We employed a gradient analysis that included 12 sites evenly distributed among three forest types (western juniper, ponderosa pine, and grand fir) along the eastern slopes of the Cascade Mountains, Oregon, which ranged in elevation from 929 to 1560 m above sea level (Fig. 1a, b). Guided by maps of forest type (Kagan et al., 2006), historical wildfires (Eidenshink et al., 2007), and land ownership we selected sites in areas with mature (> 80 years), publicly owned forest that showed minimal evidence of recent natural (e.g., fire, insect) or human disturbance (e.g., thinning). Furthermore, sites were situated within 100 m of a road to facilitate access. We included the long-term AmeriFlux Metolius Mature Pine (US-Me2) eddy covariance flux tower (Law et al., 2004) as one of our ponderosa pine sites. Additionally, we collected samples at three relatively young stands, one for each species, which included the Metolius Young Pine Burn flux tower site (US-Me6). These data are presented in Table S2, though were not included in our analysis given that the mean cambial age at breast height was 17 to 38 years, depending on site.

At each new site we established a randomly located 1 ha plot, as per the Global Terrestrial Observing System (GTOS) protocol (Law et al., 2008), though used pre-existing plots at US-Me2 and US-Me6. The plot design, similar to that used by the USDA Forest Service Forest Inventory and Analysis program, included one central subplot and three subplots arranged in a circular pattern around the center and offset from the center by 35 m. Subplot radii were held consistent within a site but varied among sites from 10 to 17 m for trees and from 5 to 17 m for saplings, depending on stem density. Stems with a diameter at breast height (DBH; 1.4 m) greater than 10 cm were considered trees, while those with DBH from 1 to 10 cm were considered saplings. We tagged every tree (n=730) and sapling (n=39) in the study with a unique identifier. Seedlings (DBH < 1 cm) were not sampled.

Field activities included stem survey measurements, collection of tree core and foliage samples, and optical measurements of leaf area, as per the GTOS protocol (Law et al., 2008). We measured DBH and total height (H) of every tagged stem and then, within each subplot, randomly selected a subset of trees for intensive measurement. Tree H was measured using a laser range finder (Laser Technology Inc., Centennial, USA).We cored five trees of the dominant species per subplot (20 trees per plot) for growth measurements. Tree cores were extracted near breast height using a 5.15 mm increment borer and, if necessary, oriented perpendicular to the slope. We then randomly selected 2 or 3 of the 5 trees per subplot (10 trees total per plot) and collected an additional core from each tree for wood density and sapwood area measurements. For each of the 10 intensively measured trees, we also excised mid-canopy, south-facing branches and collected foliage for measurements of leaf longevity (LL), specific leaf area (SLA), and foliar carbon (C) and nitrogen (N). For ponderosa pine and grand fir, we estimated LL on the excised branches using the “counting cohorts” method (Pérez-Harguindeguy et al., 2013). Since western juniper do not form distinct cohorts we calculated LL based on estimates of foliage biomass and annual leaf litter fall from a nearby site (Runyon et al., 1994). Foliage was then stored following standard protocols prior to laboratory analysis (Pérez-Harguindeguy et al., 2013). We collected additional tree cores and foliage samples when nondominant tree species were present in the subplots. The number of additional trees varied from one to six depending on the diversity and relative abundance of additional species. Lastly, we measured the canopy gap fraction on each subplot using an optical LI-2200 Plant Canopy Analyzer for conifers (LI-COR, Lincoln, USA).

We derived monthly estimates of CMI (Hogg, 1997) for each site as the difference between PPT and ET0 using gridded climate data from 1964 to 2013. The climate data were produced by the Parameter-elevation Relationships on Independent Slopes Model (PRISM; http://www.prism.oregonstate.edu/; Daly et al., 2008) based on climate station observations and spatial modeling at 4 km resolution. We calculated monthly ET0 (mm month-1) using a version of the Hargreaves equation (Hargreaves and Samani, 1985) modified by Droogers and Allen (2002): ET0=0.0013×0.408RTavg+17.0Trng-0.0123PPT0.76, which is based on average daily extraterrestrial radiation (R; MJ m-2 d-1), average daily temperature (Tavg; ∘C), daily temperature range (Trng;  ∘C), and precipitation (PPT; mm month-1). Estimates of R were derived based on equations provided by Allen et al. (1998). Modeled monthly ET0 and PPT agreed relatively well with meteorological measurements at the US-Me2 flux tower site from 2002 to 2012 (r2=0.57–0.64, P< 0.01) and at the US-Me6 flux tower from 2010 to 2013 (r2=0.58–0.65, P< 0.01). For each site we then summarized climate conditions over the growing year (gy; October in year t-1 through September of year t), which involved calculating average Tavg, as well as cumulative PPT, ET0, and CMI (e.g., denoted CMIgy). Lastly, we averaged growing-year climate conditions from 1964 to 2013 to produce 50-year climatologies for each variable (e.g., CMIgy‾).

Summary statistics in this section are for 1964 to 2013 and represent cross-site averages (±1 SD) within each of the three forest types (site-level summaries are given in Table S5). Average Tgy‾ decreased from 8.32 ± 0.80 ∘C at the juniper sites to 6.26 ± 0.72 ∘C at the fir sites, while CMIgy‾ increased from -1250 ± 120 to -280 ± 250 mm yr-1 (Table 2). Across forest types, July was the climatically driest month, while December, conversely, was the climatically wettest month. At juniper sites, CMI tended to be negative (i.e., ET0 exceeded PPT) for 9 consecutive months each year (February–October), whereas CMI was generally negative at pine and fir sites for 8 (March–October) and 7 (April–October) consecutive months, respectively.

Large interannual fluctuation in CMIgy occurred between 1964 and 2013 (Fig. 2 left panel). For instance, 1982 and 1997 were abnormally wet, whereas 1977 and 1994 were abnormally dry. At the juniper sites, CMIgy during these wet years was around 300–400 mm greater than during the dry years. At the fir sites the difference in CMIgy between these abnormally wet and dry years was more pronounced, exceeding 1000 mm. Other notable dry years included 1973, 1992, 2001 and 2003. Conditions were consistently drier than normal from 1985 to 1994, with CMIgy < CMIgy‾ for 8 or 9 years, depending on forest type (Fig. S2 in Supplement).

The 12 sites spanned a gradient in CMIgy‾ from -1240 ± 120 to 50 ± 290 mm yr-1, and across this gradient forest LAI, ANPP, and AGB increased significantly (P< 0.05) and nonlinearly with CMIgy‾, which explained 67–88 % of the variability in these stand characteristics (Fig. 3; Table 3, S2, S5). After accounting for wood interception and clumping, LAI ranged from 0.09 ± 0.03 to 6.15 ± 1.12 across sites, with average (±1 SD) LAI in each forest type increasing progressively from the open-canopy juniper sites (0.26 ± 0.13) to the pine (1.92 ± 0.51) and then closed-canopy fir sites (4.54 ± 1.49; Fig. 3a; Table 4). Annual productivity along this gradient increased from 11 ± 2 to 311 ± 60 g C m-2 yr-1 and averaged 17 ± 4, 185 ± 62, and 238 ± 54 g C m-2 yr-1 at the juniper, pine and fir sites, respectively (Fig. 3b). Aboveground biomass also tended to be quite low at the juniper sites (2.6 ± 0.4 kg C m-2), yet reached higher levels at sites dominated by pine (10.6 ± 4.1 kg C m-2) and fir (14.4 ± 5.9 kg C m-2). Across sites there was a significant (P< 0.01) positive linear relationship between LAI and ANPP (r2 = 0.75). Furthermore, AGB was positively associated with both LAI (r2=0.57) and ANPP (r2=0.68; Table 3).

Annual radial growth varied considerably from 1964 to 2013 (Fig. 2 right panel) and tracked CMI more closely at the chronically dry juniper sites than at the wetter pine and fir sites (Figs. 4, 5). Growth reductions in response to single-year drought events (e.g., 1977, 1994, and 2001) were evident at the juniper sites, whereas reductions were more modest at the pine and fir sites (Fig. 2 right panel). For instance, the 1977 drought led to a 73 % reduction in average radial growth among juniper in comparison to growth during 1976 (0.66 ± 0.05 vs. 0.18 ± 0.03 mm). In contrast, this drought drove a 35 % reduction in average growth among pine (1.74 ± 0.14 vs. 1.08 ± 0.10 mm) and a 7 % reduction among fir (1.87 ± 0.14 vs. 1.74 ± 0.14 mm).

The vast majority of juniper (91 ± 3 %) exhibited a significant positive RWI–CMI correlation (r¯RWI-CMI= 0.52 ± 0.02) and responded most strongly to CMI over the 6 to 12 months leading up to late spring and summer (Fig. 4a; Table S6). Around 51 ± 6 % of pine (r¯RWI-CMI=0.27 ± 0.02) and 12 ± 4 % of fir (r¯RWI-CMI=0.11 ± 0.02) also exhibited a significant positive correlation with CMI, yet the influence of CMI seasonality and timescale were less distinct than for juniper (Fig. 4b, c). The proportion of trees exhibiting a significant RWI–CMI association differed significantly among the three forest types (Kruskal–Wallis, X2=8.77, P=0.01), and among sites there was a significant negative relationship between r¯RWI-CMI and CMIgy‾ (r2=0.53, P< 0.05), with r¯RWI-CMI decreasing from 0.57 ± 0.14 to -0.02 ± 0.07 as conditions became wetter (Fig. 5). There was a very strong linear relationship between r¯RWI-CMI and FRWI-CMI across sites (r2=0.92, P< 0.001). Significant proportions of western juniper and, to a lesser extent, ponderosa pine also showed negative correlations between growth and VPD, Tmax, climatic water deficit, as well as positive correlations with SPEI, while grand fir were largely unresponsive to any of the climate variables that we examined (Fig. S1; Table S6).

Each species showed relatively high growth around 1980 to 1984, followed by a general decline from c. 1985 to 1994 (Fig. 2, right panel) that corresponded with a period of relatively sustained, below-average CMIgy (Fig. 2, left panel) and, as discussed below, a widespread western spruce budworm (Choristoneura freemani) outbreak. The growth decline was particularly severe among fir; average radial growth in 1993 (0.55 ± 0.04 mm) and 1994 (1.00 ± 0.09 mm) was 73 and 52 % lower, respectively, than in 1985 (2.07 ± 0.14 mm). In contrast, pine growth was 38 % lower (1.23 ± 0.09 mm vs. 1.83 ± 0.12 mm) and juniper growth 42 % lower (0.38 ± 0.03 mm vs. 0.66 ± 0.03 mm) in 1994 than in 1985. Since disturbances can obscure RWI–climate relationship, we reevaluated the correlations using measurements from 1994 to 2013 and found that the correlations changed little in comparison to those computed from 1964 to 2013 (Table S6).

Notable changes in morphological traits were evident along the water availability gradient (Fig. 6a, b, c; Tables 3, 4, S1). In general, ecosystem-average stem wood density, WD‾, decreased with increased CMIgy‾ (r2=0.36, Fig. 6a). Correspondingly, species-average stem WD of juniper (0.47 ± 0.02 g cm-3) was 4 % higher than that of pine (0.45 ± 0.02 g cm-3) and 18 % higher than that of fir (0.40 ± 0.03 g cm-3). Maximum tree height increased nonlinearly by 220 % with CMIgy‾ (r2=0.69), from average Hmax of 10.54 ± 0.53 m at the juniper sites to 33.54 ± 5.56 m at the fir-dominated sites (Fig. 6b). Many of the tallest trees at the fir-dominated sites were actually old pine, the tallest of which reached 41.2 m. Forest LA : SA also increased with CMIgy‾ (r2=0.76; Fig. 6c), ranging from 0.02 ± 0.01 to 0.19 ± 0.02 m2 cm-2. Average LA : SA was quite low at the juniper sites (0.04 ± 0.02 m2 cm-2) and 75 and 300 % higher at the pine (0.07 ± 0.02 m2 cm-2) and fir sites (0.16 ± 0.02 m2 cm-2), respectively. Lastly, there were significant relationships among Hmax, WD‾, and LA : SA across sites (r2=0.35–0.55, P< 0.05; Table 3).

Consistent with predictions from our first hypothesis, forest ecosystem LAI, ANPP, and AGB increased with CMIgy‾, yet residual variance in the regression models implies that the characteristics were influenced by additional bioclimatic or disturbance-related factors. Our results, along with those of several prior field studies (Gholz, 1982; Grier and Running, 1977), indicate that water availability is the dominant constraint on forest leaf area of mature stands in the eastern Cascades. The constraint of water availability on forest leaf area in this environment is driven by the need to maximize carbon assimilation while simultaneously minimizing water loss. Net carbon uptake by trees (i.e., photosynthesis–respiration) occurs within the limits of the hydraulic system (Ruehr et al., 2014), minimizing the risk of embolism when evaporative demand exceeds supply (Meinzer et al., 2010). While CMIgy‾ explained 68 % of the observed variance in LAI, differences among sites in soil nitrogen, stand age, and disturbance history (e.g., windthrow, insect outbreaks) might account for some of the residual variance (Gholz, 1982; Law et al., 2003). For instance, we observed low-level leaf herbivory at several of the grand fir sites, possibly due to western spruce budworm, which has affected large areas of forest in this region over the past 3 decades (Meigs et al., 2015; Willamette National Forest, 1995). Defoliation could have temporarily reduced leaf area without killing the trees, drawing down stored carbohydrates during refoliation. Additionally, there is uncertainty in both destructive and nondestructive estimates of forest leaf area (Law et al., 2001; Runyon et al., 1994) that could have further obscured the relationship with long-term climate. For example, summer maximum LAI at the Metolius Mature Pine flux site averaged about 2.5 over the past 8 years, starting around 2.1 in 1997 (Law et al., 2001), and our estimate for 2014 was about 2.0. The substantial increase in forest leaf area from the dry western juniper woodlands to the more mesic grand fir forests highlights the importance of long-term water availability in controlling light-harvesting potential and subsequent carbon accumulation.

Forest productivity is highly dependent on leaf area (Gholz, 1982; Waring, 1983; Waring et al., 2014) and therefore the observed increase in forest ANPP with CMIgy‾ can be largely explained by the concomitant increase in leaf area. Forest leaf area and CMIgy‾ respectively explained 75 % and 88 % of the variation in ANPP across sites. Forest leaf area largely determines the interception of photosynthetically active radiation (IPAR; Runyon et al., 1994), while subsequent utilization of IPAR for photosynthesis depends on physiological constraints imposed by environmental conditions (Landsberg and Waring, 1997; Runyon et al., 1994; Tezara et al., 1999). For instance, Runyon et al. (1994) found at seven sites in Oregon that annual IPAR increased linearly with LAI (r2=0.95), while subsequent ANPP depended on utilized IPAR (r2=0.99), which was calculated by reducing IPAR based on constraints imposed due to low soil moisture, high VPD, and the duration of subfreezing temperatures. Differences in biotic and abiotic constraints among our sites could explain some of the remaining variance in the ANPP–CMIgy relationship (e.g., summer VPD, extent of freezing days each year, soil fertility and ratio of above- to belowground C allocation; Schwarz et al., 2004).

Our estimates of ANPP for ponderosa pine over the past 10 years (125–262 g C m-2 yr-1) were within the range reported by Law et al. (2003) for six mature to old stands (103–473 g C m-2 yr-1) in the region and agreed to within 14 % at the Mature Pine flux tower site (210 ± 23 vs. 242 ± 57 g C m-2 yr-1). Productivity at our four ponderosa pine sites was about 20–60 % higher than at sites examined by Gholz (1982; ≈110 g C m-2 yr-1) and Runyon et al. (1994; ≈ 95 g C m-2 yr-1), whereas ANPP at our western juniper sites (11–21 g C m-2 yr-1) was considerably lower than at sites examined by either Gholz (1982; ≈ 60 g C m-2 yr-1) or Runyon et al. (1994; ≈ 80 g C m-2 yr-1). Spatial differences in soil-water-holding capacity in the rocky volcanic soils (Peterman et al., 2013) could account for some of the discrepancy in productivity, as could differences in stand density or age. Western juniper at our sites were generally quite old and had accumulated considerably higher overall AGB (2–3 kg C m-2 vs. ≈ 0.5 kg C m-2). Overall, forest ANPP increased with long-term average water availability, largely owing to the increased leaf area but potentially also due to reductions in abiotic constraints on carbon assimilation.

Forest AGB reflects the long-term integration of ANPP minus carbon loss (e.g., litterfall, herbivory, tree mortality), and therefore higher biomass observed at wetter sites was likely driven by higher rates of ANPP sustained over centennial timescales. Forest AGB increased with CMIgy‾ (r2=0.67) and achieved a carbon density, when averaged across the grand fir sites, that was 6 % greater than the average AGB estimated for temperate forests worldwide (≈ 13.5 kg C m-2; Houghton et al., 2009). We found that CMIgy‾ explained less of the variance in forest AGB than ANPP (r2=0.88), potentially as a result of differences among sites in tree mortality rates. For instance, there was little sign of recent tree mortality at the western juniper or ponderosa pine sites, yet there were several large, freshly wind-thrown trees at two of the grand fir sites (GF-1 and GF-4) that could somewhat account for lower AGB at these sites (8.8–10.4 kg C m-2) compared with the other two grand fir sites (16.9–21.6 kg C m-2). Additionally, differences among sites in growth impairment and mortality from periodic, historic outbreaks of spruce budworm and mountain pine beetle (Dendroctonus ponderosae; Meigs et al., 2015) could account for some of the unexplained variance in the AGB–CMIgy‾ regression model. Our findings, and those of prior field (Franklin and Dyrness, 1988; Gholz, 1982; Grier and Running, 1977) and modeling (Kang et al., 2014; Runyon et al., 1994) analyses, demonstrate that steep gradients in water availability are a key factor shaping forest distribution and carbon cycling in the Pacific Northwest, with regional forests further molded by disturbances such as timber harvest, wildfire, and insect outbreaks (Law et al., 2004; Law and Waring, 2015; Meigs et al., 2015).

We found evidence to support the hypothesis that interannual fluctuation in water availability exerted less influence over tree growth as conditions became wetter; however, we unexpectedly observed a pronounced growth decline across species from c. 1985 to 1994 that was particularly severe among grand fir and potentially associated with relatively sustained, dry conditions over that decade. We observed that as CMIgy‾ increased, there was a decline in the proportion of trees at each site that showed a significant correlation between radial growth and CMI as well as a decline in the magnitude of that correlation. This suggests that, on an annual basis, water tends to limit the growth of western juniper to a greater extent than that of either ponderosa pine or grand fir, a finding consistent with results from a modeling study that showed that drought imposed greater physiological reductions on the utilization of annual IPAR by western juniper than by ponderosa pine or other conifers at higher elevations in the eastern Cascade Mountains (Runyon et al., 1994). Western juniper trees occupy sites with chronically low rainfall, high summer VPD and ET, and often thin, rocky soils, leading to generally low soil moisture that is highly dependent on annual recharge. Thus, annual photosynthesis and subsequent stem wood production of western juniper track the annual water balance to a greater extent than ponderosa pine or grand fir, which occupy sites with not only higher rainfall but reduced summer VPD and ET. Furthermore, soils in the ponderosa pine and grand fir zones tend to be deeper and have greater water storage capacity than in the western juniper zone (Kern, 1995). Soil depth averaged about 0.5 m at a nearby juniper site (Anthoni et al., 1999) and about 2 m at the Metolius Mature Pine site (Law, unpublished data). Our analysis focused on changes in tree growth–climate relations over tens of kilometers, yet evidence from tree-ring and remote-sensing analyses at regional (Huang et al., 2015; Littell et al., 2008; Wilmking and Juday, 2005), hemispheric (Vicente-Serrano et al., 2014) and global scales (Vicente-Serrano et al., 2013) suggests that as landscapes become wetter, annual tree growth generally becomes increasingly decoupled from interannual fluctuations in water availability. Nevertheless, trees in characteristically wet landscapes can still be vulnerable to drought-induced stress and mortality (Allen et al., 2010).

What factor, or complex of factors, were responsible for the notable cross-species decline in tree radial growth from c. 1985 to 1994, and why was the decline most severe in grand fir? Ponderosa pine at low and high elevations in southern Oregon also experienced reduced growth during these years (Knutson and Pyke, 2008). This 10-year period was characterized by 9 years in which CMIgy was at or below the 50-year average and punctuated by very dry conditions in 1994, the second most severe single-year drought over the record examined. This period corresponded with the most extensive mountain pine beetle and western spruce budworm outbreaks to have occurred in the eastern Cascades since 1970 (Meigs et al., 2015). The inspection of aerial survey data collected annually since the 1940s by federal and state agencies (available online: http://www.fs.usda.gov/detail/r6/forest-grasslandhealth/) revealed that from 1985 to 1994 our grand fir sites showed signs of western spruce budworm defoliation for between 2 and 8 years, depending on the site. Western spruce budworm defoliation was also detected at three of our four ponderosa pine sites in 1986. Our woodland western juniper sites were not surveyed. We speculate that a multi-year drawdown of soil moisture due to insufficient recharge could have triggered the growth declines; which, in the case of grand fir and, to a lesser extent, ponderosa pine, were further exacerbated by insect attack. We suspect that grand fir growth is buffered against single-year meteorological droughts by ample soil water storage and recharge during wet years but is potentially vulnerable to soil water draw down due to sustained, multi-year periods of below average recharge. Soil moisture buffering might explain the lack of significant RWI–CMI correlations among grand fir. Although beyond the scope of this study, additional lines of evidence to support or refute this hypothesis could be drawn from tree-ring isotopes, remote-sensing observations, and detailed hydraulic and ecological modeling. This presents an intriguing possibility that although annual growth tends to become decoupled from interannual fluctuations in water as conditions become wetter, forests occupying generally wet zones could experience greater impact from infrequent, multi-year periods of below-normal soil water recharge due to increased biotic pressure (e.g., insects) and morphological investment in competition for light over water.

Dendroecological analyses often involve sampling trees at multiple sites and then producing a mean ring-width chronology, at the site or regional level, that is used to reconstruct paleoclimatic conditions (e.g., Fritts, 2001) or assess spatial variability in historic forest growth (Berner et al., 2011; Bunn et al., 2013); however, growth and climate sensitivity of neighboring trees can vary widely (Berner et al., 2013; Bunn et al., 2005; Lloyd et al., 2010), and additional ecological information can be gained by examining growth–climate relations on an individualistic basis (Carrer, 2011). Our study demonstrates that not all trees within a site responded uniformly to drought, particularly at the ponderosa pine and grand fir sites. We did not explore the underlying mechanisms responsible for differences in drought response among individual trees within a site; however, factors such as age, genotype, competition, and microclimate could have contributed to these differences (Bunn et al., 2005, 2011; Hultine et al., 2013; Loranty et al., 2010) and warrant further investigation. The substantial intra- and interspecific variation in drought sensitivity that we observed highlights the need for fine spatial and taxonomical resolution when modeling potential climate change impacts on mixed-species forests in mountainous terrain.

As conditions became wetter, forest WD‾ declined, while Hmax and LA : SA increased, suggesting that trees responded to reduced hydraulic stress by increasing investment in competition for light to the extent that xylem water potentials were maintained within an operable range. These morphological changes, together with the observed differences in forest productivity and growth–climate relations across our network of sites, highlight trade-offs associated with investment in stress tolerance versus competitive ability (Grime, 2001; Reich, 2014). Since cavitation resistance (e.g., P50) tends to increase with WD (Chave et al., 2009; Hacke et al., 2001; Jacobsen et al., 2007), the decline in stem WD‾ that we observed between the juniper and fir communities is generally congruent with prior research showing that small western juniper branches are more resistant to drought-induced cavitation (P50 = -9.0 MPa; Willson et al., 2008) than those of ponderosa pine (P50 = -4.5 to -4.8; Domec et al., 2009) or grand fir (P50 = -3.6 to -6.1 MPa; Cochard, 2006; McCulloh et al., 2011) and, furthermore, that ponderosa pine stems are more cavitation-resistant (P50 = -2.5 to -4.1; Domec et al., 2009) than those of grand fir (P50 = -1.1 to -1.0; McCulloh et al., 2011). Several previous syntheses similarly found general declines in WD (Swenson and Enquist, 2007; Wiemann and Williamson, 2002) and increases in P50 (Choat et al., 2012; Maherali et al., 2004) as conditions become wetter, yet these relationships tend to be quite weak since trees have evolved a variety of strategies to maintain hydraulic integrity, which depends on a suite of traits integrated at the organism level (e.g., stomatal regulation, xylem capacitance, embolism repair, LA : SA plasticity; Meinzer et al., 2010; Whitehead et al., 1984). High WD helps enable trees to endure hydraulic and mechanical stress, yet could come at a competitive cost in productive environments since dense wood is carbon-intensive to construct (Chave et al., 2009), generally has low hydraulic conductivity (Hacke et al., 2009; Reich, 2014), and limits the maximum height that a tree can obtain before buckling under its own weight (Swenson and Enquist, 2007).

Canopy height plays an important role in competition for light among plants (Hartmann, 2011; King, 1990) and, in trees, might ultimately be constrained by low leaf water potential inhibiting cell expansion and growth (Koch et al., 2004; Sala and Hoch, 2009) and/or photosynthesis (Koch et al., 2004; Ryan and Yoder, 1997) at the top of the canopy. In the juniper woodlands, low stem density (99–157 trees ha-1) and high water stress would place little premium on height growth, whereas ponderosa pine (127–334 trees ha-1) and grand fir (205–645 trees ha-1) formed progressively denser stands where greater canopy height would be more competitively beneficial and hydraulically feasible, which potentially explains the nearly 3-fold increase in Hmax that occurred among these forest communities. While taller trees can compete more effectively for light, the additional height causes them to experience lower xylem Ψ (Whitehead et al., 1984) and could render these trees more vulnerable to drought impacts if conditions become increasingly arid (Hartmann, 2011; McDowell and Allen, 2015).

As with Hmax, forest LA : SA increased nearly 3-fold between the western juniper and grand fir stands, demonstrating that trees responded to increased water availability by increasing investment in light absorption relative to water transport. In a review of conifer LA : SA, Waring et al. (1982) reported that LA : SA varied from 0.14 to 0.75 m2 cm-2 across 14 conifer species and noted qualitatively that taxa with low LA : SA occurred in desiccating environments (e.g., J. occidentalis), whereas taxa with high LA : SA occupied mild climates (e.g., Abies lasiocarpa). Our results provide quantitative support for this earlier observation. Several studies have additionally shown that pine inhabiting sites with contrasting climatic conditions can adjust LA : SA to maintain xylem Ψ within a narrow range that prevents cavitation (Maherali and DeLucia, 2001; Mencuccini and Grace, 1995). Low LA : SA enables trees to maintain higher leaf-specific hydraulic conductance, which helps regulate leaf Ψ (Maherali and DeLucia, 2001; Mencuccini and Grace, 1995), and is likely an important adaptation allowing western juniper to inhabit such arid sites. While low LA : SA helps ease leaf Ψ, this must be balanced against the increase in stem respiration resulting from greater investment in sapwood, given that stem respiration increases with both sapwood volume and temperature (Ryan et al., 1995). Relative to western juniper, the high LA : SA observed at the grand fir stands likely enables greater light absorption and photosynthate production, yet reduced water transport capabilities that could render these trees more susceptible to significant changes in water availability. Our findings demonstrate several adaptive responses to shifts in water availability, highlighting trade-offs between hydraulic stress tolerance and competition for light and suggesting potential changes in forest community morphological characteristics that could result from future changes in water availability due to sustained regional warming. Investigating the extent and causes of variation in traits across species and regions is a critical step towards understanding and modeling ecosystem properties and their responses to environmental change (Anderegg, 2015; Law, 2014; Westoby and Wright, 2006).

We note several limitations associated with using an observational study to elicit the response of plant form and function to variation in climate, as well as with using CMI as an indicator of water availability. The observational approach makes it very challenging to separate the effect of a single environmental factor on plant form or function given that it is not possible to control the suite of environmental factors that can influence plant processes. For instance, although we focused on CMI as a dominant driver of carbon cycling and trait characteristics, there were additional differences among sites in climate (e.g., VPD, frost frequency, snow pack), soil, and disturbance history that independently, or interactively, could affect the processes of interest. Observational studies can elucidate existing spatial and temporal variation in plant response to climate or other controlling factors and are strengthened when coupled with experimental and modeling studies focused on the underlying mechanisms of response.

We used the CMI as an indicator of plant water availability and acknowledge that while useful, it is a rather simple index with several shortcomings. The index is easy to calculate from gridded climate data and serves as the water balance calculation underpinning the SPEI (Vicente-Serrano et al., 2010); together, they have been used to examine hydroclimatic controls over forest distribution (Hogg, 1997), productivity (Berner et al., 2013; Vicente-Serrano et al., 2013, 2014), and wildfire (Williams et al., 2014). The CMI is sensitive to changes in temperature and integrates both atmospheric inputs and withdrawals; however, it does not account for spatial variation in soil depth, soil water holding capacity, or snowpack that, along with rooting depth and architecture, further determine plant water availability. Additionally, we estimated CMI at each site using gridded PRISM climate data and an empirical equation for calculating ET0 from T, PPT, and R. Different climate data sets, or means of estimating ET0, could affect the statistical relationships that we observed, yet it is the overall tendencies, rather than the absolute statistical parameters, that we are most interested in capturing.

Forests ecosystems in the Pacific Northwest are a critical element of the regional economy, culture, biodiversity and biophysics, yet future changes in water availability due to sustained regional warming could have significant adverse impacts on these unique and valuable ecosystems. Uncertainty in forest physiologic, demographic and disturbance-related responses to changing environmental settings hinder projections of future forest conditions (Fisher et al., 2010; Hudiburg et al., 2013a; Law, 2014; van der Molen et al., 2011) and thus our adaptation and mitigation capabilities. For instance, CO2 fertilization might ameliorate some impacts of reduced water availability by increasing plant water-use efficiency (WUE), yet the potential magnitude of this response is unclear. Ecological simulations suggest that warming and increased WUE could lead to a ∼ 20 % increase in plant net primary productivity (NPP) by 2100 (Hudiburg et al., 2013b; Kang et al., 2014) or, conversely, that conifer forest NPP, carbon storage, and extent could decline substantially over this period due to reduced water availability and increased heat stress, despite increased WUE (Jiang et al., 2013). These divergent projections underscore the need to further reduce uncertainty in ecological models, which will necessitate investment in high-performance computing, experiments to elucidate physiologic mechanisms driving intra- and interspecific differences in drought and heat sensitivity, regional field studies (such as ours) to provide physiological measurements for model parameterization, and a combination of remote-sensing and eddy covariance flux tower measurements to hone and validate model predictions. By understanding current climatic controls over forest ecosystem function and refining prognostic models of ecosystem processes, we can better anticipate the potential impacts of climate change and thus improve our mitigation and adaptation capabilities.