 |
 |
 |
 |
 |
|
 |
|
 |
|
 |
|
 |
|
 |
|
 |
|
 |
|
 |
|
 |
|
 |
|
 |
|
 |
|
Factors Influencing Understory Douglas-fir
Vigor
in Multi-Cohort Prairie Colonization Stands
at Fort Lewis, Washington
|
Derek John Churchill
University of Washington
Abstract
Forest stands on Fort Lewis, Washington are being managed for
multiple values under an uneven-age silvicultural system that
relies on natural regeneration. In stands that were former prairies
and have excessively drained, outwash soils, Douglas-fir (Pseudotsuga
menziesii) is the only understory conifer present and the
principal regeneration species. Factors influencing the vigor
of Douglas-fir advanced regeneration were investigated in thirteen
stands at both the individual tree and stand levels. Live crown
ratio, height-to-diameter ratio, height growth, and crown density
were combined to produce two methods of quantifying vigor: a
regression model that predicts volume growth as a percent of
maximum site potential (relative volume growth) and a simple
vigor classification system. Overstory recruitment potential
and release potential were estimated for different classes of
vigor. At the individual tree level, understory Douglas-fir with
low levels of understory competition was found to require an
average 45% full sunlight or overstory stocking of less than
150 SDI (30% full site occupancy (Long 1985)) to achieve vigor
levels where recruitment into the overstory without further release
is likely. Between 10-35% full sunlight or 150-275 SDI (30-55%
full site occupancy), regeneration was found to be growing slowly
but able to maintain release potential, especially if less than
5m in height. Below 10% full sunlight or above 275 SDI (55% full
site occupancy), regeneration was scarce and of very poor vigor.
Regeneration with high levels of understory competition was found
to require more light to achieve the same growth rates, and this
effect increased in higher light environments. A stand level
model was developed and demonstrated that while overstory density
is the dominant factor influencing understory vigor, understory
stocking, shrub cover, spatial arrangement of the cohorts are
also important. A three stage progression of overstory treatment
types is recommended to balance the tradeoffs between stand volume
growth, structural and habitat goals, and understory vigor. By
combining elements of shelterwood, group selection, and single
tree selection systems, multi-cohort, structurally complex stands
can be created and maintained in a shifting mosaic of patches.
Results suggest that uneven-age management is possible with Douglas-fir
on dry sites, although stands will be structurally different
from west-side, late-successional forests that contain shade
tolerant conifers and will require periodic stand entries to
maintain. TABLE OF CONTENTSClick here for a PDF Version of this Thesis
LIST OF FIGURES
LIST OF TABLES
ACKNOWLEDGEMENTSI would first like to acknowledge and thank Jeff Foster and Gary McCausland of the Fort Lewis Forestry Program for funding this project and giving me the opportunity to work and do research at Ft. Lewis. Jim Rhode and numerous members of the Fort Lewis Forestry field crew also provided valuable assistance along the way. I am also grateful to: Bruce Lippke and Larry Mason of the Rural Technology Initiative at the College of Forest Resources for their support throughout my 3 years at the CFR; members of my graduate advising committee, Drs. Tom Hinckley, Bruce Lippke, Eric Turnblom, Kristina Vogt, and Tim Harrington for their guidance, ideas, encouragement, and feedback throughout the project; Drs. Tim Harrington, Dave Marshall, Connie Harrington, and the summer field crew of the U.S.F.S. Pacific Northwest Research Station in Olympia, for being key partners in the project and offering invaluable advice and field assistance; Amy Miller for her help in the field and partnership in the analysis; fellow grad students at the CFR who were a vital part of both this project and my overall graduate experience, Andrew Larson, Michael Andreau, Cara Nelson, Elaine Oneil, Ann Andreau; Bianca Perla, Mitchell Almaguer-Bay, Mariano Amoroso, Jeff Comnick, and many others; Dr. Gabe Tucker for his mentorship early on in this project; Dr. Rolf Gersonde for his helpful ideas and critical review; and Dr. Charlie Halpern for the use of field equipment. Finally, I would like to thank my father Jack Churchill for his help in the field and encouragement, and most of all my wife, Wendy Finkleman for her tremendous support and understanding during this project.
IntroductionIn recent years, changing social values and increasing scientific understanding of forest ecosystems have led to greater consideration of alternatives to traditional even-aged management in forests west of the Cascade Mountains in the Pacific Northwest, especially where economic returns are not the driving management objective (Gordon 1994, Kohm and Franklin 1997, Curtis et al. 1998). After research into “partial cutting” was largely abandoned in the 1950’s (Munger 1950, Issac 1956), it was re-established in the mid 1990’s (Curtis et al. 1998, Ruel et al. 2000, Hunter 2001). Much of this research has been aimed at developing the knowledge and tools to grow conifer species under partial overstories, whether in silvicultural prescriptions designed to retain or accelerate the development of late seral structures (Franklin et al. 1997, Carey 2003), restore riparian function (Emmingham et al. 2000, Tappeiner et al. 2002), or establish successful regeneration in uneven-age management systems (Tappeiner et al. 1997, Miller and Emmingham 2001, Coates et al. 2003). Managers and landowners have been reluctant to implement alternatives to traditional even-aged management on a wider level, however, due in large part to the lack of proven management strategies to grow Douglas-fir (Pseudotsuga menziesii (Mirb.) Franco) in the understory, and the perception of an inevitable shift in overstory composition towards more shade tolerant species (Becker 1995, Emmingham 2002). Considerable debate exists in the scientific literature and
among foresters whether uneven-age management of Douglas-fir,
which is moderately shade tolerant (Issac and Dimock 1958, Herman
and Lavender 1990), is possible using single tree selection or
whether group selection is necessary (Curtis 1998, Malcolm et
al. 2001). Single tree selection typically involves removing
individual mature and low-vigor trees dispersed throughout the
stand in relatively frequent light thinnings with the goal of
creating and maintaining multiple cohorts over time (Smith et
al. 1997). While establishment of Douglas-fir seedlings in the
understory is not uncommon after light thinning entries (Bailey
and Tappeiner 1998, Harrington et al. 2003), growth rates generally
decline substantially as the sapling stage is reached (1-3m,
3-10ft), virtually eliminating recruitment into the mid- and
upper-canopy (Herman and Lavender 1990, Williams et al. 1999).
To maintain adequate vigor and recruit into the mid- and upper-
canopy, understory Douglas-fir saplings require a substantial
amount of growing space (Oliver 1995). Thus for single tree selection
to work, it is recommended that the overstory be heavily thinned
early in a stand’s development and then kept open through
successive light thinnings (Bailey 1996, Emmingham et al. 2000,
Brandeis et al. 2001b, Tappeiner et al. 2002).
The critical question in making either approach work is the amount of growing space Douglas-fir needs to maintain adequate vigor in the understory, and thus at what densities and spatial arrangements to maintain the overstory. Numerous investigators have addressed this question by correlating measures of overstory density and/or light availability with understory Douglas-fir growth, both at the stand and individual tree level (Table 1). From these studies, a very general conclusion can be drawn that a maximum of 40% of full stocking or 30-40% of full sunlight is necessary to maintain adequate vigor. However, conclusions from different studies have been variable and can be difficult to develop into comprehensive silvicultural guidelines for several reasons. First, the definition of adequate vigor is often not well defined or consistent among studies. Many authors directly or indirectly define adequate vigor as growth rates that are close enough to open grown, maximum site potential for a tree to recruit into the overstory without any release in the future (Wampler 1994, Bailey 1996, Brandeis et al. 2001b, Drever and Lertzman 2001) Yet, in uneven-age management systems with periodic overstory removals, understory trees need only to maintain sufficient leaf area, stem stability, and root system to avoid mortality and be able to respond to release in the future (O'Hara 1996, Ruel et al. 2000). Thus growth rates that are far below open grown, maximum site potential can be considered adequate for understory trees. Clearly defining vigor in terms of a tree’s ability to grow into the overstory without further release, maintain its release potential, or merely survive is a critical step in translating the variable results created by the different definitions of adequate vigor (Table 1) into guidelines for uneven-age management.
Defining vigor in terms of maintaining release potential requires an understanding of response to release. Results from studies on release of Douglas-fir, as well as inference from release studies of other western conifers, suggest that Douglas-fir can respond to release after even severe suppression and that risk of mortality, lag time before response, and post-release growth are strongly related to duration of suppression, and pre-release growth rates and live crown ratio (LC ratio) (Helms and Standiford 1985, Carlson and Schmidt 1989, Tesch and Korpela 1993, Kobe and Coates 1997, Deisenhofer 2000, Wright et al. 2000). Thus, acceptable levels of risk and desired post-release growth rates should guide definitions of pre-release vigor in terms of thresholds for pre-release growth rates or live crown ratio (Helms and Standiford 1985, Tesch and Korpela 1993, Ruel et al. 2000). Several investigators have designed vigor classification systems that have set such thresholds for Douglas-fir (Carter and Klinka 1992, Emmingham et al. 2000, Miller and Emmingham 2001). However, these systems are based on different growth metrics, involve qualitative criteria that can be difficult to replicate, or were designed for different size classes of trees. As rates of height and radial growth are affected by tree size, vigor thresholds set for seedlings based on absolute, and not relative, growth metrics may not be appropriate for saplings. The same is true for height to diameter ratio (HD ratio). Although an HD ratio of 60 is a commonly used threshold of adequate vigor (Newton and Comeau 1990), this number is questionable as the stem height of diameter measurements and the total height of the tree influence the value and implication of the ratio (Mustard and Harper 1998, Wilson and Oliver 2000). More research is needed to develop thresholds for multiple growth metrics that can be applied across different size classes and correspond to definitions of vigor that have clearly defined management implications. The second reason for the lack of comprehensive guidelines for uneven-age management of Douglas-fir is the difficulty in accounting for the large number of factors that affect growing space in complex stands, often with methods that were designed for even-aged management systems. Traditional, distant-independent stem measurements such as basal area or Stand Density Index (SDI) (Reineke 1933) are often poorly correlated at the plot level to light levels reaching the understory (Chan et al. 1997, Brandeis et al. 2001a, Aukema and Carey 2003) and thus are typically weak predictors of understory growth for individual trees (Wampler 1994, Deisenhofer 2000, Brandeis et al. 2001a). Light or canopy closure measurements have shown greater predictive power (Carter and Klinka 1992, Chen and Klinka 1997, Drever and Lertzman 2001), but are time consuming and hard to translate into stand level management prescriptions. Accounting for side shading (Oliver and Larson 1996) from intra-cohort and shrub competition is also essential to better explain what controls the growth of advance regeneration (Brandeis et al. 2001b, Duchesneau et al. 2001, Canham et al. 2004). Adding further complexity is research that suggests that Douglas-fir may be more shade tolerant on drier sites (Carter and Klinka 1992, Wampler 1994, Bailey 1996, Chen et al. 1996, Williams et al. 1999). A final factor is tree size. There is evidence that as Douglas-fir get older and taller, its light requirements to maintain growth also increase (Carter and Klinka 1992, Messier et al. 1999).
Growth models have been developed for multi-cohort stands in other forest types that factor in some or all of the complex set of variables listed above to help managers determine stocking guidelines for different stand structures and management objectives. These models include: stand-level, distant-independent models based on crown competition (Biging and Dobbertin 1995, Hasenauer and Kindermann 2002), leaf area (O'Hara and Valappil 1999), or SDI (Long 1995, Ralston et al. 2003); and spatially explicit crown and light models (Biging and Dobbertin 1995, Coates et al. 2003, Gersonde 2003). Many of these models show promise for uneven-age management of Douglas-fir. However, spatially explicit models in particular require more inventory information and technical resources than most management agencies, who are accustomed to models designed for even-age stands such as Forest Vegetation Simulator (FVS) or Organon, typically have. There is a clear need for stand level growth models that do not require spatially explicit inventory information but can account for the vertical and horizontal heterogeneity of multi-cohort stands (Monserud and Robinson 2003). In this study, I inventoried vigor and stand structure of naturally regenerated Douglas-fir advanced regeneration in dry-site conifer forests in the southern Puget Lowlands of Washington State. Methods were developed to quantitatively assess and classify vigor using metrics that can be integrated with existing inventory datasets. I sought to clearly define the implications of different levels of vigor by linking quantitative classifications of vigor with estimates of release potential and the likelihood of recruitment into the overstory without further release. These estimates were made from some data gathered in this study, but primarily from results from other investigators. I then used these vigor assessment methods to investigate and model the factors influencing vigor of understory regeneration at the individual tree and stand levels. Finally, results were combined into silvicultural recommendations for uneven-age management in dry site Douglas-fir forests. MethodsStudy area
Field sampling The 13 stands in the phase I stand inventory, totaling approximately
680 ha (1500ac), were selected using a proportional, stratified
random sample. First, the total population of prairie colonization
stands was divided into 3 overstory basal area (BA) classes:
18-29m2/ha (90-142ft2/ac), 30-39 m2/ha
(143-190 ft2/ac), and 40+ m2/ha (190+ ft2/ac).
Stands that had been thinned in the last five years or had access
restrictions were excluded. Stands were then randomly selected
from the three BA classes in proportion to the number of stands
in each BA class in the total population. This process produced
a set of stands that were a good representation of the range
of structural conditions found in the total population of prairie
colonization stands (Foster, pers comm.). All stands had had
at least two thinning entries. Site index (King 1966) ranged
from 34-38 (metric) (110-125, English) (Foster, unpublished data)
and was assumed to be 35 (metric) (115; English) for all stands
given that within stand variation is as great as among stands
due to the heterogeneity of the glacial soils. The overstory was sampled in two ways. First, trees over 15cm (6in.) diameter at breast height (dbh; diameter at 1.37m above ground level) were sampled using point sampling (variable radius plots) (Bitterlich 1947). Basal area factors used were 28, 34, or 40 (English scale) depending on the basal area class of the stand. Height, crown length, and crown width were measured on two randomly selected trees per plot. Second, a visual estimate of the percent of open sky (VOS), excluding shrubs or trees under 15cm (6in.) dbh, was made from each plot center. The same person estimated VOS in all 13 plots. This VOS measure was modified from Brandeis et al. (2001a) and Deisenhofer (2000) and consisted of estimating the percent openness of the canopy in both the northern and southern halves of the sky and averaging them. Where VOS was at least 20% higher in the southern half, the average was weighted to adjust for edge effects (Hasenauer and Kindermann 2002) using the equation below. It is treated as a unitless, relative measure, instead of an absolute measure of percent of open sky. Eq. 1: Weighted VOS = (3VOS,South + VOS,North)/4
Understory trees, defined as less than 15cm (6in) dbh and taller than 1.37m (4.5ft), were sampled within a circular fixed area plot of either 0.04 or 0.008ha (0.1 or 0.02ac), depending on the BA class of the stand. For all understory trees, dbh was measured and crown class determined based on the height of trees within a cohort: dominant, co-dominant, intermediate, or overtopped. Crown class for open grown trees not near clumps was determined by comparing the height of the tree to nearby trees in the same cohort. The following additional measurements were taken for all understory trees, or a sub-sample, depending on the number trees in the plot: diameter at 15cm (6in) above ground; total height; height to live crown, measured at the lowest whorl with two live branches with new growth; annual height increment for each of the past six years; total age determined from whorls and bud scars, crown density, and degree of crown crowding from intra-cohort competition (neighboring understory trees) and shrubs. Crown density was measured with a 1-5 rating that combined visual estimates of average branch length and diameter, number of internodal buds and branches, and number of branches and each whorl; all in the last 3 years of growth. Ranges of these variables used to rate trees are provided in Appendix A. Carter and Klinka (1992) and Miller and Emmingham (2001) developed similar systems to rate vigor based on visual crown ratings. Crown crowding, defined as any infringement on the live crown of the sample tree from neighboring understory trees in the same cohort or shrubs, was measured using a combination of methods from Howard and Newton (1984) and Wagner and Radosevich (1991, 1998). At the base of the live crown, a horizontal, two dimension circle was visualized around the tree using the longest live branch as the radius. The proportion of this two dimensional circle (projected crown area) overlapped by foliage of neighboring trees or shrubs from the base of the live crown to the top was estimated to the nearest 10 percent. This combined measure of intra-cohort and shrub competition was termed percent crown overlap. A diagram of this method is provided in Appendix B. Data from the phase I stand inventory were generated from 13 stands, 212 plots, and 637 understory Douglas-fir trees and used to broadly characterize structure of the overstory, understory, and shrub layers. After this analysis, a sub-sample of 25 plots from the 212 phase I plots was selected for further intensive analysis. A stratified random sample was used to ensure that the phase II sub-sample of 25 plots came from a balanced distribution of light environments. At each phase II plot, two understory trees were selected for
destructive sampling: the tallest tree within the cohort and
the tree closest to the average height. For each tree, dbh of
all overstory trees were recorded in a 0.09ha (0.2ac) fixed area
plot as well as in a variable radius plot, both centered on the
sample tree. Before each tree was cut down, two methods were
used to measure crown crowding from intra-cohort competition
and shrubs. First, the percent crown overlap method from phase
I was expanded to include a vertical dimension. For each competitor,
the two dimensional horizontal overlap and the percent overlap
along the vertical live crown of the sample tree were estimated.
By adding all the competitors together, an estimate was made
of the percent occupancy of competing understory trees and shrubs
within a three dimensional cylinder projected from the base of
the live crown to the top of the tree with the radius being the
longest live branch. The second method consisted of measuring
the dbh of, and azimuth and distance from the stem of the sample
tree to, the stem of each competitor within a radius set by the
farthest competitor that overlapped the crown of the sample tree.
Competitors had to be as high as the live crown of the sample
tree. This variable radius method was chosen based on Wagner
and Radosevich (1991, 1998), who found that “The optimum
radius appeared to be defined by those neighbors whose crowns
intermingled with that of the Douglas-fir.” A neighborhood
competition index (NCI) was calculated for each tree based on
the following equation from Canham et al. (2004), where DBHi is
the diameter at 1.37m of ith competitor, Disti is
the distance to the ith competitor, and
As competition from shrubs and other tree species was minimal compared to intraspecific competition, the same parameters were used for all competitors. Parameter estimates for these equations were iteratively tested to determine the most powerful values used in later analysis. After each tree was cut down, percent open sky was estimated using the VOS procedure described above and a hemispherical photograph was taken as close to 66% of tree height as could be safely reached with a 5m (16.4ft) orchard ladder using a digital Nikon camera with a 2.5 F-Stop fisheye lens mounted on a monopod with a level. Gap Light Analyzer/C, version 2.0 light modeling software (Frazer et al. 1999) was used to analyze these photographs and determine site openness and an index of total solar radiation (TSR) (Canham et al. 1990, Frazer et al. 1997). TSR combines the seasonal distribution of the sun’s path with the distribution of canopy openness to calculate a single index of available light in units of percentage of full sun for a specified growing season which was set at April 1 – Oct 15th for the Ft Lewis area. All tree measurements from phase I were also re-measured on each phase II tree. In addition, branch length and number of internodal buds were measured for the last 3 years. Percent cover of shrubs by species was estimated in a 0.008ha (0.02ac) fixed area plot around each tree. Height growth increments were measured from the top of the tree to breast height and checked against rings counted on a disc cut at dbh. Discs were also cut at ground level and at the base of the live crown. These discs were later oven dried for 48 hours, sanded, and analyzed using a high resolution computer scanner and WinDENDRO v2001a (Regent Instruments, Inc.) tree ring analysis software, which calculated age and annual radial growth along 4 radii for each disc. These radial growth measurements were combined with height growth data to calculate annual volume increment using a tapered rocket formula where the three discs were used to create two tapered cylinders and a cone at the top. Inside bark diameters were used. Height to diameter ratio history was also calculated using the ground level disc, but diameters were adjusted for bark thickness, drying, and the discrepancy between field measurements with a dbh tape and radial measurements on a disc. This was done by multiplying all inside bark diameters by the ratio of the 2004 field measured diameter to the 2004 inside bark diameter obtained from WinDENDRO.
Assessing vigor Eq 3: y = 0 .018x2 - 0.27x - 466.74 This quadratic equation was the best fit for the maximum site potential tree data (R2= 0.96, p<0.0001) until tree heights approached 1.4 meters, below which it underestimated maximum potential growth. A graph of this regression model is provided in Appendix C. For the 54 trees that were intensively analyzed in phase II, actual RVG was calculated for every year from when the tree was 1.37m to its current height. The last 5 years were averaged to create the primary vigor metric for further analysis: actual 5yr RVG. For the 637 understory trees in the stand inventory, however, coring trees to measure diameter and volume growth was intentionally not done to keep the sampling design within what could be practically integrated into the existing Fort Lewis inventory system. Exploratory analysis was thus done with the 54 tree data set to test which variables that were measured in the stand inventory best correlated with actual RVG. These variables were: past 5yr mean annual height growth increment (HG), 2004 live crown ratio (LC Ratio), 2004 height to diameter ratio (HD Ratio), and visual crown density rating (CDR). Single variables, as well as biologically meaningful products of these variables, were tested using simple linear regression. Stepwise, linear regression was then used to test multiple variables and derive a best fit model using SPSS, version 12.0 (SPSS 2003). This model was used to predict 5yr RVG for the 637 understory trees from the stand inventory. The second method of assessing vigor was a simple four class classification system. This additional method was developed to provide a more efficient and sufficiently accurate means for managers to assess vigor of understory cohorts, conceptualize vigor, and design silvicultural prescriptions. Classes were based on thresholds of HD ratio, HG, and LC Ratio that are listed in Table 2. Trees must meet the thresholds for all three metrics to be placed in a certain class or else they are placed in the next lowest class. Crown density rating was not included to allow for this vigor assessment method to be easily used with existing data sets. Also, crown density rating is a qualitative measure that requires consistency, which can be hard to achieve with multiple observers.
A HD ratio of 70, measured at 15cm above ground, was chosen as the HD threshold for high vigor (class 4) (Table 2). This correlates with an HD ratio of 60 at the root collar. This threshold value is based on observations from: Newton and Comeau (1990) who suggest that HD ratios over 60, measured at the root collar, threaten the long term growth potential of young plantation trees; Wonn and O'Hara (2001) who showed that an HD ratio of 80, measured at dbh, is a critical threshold for stem stability in interior Douglas-fir; Cole and Newton (1987) who found that decreases in height growth occurred at HD ratios above 70 in plantation trees; and Emmingham et al. (2000) who specified 60 as a threshold for a vigorous understory tree, 80 for a stable tree, and 100 for a weak tree. Height growth thresholds were based on approximately 66%, 33%, and 15% of the site index potential height growth for the average tree height found in the study (King 1966). Live crown ratio thresholds (Table 2) were based on the positive relationships of live crown ratio to post-release growth observed by Helms and Standiford (1985), Seidel (1983a), Oliver (1985), and Tesch and Korpela (1993); as well as recommendations by Emmingham et al. (2000). Table 2: Summary data for vigor classification system for understory trees. Trees must meet thresholds for all three growth metrics to be in a class, otherwise they are placed in the next lowest class.
The two methods of assessing vigor were compared to evaluate
whether the simpler vigor classification system could estimate
5yr RVG within a reasonable range. Analysis of variance (ANOVA)
and a Scheffe’s post hoc test was used to test for significant
differences between the mean RVG values for each vigor category,
for both the phase 1 and phase 2 datasets (
Overstory recruitment and release potential Similar to Tucker and Foster (unpublished), the number of years needed to reach the average overstory height of the 13 stands, 45m (148ft), was used as an indicator of the likelihood of reaching the overstory without further release, assuming that current height growth rates are maintained and that trees are currently 5m (16.4ft) in height. This indicator was a simple thought exercise used to compared vigor classes and purposely ignored the dynamic nature of forest stands and the fact that Douglas-fir height growth begins to slow as trees reach approximately 25m (80ft) on site class III sites (King 1966). As a biological limit exists regarding how slow trees can grow and remain alive (Oliver and Larson 1996, Kobe and Coates 1997), risk of mortality from suppression was then examined. As the 54 tree sub-sample was a representative sample of the total population of understory trees, the slowest radial growth rates found from analysis of lifetime radial growth rates were assumed to be close to the threshold of mortality. These minimum growth rates were compared with mortality threshold rates from other studies. Estimates of mortality risk for each vigor class were based on how close the past 5yr mean radial growth increment were to these mortality thresholds. HD ratio thresholds for each vigor class were also considered as high HD ratios are associated with a higher risk of stem failure and suppression mortality (Wilson and Oliver 2000, Wonn and O'Hara 2001). Hemispherical photographs were then used to evaluate whether trees from different classes had enough space directly above them to recruit into the overstory, assuming that existing overstory trees would expand their crowns. Percent of total solar radiation and visual examination of the photos were used. Release potential was defined in terms of a released tree’s ability to regain comparable growth rates, for a given light environment, to trees that did not undergo suppression. To assess release potential, lifetime RVG histories of the 54 intensively analyzed trees were examined for patterns of suppression and release. Trees that showed patterns of release were compared to trees growing in similar light environments that had not undergone suppression. Lifetime HD ratios were also examined to see the extent to which trees recovered from high ratios. Although the exact history of overstory removal around each tree could not be confidently determined, it was assumed that at least some of the trees growing in open conditions had been released in the past and should therefore show a growth response. Trees growing in open conditions that were growing at low RVG rates were closely examined to determine why they had not responded to release. Finally, the live crown ratio thresholds for each vigor class were also considered.
Factors influencing vigor: individual tree and stand
level At the stand level, two methods were used to compare the overall vigor of the understory between stands. First, the number of understory trees per acre in each vigor class was tabulated for each stand. Second, the stand level, mean tree 5yr RVG was calculated for all trees, and then separately for crop trees, by averaging the mean RVG of each plot with understory trees. A non-linear model was developed to explain variations in stand mean RVG and test which factors influence understory vigor at the stand level. The model was based on the concept that average growth of an understory cohort is determined by growing space available to the cohort divided by the number of trees in the cohort (O'Hara 1996, Oliver and Larson 1996). Although multiple understory cohorts did exist in some stands, all understory trees were lumped into one cohort to simplify the analysis. Stand density index (SDI) was chosen to measure growing space occupancy for both the overstory and understory, as it has been shown to be a good indicator of leaf area index (Long 1995, O'Hara 1996) and is commonly used to measure site occupancy (Long 1985, Long and Daniel 1990). Percent shrub cover was tested in the model to account for its potential effect on resource availability. The average crown overlap for the stand was also tested to account for the spatial distribution of understory trees; whether they are clumped or more evenly distributed. A final variable was tested to account for the distribution of understory trees in relation to the heterogeneity of overstory density. This variable was the ratio of average overstory SDI for plots with understory trees to the average overstory SDI of all the plots. Stands with understory trees in more open plots thus have a lower value than stands where more trees are in plots with higher levels of overstory or where gaps are occupied with shrubs instead of trees. Several model forms were tested and the best combination of variables was chosen through iterative trials. SPSS, version 12.0 (SPSS 2003) was used for this analysis.
ResultsStand inventory Assessing vigor Stepwise linear regression produced the following best fit model to predict 5 yr mean RVG using all the above variables and total height (Ht) (R2= 0.89. p<.0001, Standard Error of Estimate (SEE) =0.058, All coefficients are significant p<.005). Eq 4: Predicted RVG = 1.337 + 0.076LCR2CDR -
0.009Ht(m) - 0.028ln(HD Ratio) + .003HG(cm) Without height, HG drops out of the model due to collinearity with LCR2 *CDR. In the full model, however, the variance inflation factor (VIF) is below 10 for each variable indicating the collinearity is not significant. Height is a not a significant predictor of RVG on its own, but affects the interaction of the other variables. Residual and QxQ plots indicated a relatively constant variance and approach to normality for the overall model. A second, slightly weaker model to predict RVG that does not include crown density rating is listed in Appendix D for use with data sets where crown density rating is not available. The model in equation 4 produces negative RVG values for trees with very high height to diameter ratios and/or very low live crown ratios. Few of these low vigor trees were part of the 54 tree, phase II sub-sample used to create the model only the average and tallest trees were selected for sampling. While the model is adequate for intermediate, co-dominant, and dominants trees, it is likely to be less accurate for trees from the overtopped crown class. Although negative values do not make biological sense, they were included in subsequent analysis as indicative of RVG close to zero. The second method of assessing vigor, vigor classes based on thresholds, is compared with RVG in Figure 3. The extreme upper values in each category are due to high height to diameter ratios. For the 54 phase II trees, the means of actual RVG and radial growth for the 4 vigor categories are all significantly different from each other (p<0.05). The means of RVG from 637 trees from the phase I data set, which are also significantly different from one another (p<0.001)., all fall within the 95% confidence intervals of the phase II dataset displayed as the solid horizontal lines in figure 3a. There is also a clear separation of the 50th percentile boxes of each class within the boxplots in figure 3b. Finally, the combined phase I and II average RVG values (rounded to the nearest 5%) for each vigor class are presented in Table 2.
Overstory recruitment and release potential
Factors influencing vigor: individual tree level
Dividing crown overlap into two classes (high and low) and including it as an indicator variable generated the best regression models to predict RVG (Figures 5 & 6). The breakpoint for the two classes that produced the best models was 30% for the 3-dimensional crown overlap measured on the 54 phase II trees, and 50% for the 2-dimensional crown overlap measured on the 637 trees in the phase I dataset. Regression models are constrained to the range of the data and caution should be used making any extrapolations. The boxplots in Figure 7 were included as a secondary method of showing the relationships between RVG and VOS, SDI, and crown overlap and to better display the distribution of the data.
In the phase I VOS (Figure 6a), phase II VOS (Figure 5a), and phase II TSR (Figure 5b) regression models, ANCOVA showed statistically significantly steeper slopes and higher R2 values for trees with low crown overlap than trees with high crown overlap. Intercepts were not significant for either high or low crown overlap. The effect of crown overlap was not significant at the lowest light levels but became significant at the point where the 95% CI of the regression lines diverged, and had an increasing effect as open sky increased and overstory density decreased. The divergence in the 95% CI intervals is due to increasing differences between means, as well as a drop in variance, as light levels increased.
The phase II VOS model (Figure 5a) has a higher R2 value and lower intercept and slope than the phase I VOS model (figure 6a). The drop in R2 values is most likely due to increased variance introduced by the RVG prediction model and increased measurement error as VOS was measured at plot center in phase I and not for each individual tree as in phase II. Phase I and II regression models for SDI had similar slopes, intercepts and R2 values and so only the results for the phase I, larger data set are presented in Figure 6b. Low crown overlap had a higher R2 value and a significantly different intercept than high crown overlap, but the slope and curvature were not significantly different. Similar to the VOS models, a pattern of an increasing effect of crown overlap as SDI decreased was found. Between 200 and 250 SDI (Figure 6b), there is a clump of 6 trees above 35% RVG that were all on the edges of gaps and so were receiving more light for their SDI value than trees that were underneath a more continuous canopy. Accounting for this clump of outliers, there is a clear jump in the response of RVG at 150 SDI, which is very evident in the boxplot shown in Figure 7b. It should be noted that 0 SDI does not mean that the tree is open grown, only that no overstory trees are within the plot radius measured. The upward lip in the quadratic model beginning at 350 SDI is due to the lack of data above this level and the nature of the quadratic function. A quadratic model form was chose over an exponential form as it had a much higher R2 value. The model should be constrained to SDI values less than 350. The second vigor assessment method, the 4 vigor classes, was also related to VOS, SDI, and crown overlap using boxplots (Figure 8). These were derived to clearly show the ranges of VOS and SDI values that trees from the 4 vigor classes were found in.
VOS showed a strong relationship to TSR (Figure 9), especially for trees with lower crown overlap (R2=0.86). This was expected as VOS measured openness of just the overstory trees, while TSR measured all trees affecting the light environment of the subject tree. The relationship is not a 1:1 relationship as VOS over-predicts TSR at lower levels and under-predicts it at higher levels. This indicates that VOS is a relative and not absolute measure of actual solar radiation and should not be interpreted or applied as an absolute percent of open sky. The regression equation in Figure 9 can be used to convert the relative, visually estimated VOS into TSR in order to compare light levels with other studies.
In addition to the main vigor metric, RVG, regression models predicting live crown ratio and HD ratio were developed from the same set of explanatory variables as RVG. LC ratio and HD ratio showed the same basic relationships to VOS, SDI and crown overlap. Regression models were consistently weaker than models for RVG and so are not presented here. Instead, boxplots are shown in Figure 10 to display the distribution of LCR and HD ratio data by VOS and SDI class and crown overlap.
Factors influencing vigor: stand level
Eq. 5:
Eq. 6:
Between 85-90% of the variation in RVG was explained with only overstory SDI, understory SDI, and tall shrubs or low shrubs in each model. Non-significant coefficients were included to add predictive power. As only 13 data points were used, the asymptotic confidence intervals of the non-linear regression method are large. With more data points, they may in fact be skewed in a negative or positive direction and be significant.
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
and 
are
model parameters: 
