About This File: This file was created by scanning the printed publication. Misscans identified by the software have been corrected; forest Ecology and Management however, some mistakes may remain. ELSEVIER Forest Ecology and Management 98 (1997)49-60 Comparison of diameter-distribution-prediction, stand-table-projection, and individual-tree-growth modeling approaches for young red alder plantations Steven A. Knowe • a .* , Glenn R. Ahrens a , 1 Dean S. DeBell b Depanment of Forest Science. Oregon State UniL'ersiry. Cvn·allis. OR 97331. USA USDA Forest Sen·ice. Forestry Sciences Laboratory. Olympia. WA 98502, USA b .-\ccepted 4 io.larch 1997 Abstract A red alder planting spacing srudy was used to compare three modeling approaches that have been successfully used for other tree species. These three approaches predict stand strucrure and dynamics in plamations that are 7 to i 6 years old. with planting densities of 976 to 13 399 trees/ha. The diameter-distribution-prediction approach tended to \)ver-predict the diameter at breast height (dbh) for larger trees in stands planted at low density and to under-predict dbh for smaller trees in stands planted at high density. This approach may be useful for comparing planting densities when a tree list is not available. The stand-table-projection approach tended to under-predict dbh for smaller trees in young stands planted at low density and to over-predict dbh for smaller trees in young stands planted at high density. This approach. however, provided consistency between stand- and tree-level growth projections. and should be useful for comparing planting densities when a tree list is available. The individual-tree-growth approach provided the best representations of observed diameter distributions at all planting densities, stand ages, and growth intervals. This approach may be best suited for stands that have been thinned. stands with mixrures of species, and stands with heterogeneous size classes. © 1997 Elsevier Science B. V. Keywords: Alnus ntbra: Weibull function: Relative size; Distance-independent: Simultaneous regression: Compatible equations 1. Introduction Red alder ( Alnus rubra Bong.) is the most abun­ dant hardwood species in the Pacific Northwest. Recently, and for a variety of reasons. there has been an increased interest in managing red alder as a timber resource (Hibbs et a!., 1989, 1994). The Corresponding author. Tel: 904-225-5393: Fax: 904-2250370; E-mail: steve.knowe@ravonier.com. 1 Paper 3047. Forest Res Laboratory, Oregon State Uni­ versity, Corvallis. OR . USA. • e:uclt hardwood industry has begun to manage alder in plantations to ensure future availability of alder saw­ timber (Hibbs et al.. t 994: Raettig et a!., t 995). Also, because alder is not susceptible to laminated root rot ( Phellinus weirii), it is increasingly planted on sites infected with the disease. When n\anaged in mixture with conifers on suitable sites, alder can increase nitrogen tixation and. therefore. productivity (Tarrant and Trappe. 1971; Tarrant et al.. 1983: Binkley et al.. 1994). Very little is known about the growth and devel­ opment of alder in plantations. however. Much of the 0378-1127/97j$17.00 1997 Elsevier Science B.V. .\II rights reserved. Pfl S0378-ll27(9 7 )00075-3 S.A. 50 Knowe eta/./ Forest Ecology and Management 98 ( 1997) 49-60 growth and yield research conducted on red alder has sizes and number of surviving trees equals projected focused on natural stands and size-density relation­ stand basal area and stand survival, respectively. 1987; Hibbs and Carlton, 1989; Puettmann et al., 1992), and on thinning effects (Hibbs et al., 1989) and height-age curves (Harring­ ton and Curtis, 1986). Silvicultural regimes proposed Diameter distributions obtained for each modeling for crop trees often rely on relatively low planting limited long-term plantation data were available. the ships (Hibbs, method were compared to the observed diameter distribution across a range of planting densities and 1- for or 2-year and 5-year growth intervals. Because densities (compared with typically dense natural re­ equations presented here represent a framework for generation) in order to maximize diameter growth additional modeling as more data become available. during the alder's rapid juvenile growth phase (Hibbs Despite limitations in the quantity of data. we feel and DeBell, 1994). Preliminary results of plantation trials indicate that spacings in the range of 2 to may produce a dominant alder stand within 2 3 m to 3 the assessment of modeling approaches was not oth­ erwise limited. years, while providing adequate space for rapid di­ ameter growth (Ahrens et al., Bell, 1992; Hibbs and De­ 1994). There is a critical need for detailed quantitative information about the performance of alder at differ­ 2. Methods 2.1. Data The red alder spacing study site was established ent planting densities. Toward that end, a network of 1974 near Apiary. Oregon. Six planting densities new test plantations has been established to provide in a high-quality database for modeling growth and were included in the study: 976 (3.2 X 3.2 m), 1600 (2.5 X 2.5 m), 3086 (1.8 X 1.8 m), 4630 (1.2 X 1.8 m), 6944 (1.2 X 1.2 m). and 13 889 (0.6 X 1.2 m) yield at managed-stand densities (Hibbs et al., 1993). In anticipation of this database, an appropriate mod­ eling framework needs to be developed to ensure treesjha. respectively. These planting densities are that the future database will include the appropriate higher than those currently being considered for v ariables. As a first step, modeling methods that operational plantations (Hibbs and DeBell. have been applied to other species should be evalu­ Densities of ated. Such an evaluation may be applicable to exist­ by one plot each, and the remaining planting densi­ 976 1994). and 3086 treesjha were represented 1). ing red alder management systems-specifically, to ties were represented by two plots each (Table yield systems. With few exceptions. all plots were measured annu­ The purpose of this paper was to develop and then ally from plantation ages 7 to 12 and at ages 14 and compare modeling systems that predict stand struc­ 16 ture and dynamics in red alder plantations. Though vations. Diameters of all surviving trees on each plot years. The resulting damset consisted of 7'2 obser­ long-term data is limited, an indication of the long­ term performance of alder over a range of spacings was available from one dataset (described below). Useful as a test of model forms, this existing dataset contains a variety of initial planting spacings and ages of rapid stand dynamics. Modeling approaches included (1) for stand-level data only, a diameter-distribution-prediction method; (2) for both stand- and tree-level data, a stand-table­ projection method; and (3) Table I Summary of data used to develop and compare mode Iing ap. proaches for red alder plantations Density (treesjha) Spacing (m) Plot size (ha) Number of plots and measurement ages 976 3.2X3.2 0.0809 1600 2.5X2.5 0.0809 3086 4630 6944 13 889 1.8X 1.8 1.2X 1.8 L2X 1.2 0.6X 1.2 I plot with ages 7 to 12 only l plot with all ages l plot with ages !4 and 16 only I plot with all ages 2 plots with all ages 2 plots with all ages 2 plots with all ages for tree-level data, an individual-tree-growth modeL The stand-table-pro­ jection and individual-tree-growth methods are con­ ceptually similar in terms of generating future diame­ ters. The former does so, however, in a manner such that the sum across diameter classes of projected tree • 0.0187 0.0196 0.0180 0.0163 S.A. Knowe et al. I Forest Ecology and Management 98 I 1997) 49-60 were measured to the nearest mm. and total height of 10 trees per plot was measured to the nearest 0.1 m. 51 Shiver (1986); this function is asymptotic with re­ spect to dominant height and survival. A compatible. path-invariant basal area projection function (Eq. 2.2. Regression analysis (1h)) was derived from the prediction function by using an algebraic difference method. Parameters of All measuremenc ages were included in the devel­ opment of the diameter-distribution-prediction method, but the predictions for age 7 were not the Weibull diameter distribution function were ob­ tained by using a percentile-based parameter recov­ ery procedure as described in Section 2.3. The four considered in the comparison and assessment of diameter distribution percentiles were predicted as modeling approaches. For the stand-table-projection functions of quadratic mean diameter at breast height and individual-tree-growth methods, data were ar­ (dbh), which was obtained from basal area and sur­ ranged in nonoverlapping growth intervals ot' l to 2 viving number of trees at a given age, and plantation years. According to Borders et al. (1988), there are age ( B ailey et al., 1989; fewer problems with serial correlation of real growth 1990; Knowe and Stein, 1995). Borders and Patterson, series-derived from either remeasured plots or trees A survival function (Eq. (1f)) was selected for use -when the data are arranged in nonoverlapping in all three modeling approaches. The path-invariant growth intervals rather than all possible intervals. function developed by Clutter and Jones (1980) is Error component models for stand-level equations based on a differential equation which implies that (Gregoire, 1987) partition residuals into plot errors. the instantaneous rate of stand mortality is propor­ which remain constant for a given plot through time. tional to age and density. and measurement period errors. which remain con­ Mortality rate of other tree species may also be stant for all plots at a given measurement age. Simi­ related to site index. with greater mortality occurring larly, error component models for tree-level incre­ on the more productive sites. However. a preliminary ment equations (Stage and Wykoff. 1993) include model with age replaced by dominant height did not stand-level errors tor plots and measurement periods fit the observed data as well as the model with age. plus the errors associated with trees within plots and measurement errors. Because the individual-tree­ An algebraic difference fonnulation of the Richards (1959) equation (Eq. (1g)) was selected for growth models are based on tixed projection inter­ describing dominant height growth patterns in red vals, the 2-year growth intervals between 12 and 14 alder plantations. The resulting equation (Eq. ( 1g)) is dividing periodic growth by 2 years. These properties pennit the equation to be alge­ and between 14 and 16 years were annualized by Stand-level models for the diameter-distribution­ prediction and stand-table-projection methods, in­ cluding prediction equations for basal area and diam­ anamorphic. base-age invariant, and path invariant. braically rearranged for predicting either dominant height ( H age (A 1 ), 1) as a function of site index ( H-c), current and base age (A ) or site index as a eter distribution percentiles, and path-invariant pro­ function of dominant height. current age. and base jection equations for survival, dominant height, and age. basal area, were fit as a nonlinear system of seem­ ( 1 a) ingly unrelated simultaneous equations (Eqs. (1a), (1b), (1c), (ld), (le), (lf), (lg) and ( l h)). The path invariant property results in the same final yield regardless of whether several short growth intervals or a single long growth interval are used. Fitting the stand-level models as a system of simultaneous equa­ tions accounted for the contemporaneous correlation Do= ,\10 +A.,, Dqt + A12 A, (1b) D25 = ,\20 + A21 Dqt (lc) Dso= A 3o + A31 Dqi (ld) ( 1 e) among equations (Borders, 1989). The basal area prediction function selected for young red alder plantations (Eq. ( l a)) was a nonlin­ ear variant of the model proposed by Pienaar and ( lf) 52 S.A. Knowe et al. I Forest Ecology and Management 98 I 1997! 49-60 ( ) (-) (-) N 2 BA -=BA1N1 al H2 a H1 A2 A1 ( lg) aJ ( lh) where A 1 =plantation age (years) at the start of the growth period, A2=plantation age (years) at the end of the growth period, BA 1=predicted basal area (m2 /ha) at the start of the growth period, BA2= projected basal area (m2jha) at the end of the growth period, D0 Oth percentile of diameter distribution, D25 = 25th percentile of diameter distribution, D50= 50th percentile of diameter distribution, D95=95th p ercentile of diam e te r distribution. D " ' = /BA1/(0.00007854 X N1), H1 = dominant height (m) at the start of the growth period. H, = projected dominant height (m) at the end of the growth period, N1 =number of surviving treesjha at the start of the growth period. and N2=projected number of surviving treesjha at the end of the growth period. Using the approach applied to red alder planted in Neider plots (Knowe and Hibbs. 1996), parameters of the basal area prediction function and dominant height projection function were fit to each planting density. The resulting parameter estimates were ex­ amined for relationships with planting density; how­ ever, no consistent relationships were observed. Parameter estimates for individual-tree equations used in the stand-table-projection and individual-tree diameter growth methods were obtained by using nonlinear regression. Distance-independent tree-level models included a relative basal area projection tunc­ tion (Pienaar and Harrison, 1988; Borders and Patter­ son, 1990; Knowe, 1994; Knowe and Stein, 1995; Knowe and Hibbs, 1996) and a diameter growth equation (Hester et al., 1989; Wykoff, 1990). = 2.3. Diameter-distribution-prediction approach The Weibull cumulative distribution function has been widely used to model diameter distributions since the early applications by Bailey and Dell (1973) and others: p( D) I { ( 1- exp - D a n (2) where p( D)= cumulative probability of a diameter (D) less than or equal to D, a= location parameter, corresponding to the minimum dbh. b= scale pa­ rameter. and c=shape parameter. This approach is most useful for comparing the effects of planting density when a tree list is not available. Only stand-level information is needed to generate a diameter distribution at any age. Compo­ nents of this approach include a basal-area-prediction function (Eq. (la)), the survival function (Eq. ( lf)), diameter-distribution-percentile-prediction functions (Eqs. (lb), ( l c), (ld) and (le)), and a parameter recovery procedure for the three-parameter Weibull distribution tunction. (The basal area function se­ lected for red alder plantations is asymptotic with respect to the dominant height and survival tunc­ tions.) Recently, parameter recovery procedures have been developed to estimate the Weibull parameters (Bailey et a!., 1989; Borders and Patterson. 1990; Knowe and Stein. 1995). The main advantage of these procedures is that diameter distribution charac­ teristics. such as minimum diameter or diameter percentiles. can be predicted with more confidence than the parameters, which are indirectly estimated by analytical relationships. A percentile-based ap­ proach has been used to model fertilization effects (Bailey et a!.. 1989), the impacts of interspecific competition in conifer plantations (Knowe et al.. 1992: Knowe and Stein, 1995), and mixtures of eastern cottonwood (Populus deltoides L.) clones (Knowe et a!., 1994). The parameter recovery procedure is based on the Oth. 25th. 50th, and 95th percentiles ( D0,D25,D50, and D95, respectively). Assuming that c=3. the location parameter, a, is obtained by using the mini­ mum ( D0) and median ( D50) diameters and sample size (n): o n t.J333 Do - Dso a= 0 ll .3333- 1 ( 3) If Eq. (3) results in a negative value, then a is set to zero. The sample size ( n) is the product of the plot size and the surviving number of trees. For predict­ ing a in Eq. (3), sample size may be estimated by multiplying the number of surviving treesjha ob­ S.A. 53 Knowe et al.j Forest Ecology and ivfanagemem 98 I !9971·J9-t50 rained from the survival function by 0.032. the aver­ age plot area in hectares. The shape parameter is estimated by using the estimate for the location parameter (a) and D95 and where bli = basal area of rh tree at the start of the growth period. b:i =basal area of of the growth period. irh tree the at end b 1 =average basal area per tree at the start of the growth period. b:: = average basal area per tree at the end of the growth period. D2s: 2.343088 ( 4) c = -----In D95- a -In D25- a ( ( ) ) The scale parameter, b, is obtained by solving the second moment of the Weibull distribution for the p ositive root using the estimates for a. c, and D : A1 =age at the start of the growth period, A2 =age at the end of the growth period, and {3 =parameter to be estimated. The sign of {3 can be interpreted as the future contribution of individual trees in the projected stand. If {3 is positive, the relative contri­ bution to basal area in the future stand will increase for trees larger than the average relative size and will decrease for trees smaller than the average relative (5) where T1 = T ( l + ljc).T" = T(l + 2/c). T= the gamma function, and other terms as previously de­ tined. This procedure ensures that D4 in the pre­ dicted diameter distribution is the same as the D4 implied by prediction equations. size. Conversely. if [3 is negative. then the relative contribution will decrease for larger trees and in­ crease for smaller trees. Inspection of the parameters of the relative size projection function obtained at each planting density suggested a quadratic relationship. with the greatest parameter estimate at the 71 T2 planting density. However. only a small improvement in explained variation 2.4. Stand-table-projection approach was achieved by including a ing projected diameter distributions were not as ac­ This approach produces a future stand table that is consistent with tree- (Pienaar and Harrison. and 1988). stand-level functions Therefore, a tree list and stand information are needed. Components of the stand-table-projection approach include a relative size projection function. which contributes the tree­ level information, and the survival function and a basal area projection function, which contribute to the stand-level information. According to Borders and Patterson (1990), the advantages of the stand-ta­ ble-projection method are as follows: the functional form of the diameter distribution (e.g., Weibull) does not have to be assumed: multimodal distributions can be reproduced; and information on initial stand struc­ ture can be use�i. The tree-level function of Pienaar and Harrison ( 1988) ( < 0.1 C:C) quadratic function of planting density. and the result­ is based on relative tree size, defined as the ratio of individual-tree basal area to average basal curate as those obt:llned by using the overall parame­ ter estimate. The second part of the stand-table-projection sys­ tem developed by Pienaar and Harrison (1988) is the projected mean size obtained from stand-level basal area and survival functions. The stand-basal area­ projection function (Eq. (l h)) is an implied growth function that is functionally compatible with the basal-area-prediction function (Eq. ( 1 a)). The future diameter of individual trees can be estimated from the product of projected relative basal area and projected mean basal area per tree. which is the quotient of the projected basal are divided by the number of surviving trees. The relative size projection function for individual trees can be con­ strained to ensure consistency of the future stand table with stand-level basal area and survival func­ tions. area per tree in the stand, and changes in relative size over time. (6 ) (7) 54 S.A. Knowe eta/./ Forest Ecology and Management 98 ( 1997) -/.9-60 where b2i =projected basal area (m2 /ha) of trees in th dbh class, BA2 =projected basal area (m2 jha) of the stand, ni =number of trees in ith dbh class, and k = number of dbh classes. Stand mortality is allo­ cated to dbh classes, on the assumption that smaller trees in a stand have a greater probability of dying than do larger trees in the stand. Borders and Patter­ son (1990) found this stand-table-projection proce­ dure to be superior to a Weibull-based diameter-dis­ tribution-projection system and a percentile-based projection system with an empirically defined distri­ bution. Variants of this approach have been used to predict the effects of interspecific competition in young Douglas-fir plantations (Knowe, 1994; Knowe and Stein, 1995) and planting density in red alder plantations (Knowe and Hibbs, 1996). 2.5. Individual-tree-growth approach The distance-independent, individual-tree-diame­ ter growth function is similar to those used in ORGAl ON (Hester et al., 1989) and PROGNOSIS (Wykoff, 1990). A tree list is required to make growth predictions for fixed (annual) growth inter­ vals. The function is applicable to stands that have been thinned or to stands with multiple canopies. (8) where Ll Di =annual dbh growth (em) of j!11 tree, Di =current dbh (em) of ith tree, BAL =current basal area (m2 jha) of trees with larger dbh than the ith tree, BA =current basal area (m2 jha) of stand, and SI = site index (m). The potential annual diameter growth of individ­ ual trees is predicted as a function of current diame­ ter. Smaller trees in a stand have less growth than the larger trees, and trees in stands with low basal area exhibit greater growth than trees in stands with higher basal area. Stand mortality is allocated to dbh classes, on the assumption that the smaller trees in a stand have a greater probability of dying than do the larger trees. 2.6. Evaluation of methods The two-sample Kolmogorov-Smirnoff test (Sakal and Rohlf, 1981) was used to determine whether the obserVed and predicted or projected diameter distributions are samples from the same population. This test does not validate the prediction system, but verifies that the observed and predicted stand tables are similar for young red alder planta­ tions. The hypothesis of no difference between ob­ served and predicted diameter distributions was tested by using a=0.05. Beginning with a plantation of age 8, the ob­ served density, dominant height, and age were used to predict the diameter distribution at each measure­ ment. The cumulative diameter distribution was ob­ tained directly from the Weibull distribution func­ tion. Beginning with a plantation of age 7, in the stand-table-projection method, density, dominant height. and age at the beginning of each 1- or 2-year growth interval were used to predict the number of trees and stand basal area at the end of the interval. In the individual-tree-growth prediction method, dbh, BAL, and stand basal area at the beginning of each growth interval were used to predict growth, which was added to current dbh. In both the stand-table­ projection method and the individual-tree-growth prediction method, stand mortality was allocated pro­ portional to the inverse of relative diameter, and was then used to compute the cumulative diameter distri­ bution at the end of each growth interval. This procedure was also used to compare ob­ served and predicted diameter distributions for 5-year growth intervals (ages 7-12 and 11-16 years) ob­ tained by using the stand-table-projection and indi­ vidual-tree growth methods. Because of difficulties in allocating 5-year mortality in the stand-table-pro­ jection method, both projection methods were evalu­ ated by estimating annual growth for 5 consecutive years. At the end of each 1-year. growth cycle, tree­ and stand-level vari bles were updated prior to per­ forming the next growth cycle. Visual inspections of observed and predicted diameter distributions were made for selected plots at each planting density. S.A. Knowe et at./ Forest Ecology and Management 98 ( 1997) 49-60 55 Table 3 Parameter estimates and standard errors for tree-level projection equations for red alder plantations near Apiary, Oregon 3. Results 3.1. Auxiliary functions The survival function (Eq. (lf)) accounted for 97.6% of the variation in observed N2 and the root 206.43 treesjha. mean square error (RMSE) was Survival trends depicted by this function indicate that little mortality occurs over time in the 976 planting density. As planting density increases, how­ ever, mortality increases so that the 6944 and the 13 889 planting densities have about the same num­ ber of surviving treesjha at age 16. The height-age function accounted for 97.1% of the variation in observed H2 and the RMSE was 0.43 m. 3.2. Diameter-distribution-prediction approach Using the parameter estimates in Table 1. the Equation a Parameter Estimate Standard error Eqs. (6) and (7) Eq. (8) Eq. (8) Eq. (8) Eq. (8) Eq. (8) {3 -0.306601 0.298965 0.577319 -0.002186 -0.009640 -0.330934 0.021452 0.139387 0.067428 0.000258 0.0003 35 0.008471 a s0 a, <52 83 0 Corresponds to equation number used in text. height growth (Eq. (lg)) indicate that stand basal area begins to converge by age 15 for planting 3086 treesjha but still di­ verges for lower planting densities. The 6944 and the 13 889 planting densities have about the same basal densities greater than area at age 1 6. which retlects the combined effects of basal area prediction function accounted for 95.5% mortality and reduced-diameter growth at the highest of the variation in observed basal area and the planting density. RMSE was 1 .2 1 m2'jha. Trends in basal area over The red alder spacing study data indicated that time (Eqs. (la) and (l h)) obtained by using the age surviving number of trees (Eq. (lf)) and dominant diameter percentile. D0• The percentile prediction was important in predicting only the minimum equations (Eqs. (!b). (lc), (1d) and (1e)) accounted 82.6 to 97.2% of the variation in observed per­ centiles and RMSE ranged from 0.42 em for D'25 to 0.86 em for D95. Increased planting density not only for Table 2 Parameter estimates and standard errors for system of stand-level prediction and projection equations for red alder plantations near Apiary, Oregon Equation' Parameter Estimate Standard error Eq. ( Ia) Eqs. (1 a) and(l h) Eqs. (l a) and(Ih) Eqs. (Ia) and ( lh) Eq. (lb) Eq. (!b) Eq. (!b) Eq. ( l c) Eq. (lc), Eq. (l d) Eq.(ld) Eq. (l e) Eq. (!e) Eq. (lf) Eq. (lf) Eq. (lf) Eq. (l g) Eq. (lg) ao a, a2 aJ 0.053966 0.325035 0.903040 0.412420 -4.324698 0.285263 0.555794 -1.449393 0.917854 -0.518221 l . O l l046 1.940934 1.175086 -2.028169 7.282x w-6 2.257404 0;185226 2.574843 0.01019 0.01494 0.09392 0.07225 0.49295 0.06029 0.06234 0.20248 0.02065 0.16503 0.01653 0.36726 0.03657 0.51245 2.138X to-s 0.71305 0.01749 0.31157 ,\10 A,, ,\12 A;o ,\21 AJo ,\3/ ,\40 ,\41 a, 82 1)3 {3, {3 ' Corresponds to equation number used in text. reduced the mean diameter, but also increased the degree of positive skewing of simulated diameter distributions, and increased the coefficient of varia­ tion in diameters. 3.3. Stand-table-projection approach The relative basal area projection functio n was based on 4183 observations and accounted for 96.4% of the variation in observed future relative size: the R.\1SE was 0.10 m1jm1. Parameter estimates and 3. standard errors are shown in Table The second component of the stand-table-projec­ tion approach is the basal area projection function (Eq. (1h)), which was based on accounted for 62 observations: it 92A% of the variation in projected 1.46 m2jha. stand basal area and the RMSE was This projection function depicts similar trajectories in basal area over time as the compatible prediction function (Eq. (la)). S.A. 56 Knowe eta/./ Forest Ecology and J\tlanagement 98 ( 1997) .J9-60 3.4. Individual-tree-growth approach ing density at age 1 6, were significantly different. For the individual-tree-growth approach, The diameter growth function was based on observations and accounted for 59.8% of the varia­ tion in annual diameter growth: the RMSE was the ob­ served and projected diameter distributions for one 3698 plot (1.6% of the total), corresponding to the 1600 treejha planting density at age 1 6 only, were signifi­ 0.25 em. Site index (SI) was not included in the final cantly different. model because a negative coefficient was estimated Visual inspections of diameter distributions for l ­ l) in a preliminary model. Logically, diameter growth or 2-year growth intervals (Fig. should increase rather than decrease with increasing diameter-distribution-prediction approach tended to revealed that the site productivity. The negative coefficient may have over-predict dbh for larger trees at all ages in stands been an artifact of the restricted range in SI and was with the possibly confounded with planting density as a con­ predict dbh for larger trees in stands with the sequence of using data from only one location. Po­ planting density (Fig. lD), and to over-predict dbh 976 planting density (Fig. lA), to under­ 6944 tential growth depicted by the diameter growth equa­ for smaller trees in stands with the 1 3 889 planting tion indicates that the maximum annual growth of density (Fig. IE). In the 976 planting density (Fig. 4.3 em/year is attained for trees with 1 1 -cm dbh. 1 A), the stand-table-projection approach tended to Only under-predict dbh for smaller trees at younger ages. 50% of the potential diameter growth is ex­ pressed for trees with 7 to 11 trees in stands with more than m:! /ha BAL or for 5 However, in the m2jha basal area. (Fig. ID and 6944 E). over-predicted at age 3.5. Evaluation of methods Observed were by planting densities 13 889 16. The individual-tree-growth approach provided the best representation of ob­ served diameter distributions at all planting densities and predicted diameter distributions compared and diameters for smaller trees were using two-sample and at all ages (Fig. 1 ). Kol­ For the 5-year growth intervals. the predicted and For the l - and observed diameter distributions for three plots ( 17.6% 2-year growth intervals, the predicted and observed of the total) obtained by using the stand-table-projec­ mogorov-Smirnoff test the (a = 0.05). diameter distributions obtained by using the diame­ tion method were significantly different (Table 4). ter-distribution-prediction approach for three plots The rejected plots included one for the (4.8% of total) were significantly different (Table These correspond to the density at ages 1 4 ( 1 plot) 13 899 and 16 4). density in projections from ages treejha planting one each for the (2 plots). For the 4360 7 and 1 3 in projections from ages 1 1 976 planting to 1 2 years and planting densities 889 to 16 years. Significant stand-table-projection approach. the observed and differences between the predicted and observed di­ projected diameter distributions for one plot ( 1.6% of ameter distributions were detected for only one plot the total), corresponding to the 13 899 treejha plant­ (5.9% of the total) obtained by using the individual- Table+ Number of occurrences with a significant difference between the observed and predicted diameter distribution, · two-sample Kolmogorov-Smirnoti test as determined by the ( a= 0.05), for the diameter-distribution-prediction method, the stand-table-projection method, and the individual-tree-growth method developed for red alder plantations near Apiary, Oregon Number of rejected distributions (percent of rejected distributions Method 1- or Diameter-distribution-prediction Stand-table-projection Individual-tree-growth • Percentages are based on 2-year interval 3(+.8%) 1(1.6%) I( 1.6%) (n = 62) 5-year interval •) (n = 17) 3(.17.6%) 1(5.9%) 62 observations with 1- or 2-year growth intervals and 17 observations with 5-year growth intervals. S.A. Knowe era{. j Foresr Ecology and Management 98 ( 1997) -19-60 Visual inspections of predicted and observed di tree growth method. This plot corresponded to the 1600 planting 16 years. density and projected from ages 11-12 9- 0 7-6 1.0 0.9 --- Observed 0.8 --- We1bull prediction 0.7 --- Stand Iailie ,., Stand Iallie c 1l 0 OrQJectfOO " cr 0.6 ------ lndiVIdual·tree growtn " g 0.4 3 (.) " E " 0.3 0 0.2 4 2 6 10 I4 12 18 16 dbh (em} Projection B. 0.8 --- Wetbud prediction --- Stano :aOla :0 cr " 0.6 ------ lndivldual·tree " 0.5 " ; '3 " :; (.) .orojectlon growth 0.4 0.3 0 >nterval 14-16 1.0, t 0.91 growtn . ; :; 0.3 (.) 0.2 -- 14 16 18 20 22 24 16 Pro1ectton 9·10 18 20 interval 11-12 4-16 Observed Wetbuil predlctton Stand table pTOJ8CtiOO growth 0.4 0.3 0.2 0.1 12 14 ------ !nlj1v1dual·lree 0.0 +-...... 0 2 10 12 7·8 0.0 4 10 0. 0.1 2 a 6 dbh (em) 0.81l --- " 0.5 0.5 'I> 4 2 j ,.. " 0.7 .... ___ c D ::J cr pro1ection 0.4 0 14·16 ------ lndivtdual·tree 0.5 0.0 22 ---Observed 0.7 ,.. 20 11·12 9-10 7·8 c <.J interval 9·10 11-12 0.1 0.0 0 0.9 Projection 0.2 0.1 1.0 stand table projection 7-8 ---Observed 0.5 the c. --- Weibull prediction cr 0.6 ;:: that 0.8 ,., --" 0.7 "' " 3 indicated 0.9 c 2) interval Projection "' > ameter distributions for the 5 year growth intervals (Fig. A. 1.0 to 11 57 ;._.:::;;,.:Je!...,.;::..Z. : 6 a --,..--10 12 ------, 14 :6 18 dbh (em) abh (em) Projection .01 0.9 0.8 ] -- Observed --- Weibutl prediction "jJ --- interval 11-12 14·16 Stand Iallie iJ' 0. 7 5- 0.61------ lndivtdual· ; 9·10 7·8 E. P!Oiecttor\ i ::l' -···:: o.3 (.) 02 . 0.1 1 v' · f·' 0.0 +-��o:!!!:::,.....::::-.,--.:;...r--_..,�--,�-----'r�--, 6 0 s 10 12 14 16 18 20 abh (em) Fig. 1. Comparison of observed and predicted diameter distributions obtained by a diameter-distribution-prediction method, a stand-table­ ·(A) 976 planting density: (B) 1600 planting (C) 3086 planting densil)'; (Dl 6944 planting density: :1nd (E) 13889 planting density. Note differences in X-uxis and projection projection method. and an individual-tree-growth method and 1- or 2-year growth intervals: density: intervals. S.A. 58 Knowe et a/./ Forest Ecology and Management 98 ( /997) 49-60 tree-growth method tended to over-predict dbh for smaller trees in young stands and at high planting densities, and to under-predict dbh for smaller trees in older stands and at low or high planting densities. method tended to under-predict dbh for smaller trees in young stands and at low planting densities, and to over-predict dbh for larger trees in older stands and at intermediate planting densities. The individual- Projection A. 1. 0 0.9 --- ObseiVed 0.8 --- Stand table protection c: 0.7 " ------ }' (I' lndtvldual·tree grow<h " :J E 0.4 .<: (.) :J 0' 0.6 " 0.5 :; E 0.4 )/ /) I,J :J !I j., 0.3 (.) :J /l 0.2 0.2 _; I ./_../ 0.1 ___ 0. 0 0 2 4 6 j..../ ,...1 8 10 12 0.1 0.0 1 4 16 dbh (em) 18 20 Projection B. 7-12 22 >. 0 " > · § (.) 0.9 ---Observed --- Stand lal>le PIOiectJon :; 0.7 - Vi '0 " 0.6 :g r:r 0.6 0.5 0.4 16 18 Projection 20 interval t 1-16 - - - Individual-tree growth -- 0.5 - 0.4 E 0.3 ;; :; 0.3 14 7-12 0.8 c: - :J 12 o. 1. 0 Stand table projection 10 8 6 interval 11-16 0.7 ;; 4 dbh (eml --- CbseiVed 0.8 2 0 1. 0 0.9 Individual-tree growth > ,I I 0.3 0.8 0.7 c: d j} --- Obs!!IVed --- Stand table projedion interval 11·16 7-12 0.9 " " >. • r:r 0.6 0.5 c. 1. 0 LJ d :J Projection interval 7-12 (.) :J 0.2 0.2 0.1 0.1 0. 0 0 0. 0 2 4 6 10 12 14 16 18 20 22 24 ;' 0 4 2 6 8 10 12 14 16 18 dbh (em) dbh (em) Projection E. 1. 0 7·12 0.9 -- Obs8M>d 0.8 -- Stand table pcojection 51 0.7 ------ lndlviduaHree growth r:r 0.6 >. :; :J " .<: ;; :; E (.) :J interval 11·16 0.5 0.4 0.3 0.2 0.1 0. 0 0 2 4 6 a 10 12 14 16 18 20 dbh (em) Fig. 2. Comparison of observed and predicted diameter distributions obtained by a diameter-distribution-prediction method, a stand-table­ projection method. and an individual-tree-growth method and 5-year growth intervals: (A) 976 planting densicy: (B) 1600 planting density: (C) 3086 planting density; (D) 6944 planting density; and (E) 13 889 planting density. Note differences in X-axis. S.A. Knowe et al. /Forest Ecology and Management 98 (}997) 49-60 4. Discussion and conclusions A yield system for red alder was not previously available because of lack of long-tenn data for vari­ ous planting densities. The predicted stand tables in all three modeling methods depicted increases in both mortality and reductions in tree diameters as planting density increased. Because limited data were available for fitting the equations, the system should be viewed as a framework for improvement. The diameter-distribution-prediction approach tends to over-predict dbh for larger trees in stands planted at low density and to under-predict dbh for smaller trees planted at high density. Adding an expression of density to the minimum diameter percentile ( D0) and 95th percentile (D95) may improve the accuracy of the diameter-distribution-prediction approach. However, this approach does give a reasonable rep­ resentation of the smoothed diameter distribution. and may be useful for comparing planting densities when a tree list is not available. The stand-table-pro­ jection approach tends to under-predict dbh for smaller trees in young stands planted at low density and to over-predict dbh for smaller trees in older stands planted at high density. This approach pro­ vides for consistency between stand- and tree-level growth projections, and should be useful for compar­ ing planting densities when a tree list is available. The individual-tree-growth approach provides the best representation of observed diameter distribu­ tions at all planting densities. This approach may be best suited for stands that have been thinned, stands with mixtures of species, or stands with various size classes. It should be noted that stand-level models have been developed to incorporate the effects of thinning (Pienaar and Shiver, 1986), site preparation (Knowe and Stein, 1995; Pienaar and Rheney, 1995), release from competition (Knowe et al., 1992; Knowe, 1994), and clonal mixtures (Knowe et al., 1994). Our results were consistent with those reported for similar studies in loblolly pine (Pinus taeda L. ) and young Douglas-fir (Pseudotsuga menziesii). In comparisons of modeling approaches, Borders and Patterson ( 1990) and Kno we (1994) obtained a better representation of diameter distributions when using the stand-table-projection approach than when using diameter-distribution-projection and prediction meth­ 59 ods. Knowe and Stein (1995) reported only a slight improvement in representation of diameter distribu­ tions obtained by using the stand-table-projection approach as compared with the diameter-distribu­ tion-prediction method. As in the current study, Borders et a!. (1988) noted decreasing precision with increasing growth interval for stand-level projection models similar to the functions developed for red alder plantations. Also, Knowe (1994) reported similar decreases in precision for a stand-table-projection method devel­ oped for young Douglas-fir plantations with interspe­ cific competition. However, Knowe and Stein ( 1995) did not observe appreciable differences in precision for 1-, 2-, and 3-year growth intervals in young Douglas-fir. Knowe (1994) and Knowe and Stein (1995) reported that more significant differences be­ tween predicted and observed diameter distributions were detected at younger ages than at older ages when the diameter distribution prediction method was compared with the stand-table projection ap­ proach. In the present study, we found more signifi­ cant differences in distributions at older ages and at extremes in planting density. Knowe and Hibbs ( 1996) observed consistent re­ lationships between parameters of growth projection models and planting density of young red alder in Neider plots. We did not observe similar relation­ ships in the present study. The narrower range of planting densities in the present study (976-13 889 trees/ha vs. 238-10 118 treesjha in the Neider plots) may have obscured consistent relationships. Also, the nonindependence of observations in Neider plots may impart consistent and systematic variation in parameters of growth models developed by using data from Neider plots. References Ahrens, G.R., Dabkowski. A., Hibbs, D.E.. 1992. Red alder: guidelines for successful regeneration. Spec. Pub!. 24, Forest Research Laboratory,-Oregon State University, Corvallis. OR, ll pp. Bailey, R.L.. Dell. T.R., 1973. Quantifying diameter distributions with the Weibull function. For. Sci. 19, 97-104. Bailey, R.L., Burgan', T.M., Jokela. E.J., 1989. Fertilized midrota­ tion-aged slash pine plantations-stand strucrure and yield pre­ diction models. South. J. Appl. For. 13, 76-80. S.A. Knowe et at. / Forest Ecology and Management 98 I 1 997) -19-60 60 Binkley, D .. Cromack, K .. Jr., Baker, D.D .. 1 994. Nitrogen tixa­ tion by red aider: biology, rates, and controls. In: Hibbs, D.E., DeBell, D.S., Tarrant, R.F. (Eds.), The Biology and Manage­ ment of Red Alder. Oregon State University Press, Corvallis. OR, pp. 57-72. Borders, B.E .. 1 989. Systems of equations in forest stand model­ ing. For. Sci. 35 548-556. Borders, B.E., Patterson, W.D.. 1 990. Projecting stand tables: a comparison of the Weibull diameter distribution method, a percentile-based projection method, and a basal area growth projection method. For. Sci. 36, 413-42 4. Borders, B.E., B ailey, R.L., Clutter, M.L.. 1 988. Forest growth models: parameter estimation using. real growth series. In: IUFRO Forest Growth Modeling and Prediction Conference, 24-28 August 1 987, Minneapolis, MN, pp. 660-667. Clutter, J.L.. Jones, E.P., 1 980. Prediction of growth after thinning old-field slash pine plantations. Res. Pap. SE-2 1 7, USDA Forest Service, Southeastern Forest Experiment Station, Asheville. NC, 14 pp. Gregoire, T.G .. 1 987. Generalized error structure of forestry yield models. For. Sci. 33 , 423-444. Harrington. C.A., Curtis, R.O.. 1 986. Height growth and site index curves for red alder. Res. Pap. PNW-358. USDA Forest Service, Pacific Northwest Research Station. Portland, OR. 14 pp. Hester, A.S., Hann. D.W., Larsen, D.R. 1 989. ORGANON: southwest Oregon growth and yield model user manual. Ver­ :;ion 2.0. Forest Research Laboratory. College of Forestry. Oregon State University, Corvallis. OR. 59 pp. Hibbs. D.E.. 1 987. The self-thinning rule and red alder manage­ ment. For. Ecol. Manage. 18. 273-28 1. Hibbs, D.E., Carlton, G.C . 1 989. A comparison of diameter- and , . . volume-based stocking guides for red alder. West. J. Appl. For. +. 1 1 3- 1 1 5 . Hibbs. D . E.. DeBell. D.S., 1 994. Management o f young red alder. In: Hibbs. D.E., DeBell. D.S., Tarrant. R.F. (Eds.). The Biol­ ogy and Management of Red Alder. Oregon State University Press. Corvallis. OR. pp. 202-2 1 5 . Hibbs, D.E., DeBell, D.S., Tarrant, R.F. (Eds.), 1 994. The Biol­ ogy and Management of Red Alder. Oregon State University Press. Corvallis, OR, 256 pp. Hibbs, D.E., Emmingham. W.H., Bondi. M.C., 1 989. Thinning red alder: effects of method and spacing. For. Sci. 35, 1 6-29. Hibbs, D.E., Ahrens, G.R.. B uermeyer, K.. Giordano, P.A .. 1 993. Hardwood Silviculture Cooperative: Annual Report, 1992-93. Forest Research Laboratory, College of Forestry, Oregon State University, Corvallis, OR. 17 pp. Knowe, S.A.. 1 994. Incorporating the effects of interspecific competition and vegetation-management treatments in stand table projection models for Douglas-fir saplings. For. Ecol. Manage. 67. 37-99. Knowe. S .A . . Hibbs, D.E . 1 996. S tand structure and dynamics of young red alder as affected by pl:lnting density. For. Ecol. Manage. 82. 69-85. Knowe, S.A .. Stein, W.I.. 1 995. Predicting the effects of site preparation and protection on development of young Douglas­ fir plantations. Can. J. For. Res. 25, 1538- 1547. Knowe. S.A., Harrington, T.B.. Shula. R.G., 1 992. Incorporating the effects of interspecific competition and vegetation manage­ ment treatments in diameter distribution models for Douglas-fir saplings. Can. J. For. Res. 22. 1255- 1 262. Knowe, S.A Foster, G.S., Rousseau, R., Nance. W.A.. 1994. Eastern cottonwood clonal mixing study: predicted diameter distributions. Can. J. For. Res. 24. +05-4 14. Pienaar. L.V .. Harrison. W.M.. 1 988. A stand table projection approach to yield prediction in unthinned even-aged stands. For. Sci. 34. 304-808. Pienaar, LV .. Rheney, J.W .. 1 995. Modeling stand level growth and yield response to silvicultural treatments. For. Sci. + l . 629-638. Pienaar. L. V.. Shiver. B .D.. 1 986. Basal area prediction and projection equations for pine plantations. For. Sd. 32. 623­ 633. Puettmann. K.J.. Hibbs. D.E., Hann, D. W., 1 992. The dynamics of mixed stands of Alnus ntbra and Pseudotsuga men::.iesii: extension of size-density analysis to species mixture. J. E�.:ol. 80. +49-458. Ruettig. T.R., Connaughton, K., Ahrens, G.R.. 1995. Hardwood supply in the Pacific Northwest: a policy perspective. Res. Pap. PNW-+78, USDA Forest Service, Pacitic Northwest Re­ search Station, Portland, OR, 80 pp. Richards. F.J., 1 959. A flexible growth function for empirical use. J. Exp. Bot. 10, 290-300. Sokal. R.R .. Rohlf, F.J., 1 98 1 . Biometry: The Principles and Practice of Statistics in Biological Research, 2nd edn. W.H. Freeman and Co., San Francisco, CA. 859 pp. Stage, A.R .. Wykoff, W.R.. 1 993. Calibrating a model of stochas­ tic effects on diameter increment for individual-tree simula­ tions of stand dynamics. For. Sci. 39, 692-705. Tarrant, R.F.. Trappe, J.M., 1 97 1 . The role of Alnus in improving the forest environment. Plant Soil 22, 335-348. Tarrant, R.F., Bormann, B.T., DeBell. D.S., Atkinson, W.A., 1 983. Managing red alder in the Douglas-tir region: some possibilities. J. For. 8 1 , 787-792. Wykoff. W.R 1 990. A basal area increment model for individual conifers in the northern Rocky Mountains. For. Sci. 36. 1 077­ 1 104. . .• .•