized with a hydraulic trait database and high-resolution surveys of species distribution and abundance to diagnose current and future projected climate stress of US forests.
We find that climate change will likely increase chronic water stress in US forest, potentially resulting in shifts in the composition and biogeography of forests toward more drought-tolerant trees, even with within-species leaf area acclimation. Without shifts in forest composition or acclimation, increased water stress may accelerate tree mortality, offsetting potential carbon sink gains from CO 2 fertilization. F I G U R E 5 Species range shifts could buffer against future climate-induced hydraulic stress in a majority of locations, but observed trait velocities are not currently keeping pace with the required compositional shifts to mitigate increases in stress. Shift in P50 under future climates relative to current community mean P50 required to avoid increased hydraulic, without (a) and with (b) adaptive leaf area acclimation in future climates. Gold, orange, and brown indicate at least some shift toward more drought-resistant (more negative) P50s is required to avoid increased chronic daily stress. Dark gray values are areas where no P50 values within the region are sufficient to avoid increased hydraulic stress. Turquoise indicates that no shift in P50 is required to avoid increases in stress. Purple, green, and pink indicate that shifts toward more drought vulnerable P50 values (less negative) are possible without increasing stress. Circles indicate the relative abundance of grid cells associated with each color. (c) The distribution of P50 velocities observed (gray) from the US Forest Inventory across the continental United States compared to model required values in (a, b) for predictions with a fixed leaf area (blue) and acclimated leaf area (red). (a, b) White stippling notes areas where the points are where the 5-95th percentile overlaps zero. Map lines delineate study areas and do not necessarily depict accepted national boundaries. 2018-67012-31496.

CO N FLI C T O F I NTE R E S T S TATE M E NT
None.

DATA AVA I L A B I L I T Y S TAT E M E N T
The analysis code supporting this publication is available online at