Cross-scale interactions affect tree growth and intrinsic water use efficiency and highlight the importance of spatial context in managing forests under global change

Abstract

1. We investigated the potential of cross-scale interactions to affect the outcome of density reduction in a large-scale silvicultural experiment to better understand options for managing forests under climate change.

2. We measured tree growth and intrinsic water-use efficiency (iWUE) based on stable carbon isotopes δ¹³C) to investigate impacts of density reduction across a range of progressively finer spatial scales: site, stand, hillslope position, and neighborhood. In particular, we focused on the influence of treatments beyond the boundaries of treated stands to include impacts on downslope and neighboring stands across sites varying in soil moisture.

3. Trees at the wet site responded with increased growth when compared with trees at the dry site. Additionally, trees in treated stands at the dry site responded with increased iWUE while trees at the wet site showed no difference in iWUE compared to untreated stands.

4. We hypothesized that water is not the primary limiting factor for growth at our sites, but that density reduction released other resources, such as growing space or nutrients to drive the growth response. At progressively finer spatial scales we found that tree responses were not driven by hillslope location (i.e., downslope of treatment) but to changes in local neighborhood tree density.

5. Synthesis. This study demonstrated that water can be viewed as an agent to investigate cross-scale interactions as it links processes operating at coarse to finer spatial scales and vice versa. Consequently, management prescriptions such as density reductions to increase resistance and resilience of trees to climate change, specifically to drought, need to consider cross-scale interactions as specific magnitude and mechanisms of growth responses can only be predicted when multiple scales are taken into account.

Introduction

Global climate change, especially as reflected in increased drought events, is projected to affect ecosystems and valuable services at multiple scales and processes (Garcia et al. 2014). A main challenge in planning for global change is to identify the scales over which ecosystem processes will be impacted. Global-scale processes such as CO₂ fertilization or temperature increase affect water-use efficiency and biomass-allocation patterns (Keenan et al. 2013, Reich et al. 2014). Regional processes such as drought and species migration affect plant growth, mortality and disturbance patterns (Raffa et al. 2008, McDowell et al. 2011). Local processes such as competition or physical damage affect plant growth and mortality patterns (D'Amato and Puettmann 2004, Holzwarth et al. 2013). Just as, or even more important, factors at different scales interact, e.g., global and regional impacts can be moderated through decoupling of the environment at finer, local scales. For example, regional warming can be moderated by local topographic features (Daly et al. 2010) or site quality (Lévesque et al. 2014), and plant interactions (De Frenne et al. 2013, Spasojevic et al. 2014).

Separately assessing these impacts and the spatial scales over which they operate may lead to mismatches of potential management interventions with processes that affect ecosystem services such as water yield and carbon sequestration (Peters et al. 2004, Morris et al. 2014). Examining factors at only one spatial scale in isolation can lead researchers and managers to miss important options or potential synergies. It also ignores cross-scale interactions, where processes at one scale, like regional climate, interact with processes at another scale, such as plant competition (Peters et al. 2007). Cross-scale interactions can be especially important when considering climate change as they give rise to nonlinear or threshold responses that can overwhelm the effects of processes at other scales (Peters et al. 2004, Raffa et al. 2008).

Plant density strongly influences the intensity and importance of processes that act at fine spatial scales, such as competition. Consequently, natural resource managers have commonly managed density by regulating plant spacing through seeding and planting or reducing postestablishment plant density by thinning operations (Puettmann et al. 2009). By managing competition, thinning has been used to achieve a variety of objectives such as to increase tree growth and vigor (Waring and Pitman 1985, Breda et al. 1995), understorey species richness and adaptive capacity (Neill and Puettmann 2013), diversifying overstorey structure (Bauhus et al. 2009), or reducing canopy fuel loads (Allen et al. 2002). The primary effect of density reduction in forests is to increase the amount of resources (water, light, nutrients, growing space) available to residual vegetation (Bréda et al. 1995, Boyden, et al. 2012). In response to increased resource availability trees will adjust crown architecture, and leaf, sapwood and root areas to capture these resources (Aussenac 2000). Tree responses not only depend on the amount and spatial heterogeneity of residual available resources right after treatment, but also vary over time. Density reduction through thinning has been proposed in temperate and boreal forests as an option to reduce the effects of climate change on forests, specifically drought, through its impact of increasing available resources, especially water (Kohler et al. 2010, Chmura et al. 2011, D'Amato et al. 2013, Park et al. 2014).

Douglas-fir forests managed for commodity production and other ecosystem services provide an ideal opportunity to study the influence of density reduction at multiple spatial scales in the context of adaptation to climate change. Douglas-fir has a broad natural distribution and is widely planted in many different ecosystems, including outside its natural range, in a correspondingly wide range of climate conditions (Chen et al. 2010, Chmura et al. 2011, Lévesque et al. 2013, Lavender et al. 2014). Additionally, in forests where Douglas-fir is a dominant species, an abundance of research exists on potential management options in uncertain future climates, for a review; see Chmura et al. (2011).

Retrospective studies on effects of density reduction, other disturbances, and climate variability on trees often use dendrochronology. Additionally, isotopic analysis of tree-ring cellulose allows more detailed investigations of climate and growing condition influences on plant physiological performance over time. Specifically, stable carbon isotope analysis δ¹³C) from tree rings, combined with dendrochronology, has become a common tool used to retrospectively investigate impacts of climate and forest management on the intrinsic water-use efficiency (iWUE) of trees (McDowell et al. 2006, Brooks and Coulombe 2009, Brooks and Mitchell 2011). Using δ¹³C to calculate iWUE facilitates comparisons of the physiological responses to changing resources over time among trees growing at different sites (McCarroll and Loader 2004, Lévesque et al. 2013). In general, changes in iWUE indicate a shift in the physiological balance between photosynthesis and stomatal conductance, and are often caused by changes in moisture availability. Thus, isotopic records in tree ring chronologies indicate physiological responses to moisture conditions at the time the carbon was fixed (McCarroll and Loader 2004).

We initiated the current study to investigate multi-scale processes to understand the mechanisms responsible for tree responses to density reductions. In a previous study of Douglas-fir in western Oregon, Ruzicka et al. (2014) identified that trees within and downslope of treated stands showed higher post-treatment growth than trees in untreated stands and downslope of gaps. In this study, we follow up using changes in growth and iWUE to examine if cross-scale interactions provide insight into management options that reduce the impact of climate change on forests. Our analysis is based on a hierarchy of scales, where we first examine coarse-scale patterns, i.e., basic information about growth differences across three sites, which varied in local climate. We then examined progressively finer-scale processes to investigate potential crossscale interactions, starting with a comparison of tree growth in treated and untreated stands within sites. Lastly, to understand responses at the stand-scale, we examined finer-grained interactions from hillslope location to the local neighborhood scale.

Methods

Our study utilized sites established as part of the Density Management and Riparian Buffer Study of western Oregon, USA (DMS; Cissel et al. 2006, Ruzicka et al. 2014) (Figure 1). We chose three DMS study sites (Keel Mountain, North Soup, OM Hubbard) to represent a climate gradient (wet, intermediate, dry, respectively; Table 1). All three sites experience a Mediterranean climate with cool wet winters and warm dry summers. Average precipitation ranged from 1417 to 1968 mm per year, mostly occurring from November through April. Keel Mountain was the wettest site experiencing, on average, the most precipitation and lowest growing season vapor pressure deficit (VPD) while OM Hubbard was the driest with the lowest yearly rainfall and highest VPD. The three sites in this study were composed of several stands, over 200 ha in area, and approximately 200 km apart from each other. Forests on all sites were dominated by conifer species, primarily Douglas-fir (Pseudotsuga menziesii (Mirb.) Franco) with a smaller component of western hemlock (Tsuga heterophylla (Raf.) Sarg.) and western redcedar (Thuja plicata Donn ex D. Don), which were more prevalent at Keel Mountain in the western Cascade Range (Figure 1). Soils were primarily humic ultisols and inceptisols with high infiltration rates typical of western Oregon. The soils at OM Hubbard contain a higher clay content than the other two sites. Cissel et al. (2006) contains a complete DMS overview including detailed site histories and descriptions.

Figure 1:

location map

Table 1.

Characteristics and treatment conditions for three western Oregon, USA study sites.

	Keel Mountain (wet)	North Soup (intermediate)	OM Hubbard (dry)
Latitude	44°31’41.0”	43°33’57.0”	43°17’30.0”
Longitude	122°37’55.0”	123°46’38.0”	123°35’00.0”
Elevation (m)	654–756	176–411	436–783
1Mean annual precip. (mm)	1968	1735	1417
1Mean growing season vapor pressure deficit (kPA)	1.23	1.5	1.64
Treatment start	Dec-1997	Aug-1998	Sep-1997
Stand age at treatment	44	48	39

¹ Weather data from PRISM Climate Group http://www.prism.oregonstate.edu/explorer/ for the study time period.

The upland treatments (density reduction through thinning and gap creation) covered large areas (20-49 ha), creating the different stands used in this study. OM Hubbard and Keel Mountain were treated in 1997 while North Soup was treated in 1998. We examined two different tree densities within the variable density treatment area: 1) areas thinned to a residual density of 100 trees per hectare (tph; thinned) with adjacent riparian stands located downslope of thinned areas (downslope thinned); and 2) 0.4-ha circular gaps in which all trees were harvested in the gap with adjacent riparian stands below the gaps (downslope gap). An unthinned stand was also included as a reference or control stand (unthinned). All thinning preferentially removed smaller trees, except for minority conifer species and hardwoods, which were retained for structural and species diversity.

At all sites, we collected data from overstorey trees in plots that were part of 13 pre-established trans-riparian transects (Cissel et al. 2006, Anderson 2007), aligned perpendicular to headwater streams. Riparian buffer widths ranged from 16 m to 32 m as measured from the middle of the stream channel. The buffer was designed with a 15 m minimum width, but could be wider to accommodate local conditions of riparian vegetation and topography. The first and second plot centers were 4.5 m and 14 m from the stream center, respectively, and typically were located in the riparian buffer (downslope thinned, downslope gap and unthinned stands). The third and fourth plots, (thinned and unthinned stands) were 22.7 m and 41 m upslope of the stream center, respectively. From June to August 2011, we selected and recorded locations of the three co-dominant and apparently healthy Douglas-fir closest to plot centers. The numbers of sample trees by treatment were 9 (three per site) in thinned stands, 14 (6 wet 3 intermediate 5 dry) in downslope of thinned 21 (6 wet, 7 intermediate, 8 dry) in downslope of gaps, and 32 (12 wet, 13 intermediate, 7 dry) in unthinned stands. Unequal sample sizes resulted because we excluded some trees due to an inability to cross-date cores or other sampling problems. The majority of excluded trees were downslope gap and thinned stands at the dry (6) and intermediate (14) sites. For four transects, the ridge top was less than 41m from the stream center, limiting measurements to three plots. In addition, one plot in the control stand at Keel Mountain did not have any Douglas-fir, resulting in a total of 36 plots. Our final sample size was 76 trees over three sites.

For each tree, we measured diameter at breast height and distance from treatment edge (downslope stands only), which was defined as the closest cut stump. To estimate the change in local competition, we counted the number of trees and stumps within 11.5 m of each tree. We measured slope distance, not horizontal distance, to capture the physical length of soil between trees and the treatment edge. Trees in riparian buffers were between 2 m and 30 m downslope of stand edges.

We collected one 7 mm increment core to the pith and three 12 mm cores to capture at least 30 annual rings, one from each cardinal direction. Each core was sanded until rings and early-latewood boundaries were clearly visible. We measured ring width using an optical scanner with 2400 DPI resolution and loaded into WinDendro to measure early, late and annual ring width. Tree rings were cross-dated visually using pointer years and statistically using the cross.date function in the dplR program library (Bunn 2010). After cross-dating to ensure intra-and inter-tree accuracy, we averaged the four cores from each tree together by year into a raw ring width chronology for each tree. We converted raw tree ring widths to basal area increment (BAI) for earlywood, latewood, and annual ring growth (Phipps 2005) after subtracting the width of the bark from the final tree diameter (Larsen and Hann 1985). For both BAI and isotope analysis we sampled rings from 5 years pre-treatment to 12 years post-treatment for each tree in our analysis.

For δ¹³C measurements, we separated early and latewood from the three 12-mm cores from each tree along ring boundaries established in the dating process using a grinding bit in a handheld rotary tool, which transformed the wood tissue to a fine powder. We combined the powdered samples from the replicate cores for each tree by year and then extracted to α–cellulose (Leavitt and Danzer 1993). The stable isotope lab at Oregon State University College of Earth, Oceanic, Atmospheric Sciences in Corvallis, OR determined the values of δ¹³C in tree ring cellulose. Samples were combusted to CO₂ in a Carlo Erba NA1500 elemental analyzer then introduced into a DeltaPlusXL isotope ratio mass spectrometer. Measurement precision was better than 0.1 ‰ as determined by repeated measures from sample replicates. We could not analyze some samples due to insufficient sample size to accurately measure for isotope composition or sample corruption. Sample corruption occurred on a few trees with very narrow growth rings that were ground using a ball mill where iron filings were accidentally introduced to the sample. The final sample size was 1347.

We reported δ¹³C values relative to the Pee Dee Belemnite standard (PDB) where the ratio of ¹³C to ¹²C atoms in a sample (R_sample was compared with the ratio of the standard (R_standard) in equation 1:

$${\delta }^{13}C=\left(\frac{R_{\text{sample}}}{R_{\text{standard}}}-1\right)$$ (1)

We analyzed only latewood for δ¹³C because latewood provides a more robust indicator of the growing conditions for a particular year and because it provides a record of the growing conditions for the water-limited season in a Mediterranean climate. Earlywood potentially retains the isotopic signature of photosynthesis carried out in the previous growing season and winter, and from other non-structural carbohydrates stored for longer periods (McCarroll and Loader 2004).

Isotope theory

Intrinsic water-use efficiency is defined as the ratio of photosynthesis and stomatal conductance (A/g_s) and can be estimated using δ¹³C. First, we calculated discrimination (Δ¹³C) against the heavier isotope to remove year-to-year variation of δ¹³C in the atmosphere using equating 2: (Farquhar et al. 1982)

$${\Delta }^{13}\text{C}=\frac{{\delta }^{13}{\text{C}}_{\text{air}}-{\delta }^{13}{\text{C}}_{\text{cell}}}{1+{\delta }^{13}{\text{C}}_{\text{cell}}}$$ (2)

where δ¹³C_air and δ¹³C_cell are the δ¹³C values of the atmosphere and cellulose samples. We obtained [CO₂] and δ¹³C_air concentrations from the National Oceanic and Atmospheric Administration Cooperative air sampling network, Midway Island station for the years 1991–2011 (http://www.esrl.noaa.gov/gmd/ccgg/flask.php). To account for seasonal changes in the concentration of CO₂ and δ¹³C we converted monthly data to seasonal averages corresponding to the seasonal climate averages associated with early and late periods of carbon fixation (April-June and late July-Sept; Barnard et al. 2012, Beedlow et al. 2007). Δ¹³C is regulated by the ratio of internal [CO₂] to atmospheric [CO₂] (C_i/c_a) as described by equation 3 (Farquhar et al. 1989):

$${\Delta }^{13}\text{C=}a+\left(b-a\right)(\frac{c_i}{c_a})$$ (3)

where a is the fractionation from diffusion through the stomata (4.4 ‰) and b is the fractionation by rubisco (~27 ‰). iWUE is then estimated using the atmospheric concentration of CO₂ (c_a) and the difference in diffusivity of CO₂ and water in air (1.6) in equation 4: (Farquhar et al. 1989)

$$\text{iWUE=}\frac{\text{A}}{g_s}=\frac{(c_a-c_i)}{1.6}$$ (4)

Isotopically derived values of iWUE from tree rings facilitate comparisons of long-term trends because the carbon in latewood is integrated over the period of latewood formation, averaging iWUE over the late growing season (McCarroll and Loader 2004). Year-to-year values of iWUE can then be compared to indicate different physiological responses to water availability or changing growing conditions. Implicit in these models are assumptions that mesophyll conductance remains constant and non-limiting in leaves throughout the study (Seibt et al. 2008), and any fractionation events in the trunk or in phloem are consistent (Offermann et al. 2011) in terms of their effects on δ¹³C in tree rings. For most studies, these assumptions appear to be valid (Cernusak et al. 2013, Cernusak and English 2015). Additionally, climatological studies have identified that a juvenile effect may influence the δ¹³C in tree rings in young trees due to respiration of carbon and/or tree height (McCarroll and Loader 2004). However the respired CO₂ influence is confined to young seedling within 1m of the forest floor (Buchmann et al. 2002), and Monserud and Marshall (2001) found no influence of height on tree ring carbon isotope discrimination for Douglas-fir. Trees in our study were taller than 25 m at the onset of the study so beyond any juvenile effect. Additionally, our study design comparing treated and control stands from the same cohort would account for any age influences over time.

Data analysis

We used a hierarchal framework for data analysis at multiple scales. The results of analysis at coarse scales were used to inform the analysis at finer scales. Data analysis in this framework allowed us to explore variation found at coarse scales in the search of a more parsimonious explanation for patterns observed at each scale.

The first and largest scale analyzed was at the site level which was done to assess site differences in tree growth and intrinsic water-use efficiency. Trees growing in the unthinned riparian stands were used for this analysis to avoid any treatment effects (n=32). Two-way repeated measures ANOVA was used to investigate inherent site differences in five metrics: annual ring growth, or basal area increment (BAI), earlywood and latewood BAI, latewood iWUE and ratio of earlywood to annual ring width (earlywood proportion). A random error structure was included to account for our nested design (plots within transects) and an autoregressive covariance structure to model the autocorrelation present due to repeated measures by year. The five models (annual BAI, earlywood BAI, latewood BAI, latewood iWUE and earlywood proportion) were fitted using the lme function from the nlme package (Pinheiro et al. 2011) and computed in R 3.1.1 (R Core Team 2012). The function varIdent was used to allow for different variances in sample size between sites (Zuur et al. 2009). The Bonferroni method was used for multiple comparisons to obtain correct p-values when computing contrasts.

The second scale analyzed for this study was at the stand level. Multiple stands were created at each site using the treatments described in the methods (thinned, downslope thinned, downslope gap, unthinned). A two-way repeated measures ANOVA was used to investigate stand differences among sites after treatment in four metrics: annual ring growth (BAI), earlywood BAI, latewood BAI, latewood iWUE. The interaction between stand and site was included in the model. The four (annual ring growth (BAI), earlywood BAI, latewood BAI, latewood iWUE) models were fitted using the lme function from the nlme package (Pinheiro et al. 2011) and computed in R 3.1.1. The models were fit using the same error structures as the site level variables. Contrasts between stands within each site and between sites were calculated using the testInteractions function from the phia package (De Rosario-Martinez and Fox 2013). The Bonferroni method was used for multiple comparisons to obtain correct p-values when computing site contrasts.

Analysis at the stand scale highlighted the difference between stands among sites, but also suggested that finer scale processes might better explain differences at the stand scale. A previous study at our sites (Ruzicka 2014) described effects from upslope thinning treatments extended about 15 m from the edge into the stands downslope of thinning. We focused the next investigations on the stands downslope of thinning and gaps to test for possible cross-scale interactions that influence stand-level responses. The finer scales analyzed for this study used only trees (n=35) downslope of thinned and gap stands to investigate differences between these stands.

Trees in the thinned and downslope thinned and gap stands were divided into groups; trees that grew within 15 m from the edge of the treated stand, and those that did not. A type II ANOVA was used to compare differences in post-pretreatment normalized growth between the two groups. A zero value in the difference would indicate no change in growth after treatment. An interaction between hillslope location (stand edge and not edge) and site was included to examine if any difference between hillslope locations was based on the sites. The model was fitted using the aov function in the base R 3.1.1.

The interaction of the factors acting at hillslope and local neighborhood scales was investigated to determine why some trees in downslope of thinned and gap stands responded with increased growth and others did not. The annual ring BAI growth difference (post minus pre) for trees in the downslope thinned and gap stands were analyzed using a mixed effect model with site and the proportion of trees removed during treatment as well as the interaction thereof. The proportion of trees cut from the local neighborhood in the thinning process was arcsine square root transformed to improve model fits when using a proportion. Similar to the site and stand level, the model was fit using the lme and varIdent functions in R.

The final investigation was to determine if the response of trees at the local scale was related to changes in the tree’s intrinsic water-use efficiency. The trees were divided into two groups; trees within 15m that responded with increased growth to density reduction from treatments and those that did not. A type II ANOVA was used to compare the difference (postpretreatment) in latewood iWUE between the two groups. An interaction was included to examine if any difference was based on the site. The model was fit using the aov function in R 3.1.1 (R Core Team 2012). The assumptions of normality and homogeneity of variances was tested graphically for all ANOVA tests at the hillslope location and local scale. Additionally, p-values from tests done at the hillslope location and local scale were corrected using the false discovery rate to account for multiple tests (Verhoeven et al. 2005).

Results

Site-scale differences (e.g., precipitation, growing season vapor pressure deficit: Table 1) were reflected in growth and intrinsic water-use efficiency of unthinned trees (Table 2, Figure 2). Annual ring BAI and latewood BAI was lower at drier sites in the unthinned riparian stands. Trees growing at the wet site (Keel Mountain) had the highest average annual growth, while trees at the dry site (OM Hubbard) had the lowest (Table 2). Earlywood BAI growth was not significantly different among the three sites (Table 2). In contrast, trees growing at the dry site had significantly less latewood BAI growth with almost 60 percent of diameter growth completed before the seasonal summer drought lead to a shift to latewood production (Brix 1972) (Table 2). Tree iWUE was also higher at the dry and intermediate sites but only the intermediate site was significantly higher than the wet site (Table 2). Latewood iWUE at the dry site was not significantly different from the other two sites as a result of two trees at that site having the lowest average intrinsic water-use efficiency from all trees in this study (Figure 2). We did not find any special circumstances that may have influenced these trees. However, when these two trees were removed from analysis trees the dry site had significantly higher iWUE than trees on the wet site but not significantly different from intermediate site trees (data not shown). No trees were removed for the analysis presented in Figure and Table 2.

Table 2.

Site differences in unthinned Douglas-fir BAI and intrinsic water-use efficiency. Type III Repeated Measures ANOVA was used to determine differences between sites for all five comparisons. Mean annual values for each site (standard error) are indicated in each column. Significant contrasts among the sites are indicated by stars (*p<0.1, **p<0.05) after Bonferroni correction while values with the same letters are not different.

Site	BAI (total ring) (mm2)	Earlywood BAI growth (mm2)	Latewood BAI growth (mm2)	Proportion of earlywood	Latewood iWUE (μmol mol-1)
Wet	3759a* (457)	1821a (284)	1916a** (220)	0.47a** (0.02)	103a** (2)
Intermediate	2887ab (645)	1397a (398)	1554ab (313)	0.46a** (0.03)	112b** (2)
Dry	2378b* (739)	1445a (466)	904b** (354)	0.60b** (0.03)	107ab (3)

Figure 2.

Latewood iWUE and the proportion of earlywood (as part of total ring BAI) by year for Douglas-fir trees growing in untreated stands of three western Oregon, USA study sites.

Stand-level differences indicated a cross-scale interaction with the site scale as trees responded differently to stand-level treatments at each site (Figure 3). Annual BAI increased significantly more after thinning at the wet site compared to trees in the thinned stand in the dry site (Table 3). Similar patterns were found when examining earlywood and latewood BAI individually (data not shown). Trees in stands downslope of gaps showed an increase in growth after treatment at the intermediate site compared to trees at the wet and dry stands (Table 3). The variation in the magnitude of thinned-stand responses as well as the site differences in stands downslope of treatments, suggested that cross-scale interactions at finer spatial scales likely affected tree growth.

Figure 3:

Stand level annual BAI growth by year for Douglas-fir trees within three western Oregon, USA study sites. Growth was normalized by post minus pre-treatment to account for inherent differences in tree growth (see methods). Error bars represent standard errors. A value of zero for growth would indicate no difference from the pretreatment average. Year zero represents the year that density reduction treatments were applied.

Table 3.

Mixed model Type III Repeated measures ANOVA results for the difference across sites in annual BAI for Douglas-fir stands among three western Oregon, USA study sites. Mean values (standard error) indicate the average change in BAI after treatment for trees within stands. Values were normalized using the pretreatment means; thus a zero value would indicate no change in growth after treatment. Significant contrast differences are indicated by stars (*p<0.1, **p<0.05) after Bonferroni correction. Significant differences between stands with in a site are indicated by different capital letters (A,B, column comparisons), while significant differences among sites are indicated by different lower-case letters (a,b, row comparisons).

Stand X Site	Wet BAI (mm2)	Intermediate BAI (mm2)	Dry BAI (mm2)
Thinned	2871.6aB** (364.5)	1933.7abB* (183.7)	787.6b**A (166.10)
Downslope thinned	798.5aA (166.8)	971.6aA (110.7)	1382.8aB* (122.82)
Downslope gap	104.9aA (81.6)	626.3aA (152.4)	−181.4aA (50.7)
Unthinned	52.4aA (73.1)	194.2aA (87.2)	55.1aA (58.2)

Trees in the thinned stand at the dry site increased in iWUE after thinning (Figure 4), which is opposite of expectations if water resources available to individual trees had increased after thinning. Trees downslope of the thinned stands at the intermediate site also had a significantly higher iWUE than trees at the same hillslope locations on the wet site but not the dry site (Table 4). An increase in iWUE along with an increase in growth in response to thinning indicated that the photosynthetic rate increased relative to stomatal conductance (increased A/g_s). Remarkably, the iWUE for trees on the wet site for all treatment stands was unchanged after density reduction. Together these two lines of evidence suggested that increased water from density reduction was not driving the growth increase of trees in this study. iWUE would likely have decreased after density reduction if an increase in water availability led to the increased growth, i.e., stomatal conductance would have increased more relative to the rate of photosynthesis (decreased A/gs) indicating that water was less limiting to tree growth. There is no evidence to suggest that treatment differences in iWUE were due to changes in local climate as the sites varied in sync and the relative climate differences between sites stayed the same.

Figure 4:

Stand-level latewood iWUE by year for Douglas-fir trees within three western Oregon, USA study sites. iWUE was normalized by post minus pre-treatment to account for inherent differences in tree physiology (see methods). Error bars represent standard errors. A value of zero for iWUE would indicate no difference from the pretreatment average. Year zero represents the year density reduction treatments were applied.

Table 4.

Mixed model Type III Repeated Measures contrasts indicating within treatment difference in latewood iWUE for Douglas-fir stands among three western Oregon, USA study sites. Mean values (standard error) indicate the change in latewood iWUE after treatment for trees within stands. Values were normalized using the pretreatment means; thus a value of zero would indicate no change in iWUE after treatment. Significant contrast differences are indicated by stars (**p<0.05, *p<0.1) after Bonferroni correction. Significant differences between stands with in a site are indicated by different capital letters (A,B, column comparisons), while significant differences among sites are indicated by different lower-case letters (a,b, row comparisons).

Stand X Site	Wet iWUE μmol mol−1	Intermediate iWUE μmol mol−1	Dry iWUE μmol mol−1
Thinned	4.54aA (1.11)	6.99aA (0.98)	22.27b**B** (1.22)
Downslope thinned	5.93aA (0.65)	14.01b*A (0.95)	11.52aA (1.23)
Downslope gap	7.49aA (0.83)	11.98aA (0.63)	8.54aA (0.67)
Unthinned	6.28aA (0.48)	8.29aA (0.46)	10.26aA (0.69)

For trees downslope of the thinning treatments, only trees within 15 m of the edge increased in BAI in response to upland density reduction. However, our experimental design was not set up to test if the edge effect was related to distance from the edge or if the responses were only related to changing local density i.e., there was no density reduction in riparian areas so the density didn’t change beyond the treatment areas. When averaged across sites, a significant edge effect was evident as trees within 15 m of density reduction treatments responded with increased BAI (p=0.027). However, when compared between sites, edge growth response was marginally greater but not statistically different in edge trees at the wet and intermediate sites compared to the dry site where the edge response was more variable. It is likely that the variability between and within stands indicates that resources were not released uniformly across a stand after density reduction. Higher variability observed in the edge response at the hillslope location scale, especially at the dry site, suggested that interactions with finer spatial scales can explain the apparent edge effect (Supplement Figure 2).

At the local scale, trees in the downslope thinned and gap stands responded with increased BAI to greater density reduction in the local neighborhoods (p=0.004). The effect was consistent at all three sites with no significant difference or interaction between them (Figure 5). In general, a similar number of trees were removed from each site (trees per hectare target thinning) but the mean tree size was smaller at the time of thinning at the dry site (average DBH was 27 cm, 25.4 cm, and 19.4 cm at the wet, intermediate, and dry site, respectively).

Figure 5:

Difference in post minus pre-treatment entire ring BAI for individual trees growing downslope of the thinned and gap stands plotted over the proportion of trees removed from local neighborhoods.

The effect of the cross-scale interaction between stand and local neighborhood was reflected by variability in tree physiological response. Trees that increased growth within the 15 m edge showed a highly variable iWUE response to upland forest management. iWUE generally increased in trees downslope of thinned and gap stands that responded at the intermediate and dry sites, with more dramatic increases at the intermediate site and more variability at the dry site although the differences or the site interaction were not significant (data not shown). This result provided further evidence that changes in water availability after density reduction were not the primary driver of the growth response in trees.

Discussion

In summary, by using a hierarchal analysis framework, our results indicated that impacts of global climate change can be influenced by local conditions, as cross-scale interactions control the growth response to density reduction. Density at the local neighborhood scale overrode treatment effects at the larger scales. Conversely, factors acting at the regional scale (site) moderated the magnitude of growth response at the stand scale. This suggests a cross-scale interaction, where patterns observed at larger scales (stand) were mediated by patterns observed at a small scale conditions (local neighborhood, Figure 6).

Figure 6:

Conceptual model for cross-scale interactions examined in our study from the largest scale in the upper left. Site scale differences in growth impact the magnitude of the growth response of different stands to density reduction. Local neighborhood density can override larger scale processes when investigating tree response to density reduction. Arrow width represents the magnitude of the cross-scale interaction for the response of Douglas-fir iWUE or growth. The dry site (orange arrow) responded to thinning with increased water-use efficiency. The wet site (blue arrow) responded to thinning with the greatest growth increase followed by the intermediate site (green arrow) and dry site. Multiple site trajectories of “yes” responses are shown by black arrows.

Our study is an example, how viewing forests as complex adaptive systems can provide insights into ecosystem processes (Messier et al. 2015), e.g., by highlighting and emphasizing cross-scale, hierarchical interactions. Specifically, it provided support for the importance of assessing cross-scale interactions to understand ecological processes, such as tree growth. This is especially important in an era of climate change, when mitigation strategies at the global scale are not likely to be effective, thus placing more emphasis on managing at fine scales, e.g., stand levels or smaller in forestry operations (Puettmann et al. 2009). Our study adds to the growing body of evidence that suggests that this approach has merit, as it shows that impacts of large-scale phenomena, such as climate change, can be mediated by fine-scale processes, such as competition or facilitation (De Frenne et al. 2013, Spasojevic et al. 2014). In this study, we identified that multiple processes interact across scales to affect the response of trees to density reduction. Site-scale differences influenced the magnitude of the growth response of different stands to treatment with a larger growth response in trees at wet sites. However, local neighborhood density was the primary driver behind the larger-scale patterns when investigating whether or not a downslope tree responds with increased growth to upslope density reduction (Figure 6).

Cross-scale interactions may also act at scales beyond those used in our study. Douglas-fir obtains maximum size and competitive advantage with cool wet winters followed by relatively cool dry summers typical of the Pacific Northwest (Waring and Franklin 1979). Due to widespread planting however, Douglas-fir growth can also be compared across very different climate patterns. For example, modeling studies have hypothesized that Douglas-fir in New Zealand grow faster because of lower summer evaporative demand and higher summer moisture availability compared to sites in the Pacific Northwest (Waring et al. 2008). In Europe, planted Douglas-fir grow faster at a mesic, continental site with summer rains than in a Mediterranean climate with a summer moisture deficit (Lévesque et al. 2013). The presence of cross-scale interactions suggest that results from our study, in a region with seasonal moisture patterns of a Mediterranean climate may not apply to these areas where Douglas-fir was introduced or even to other regions where Douglas-fir is found naturally (e.g., the Rocky Mountains). Higher summer moisture availability in many other regions (Waring et al. 2008) would likely modify the patterns found in this study, suggesting caution when extrapolating our results to other regions in the world without accounting for differences in amount and seasonal timing of water availability.

Our study suggested that even within the Pacific Northwest, sensitivity of Douglas-fir growth response to density reduction varies along a fairly narrow range of site conditions compared to the natural (and even more so, the anthropogenic) distribution of the species. In our study, trees on different sites appeared to have established different sensitivities to local climate. Trees growing at the dry site likely will experience more stress and less growth with changes, such as lower precipitation in the winter and spring because the majority of their diameter growth occurred in the spring prior to summer dry periods.

Our results supported general findings that report differences in growth rates and intrinsic water-use efficiency along a climate gradient at regional Pacific Northwest and sub-continental western North American scales (Nakawatase and Peterson 2006, Littell et al. 2008). In general, trees growing at drier sites had a lower growth and higher intrinsic water-use efficiency (higher δ¹³C) when compared with trees on wetter sites (Panek and Waring 1997, Roden et al. 2005). Also, higher intrinsic water-use efficiency in trees on drier sites is relatively consistent when examined at continental scales for Douglas-fir (Chen et al. 2010). In the Pacific Northwest, photosynthesis in the winter and early spring contribute a large amount of carbon to tree reserves (Waring and Franklin 1979) and substantial water can be stored in tree trunks of Douglas-fir for use during the seasonal summer drought (Phillips et al. 2003). Trees may have also responded to extreme summer droughts by closing stomata and thus decoupling growth from regional climate measures (Panek and Waring 1997). Our results highlight that Douglas-fir can be locally adapted to site-level conditions through these and other ecophysiological mechanisms such as leaf morphology or crown architecture (Aussenac 2000). Earlier genetic studies suggested that Douglas-fir was adapted to environmental conditions at even smaller spatial scales, e.g., on north versus south slopes of the same ridge (Hermann and Lavender 1968). We identified cross-scale interactions (Figure 6) that indicate site-scale adaptations should be taken into account when examining changes in growth or intrinsic water-use efficiency at fine spatial scales because these responses can vary depending on regional water regimes (Littell et al. 2008, Chen et al. 2010).

Density reduction typically increases growth rates in residual trees because of an increase in available resources with larger increases at better sites (Aussenac 2000). An increase in growth from density reduction without a change in intrinsic water-use efficiency at the intermediate and wet sites in our study suggested that photosynthesis and stomatal conductance increased in tandem (Farquhar et al. 1989). On these sites, we speculate that the availability of resources that increase photosynthesis (light and nutrients) increased proportionally to resources that increase stomatal conductance (water). In contrast, trees in the treated stand at the dry site showed increased earlywood, latewood and entire ring growth and higher intrinsic water-use efficiency after treatment, indicating that impacts of an increase in available growing space or other resources that would favor photosynthesis was greater than impacts of increased water availability (Brooks and Mitchell 2011). Contrary to expectations, our results indicated that density reduction at dry sites did not result in increased growth due higher water availability in the late summer, rather growth responses appeared to be driven by an increase in other resources such as light or nutrients (McDowell et al. 2006; McDowell et al. 2003). This notion is supported by other studies which showed increased intrinsic water use efficiency in Douglas-fir due to nitrogen addition that likely increased photosynthesis relative to stomatal conductance (Brooks and Coulombe 2009, Brooks and Mitchell 2011). Our results also question whether an increase in absolute growth after density reduction on the dry site has resulted in higher tree vigor, i.e., trees being further away from a tipping point, e.g., drought related mortality (Reyer et al. 2015).

Our findings also show how cross-scale interactions can help interpret confusing or contradictory findings. For example, when diameter growth in stands was averaged across our sites, trees downslope of treatment responded with increased growth, while trees in stands downslope of gaps did not. In our previous work, we had hypothesized that mechanisms for this included an expansion of understory plants in the gaps as they were competing for nutrients, especially nitrogen, with the trees growing at the treatment edge (Ruzicka et al. 2014). A more detailed investigation at finer spatial scales coupled with information about intrinsic water-use efficiency presented in the current study failed to support that hypothesis. Instead, our results suggested that competition in areas with high local neighborhood tree densities may override any potential impacts from uphill treatment or gap creation. In this study, the hillslope scale (distance from treatment edge) was likely a proxy for competitive processes operating at the local neighborhood scale. Other studies have also found that processes such as habitat availability operating at forest edges are dependent on the spatial scale examined (Donovan et al. 1997), as well as at cross-scale interactions which can lead to the apparent variability of responses when investigating edge effects (Ries et al. 2004). The importance of small scale variability has been shown in previous studies at DMS sites, as gap creation increased the amount and variability of nitrogen at the edges of gaps (Thiel and Perakis 2009). This increased small scale variability in nutrient availability has been hypothesized as a factor for understory responses that depended on small-scale processes operating at gap edges (Fahey and Puettmann 2008). Our study reported variability not just acting at fine scales (e.g., hillslope location or neighborhoods), but that variability was influenced by larger scale factors, as it differed among sites and trees on the dry site had a more variable response than trees on the wet site.

Competition and other plant interactions influence plant establishment and growth at small spatial scales (i.e., local plant neighborhoods; D’Amato and Puettmann 2004). Our study suggested a cross-scale interaction between competition operating at the local scale and the growth response downslope of treated and created gap stand edges. Although competition was reduced in the overstorey in stands after density reduction and near the edge, not all edge trees experienced the same change in local conditions. Even if they did, our study suggested that trees would exhibit slightly different patterns of growth and iWUE after treatment depending on crossscale interactions with site. Gap creation and density reduction create spatial heterogeneity in resource availability driven by residual plant structure, substrate and associated plant responses (Boyden et al. 2012). For example, regional-scale patterns in regeneration and understory plant diversity after density reduction varied based on local competitive neighborhoods and cross-scale interactions between overstorey cover and understorey competition (Burton et al. 2014, Dodson et al. 2014). Other studies have also found that response to climate gradients can change based on interactions at different spatial scales or plant processes (Galván et al. 2014). Potential mechanisms for this disparity are rooted in local genetic adaptation (Hermann and Lavender 1968) and changes in resource availability over space and time as plants adjust to different growing conditions (Aussenac 2000).

Conclusions and management implications of cross-scale interactions

Our study identified a hierarchy of cross-scale interactions that reflect the changes in growth and intrinsic water-use efficiency following density reduction. Variability and patterns at each spatial scale investigated in our study has been driven by cross-scale interactions with the other spatial scales examined. There is great value in placing a project area and organisms under management consideration within the context of the larger landscape (Figure 6). It was likely that unexplained variability in tree response to changes in neighborhood density, especially at the different sites was at least partially due to factors measured at very fine scales such as microclimate, soil resource availability, and genetics not investigated in this study.

Management actions to create a robust, climate-smart landscape need to examine the processes affected at multiple spatial and temporal scales. Our study identifies several potential avenues where cross scale interactions can be utilized for management actions. Managers can prioritize stands for treatment based on different management goals and current conditions. Wet sites can be targeted for density reduction if increased growth and vigor of trees is the primary objective. Trees in dry or dense stands may respond at lower levels of residual density compared to wet stands. Management techniques used to create structural heterogeneity will need to account for changes in the local neighborhoods of individual trees to understand stand level responses (Dodson et al. 2012). Local density controls can be used to create a mosaic of different conditions that vary over spatial scales. In addition, tree responses are not constant but change over time. For example, D'Amato et al. (2013) showed the initial beneficial responses reversed over time, i.e., compared to unthinned stands, eventually trees in treated stands became more susceptible to drought, likely due to their increased crown size and leaf area.

Finally, the finding that cross-scale interactions are important in understanding vegetation response suggests that the inference scope for studies of tree response to density reductions need to be placed into a larger spatial context (Raffa et al. 2008). Inference made on studies using single stands or averaging across regions cannot be simply extrapolated to global scales (Heffernan et al. 2014), other regions of the world (Soranno 2014) or easily downscaled to local management options without the context of multiple interacting scales (Peters et al. 2004).