Kyriaki Kitikidou*, Argyro Papageorgiou, Elias Milios, Athanasios Stampoulidis

Silva Balcanica, 15(1)/2014 A bark thickness model for Pinus halepensis in Kassandra, Chalkidiki (Northern Greece) Kyriaki Kitikidou*, Argyro Papageorgiou, Elias Milios, Athanasios Stampoulidis Democritus University, Department of Forestry and Management of the Environment and Natural Resources Orestiada Abstract Timber value cannot be appropriately estimated without knowing bark thickness. In this study, an S-curve model to estimate bark thickness of Pinus halepensis was developed. Data from forty, randomly selected trees were collected from Kassandra, Chalkidiki (Northern Greece), and ten regression models were compared with ten statistical criteria. Results showed that the S-curve model provided the most realistic estimates of bark thickness, even though other models that have been tested gave good values for the criteria used for models comparison. Key words: bark thickness, diameter at breast height, Pinus halepensis, regression models. INTRODUCTION Estimating bark thickness is a significant component of the research of forest growth and yield. Timber volume and prize cannot be correctly estimated without knowing the thickness of bark (Meyer, 1946). Whilst bark accounts for a small percentage of tree volume, managers acknowledge that the bark they buy when prize estimations are based on standing trees, carries some weight to their profit (Marden et al., 1975). For many years, bark had been an unwanted by-product, since its disposal, typically through burying or burning, often increases the cost of milling (Haygreen, Bowyer 1996). Bark has been used for centuries, on a small scale, for medicinal purposes, food, baskets, boats, and tannins (Small, 1884). The most basic use of bark is to produce energy or heat through burning. The outside bark volume of standing trees can be estimated from measurements of their height and diameter at breast height (DBH) outside bark. With bark thickness models, the volume of wood inside the bark can also be estimated; therefore, models that provide accurate estimates of the amount of bark in standing trees are useful (Farr, 1967). Bark thickness varies by tree species. For example, white spruce (Picea glauca Moench) and black spruce (Picea mariana Mill.) are considered thin-bark species with average thickness of 0.6 to 1.3 cm (Viereck, Little 1972), while white fir (Abies concolor Gord. and Glend.) is a thick-bark species with an average bark thickness of 10.2 to 17.8 cm (Harlow et al., 1996). Ne eman et al. (2004) report a bark thick- 47

ness for Aleppo pine (Pinus halepensis) of 3.8 cm on average, while Voulgaridis et al. (1985) measured a bark thickness range of 2 to 2.5 cm, for 45 to 50 years old trees. Models for estimation of bark thickness from DBH have not been reported for Aleppo pine in Greece so far. The objective of this study was to develop a model to estimate bark thickness of Aleppo pine in Greece. Aleppo pine is the dominant tree of a large percentage (26%) of coniferous forests in the country, and it is very well adapted to fire (Daskalakou, Thanos 1996). A bark thickness model was developed based on sampled trees from Kassandra peninsula (Northern Greece). MATERIALS AND METHODS Forty trees were selected from two site types found in an area that burned in 1990, within Kassandra peninsula (Fig. 1). The burned area was 575 ha, and 386.5 of them were forested (data from Forest Service of Kassandra). Twenty experimental plots were randomly located, in two site types, in areas with a cover, created by the tree canopy projection, which is over 80% (Papageorgiou, 2011). Site types were determined as follows: site type A, where the soil depth was 50 to 55+ cm, and site type B, where the soil depth was 35 to 40 cm. In each site type 10 plots of 25 m² (5mx5m) were established randomly. Site type A represents the productive sites, while site type B represents the medium productivity sites (Papageorgiou, 2011). Fig. 1. Study area (source: Google Maps <https://maps.google.com/>). In each plot, among other measurements, the dominant and co-dominant trees were determined and one dominant and one co-dominant tree were sampled, during the winter of 2010-2011, using the stratified sampling method (Matis, 2004). The characterization of trees as dominant or co-dominant was based on Kraft classification of tree crowns (Oliver, Larson, 1996] Smith et al., 1997). As dominants are characterized the trees with crowns extending higher than the general level of the 48

crown stratum, but are possibly somewhat crowded from the sides by other trees. On the other hand, co-dominants were considered the trees that form with their crowns the general level of the crown cover, they are not restricted from above, but they have their crowns more or less crowded by other individuals from the sides. Smith et al. (1997) and Oliver, Larson (1996) provide analytical definitions of the terms dominant and co-dominant. From each sampled tree, cross-sectional discs were cut and removed from various heights. In each cross-sectional disc measurements were made in the frame of other research studies. For the present study, at the height of 1.3 m, the mean diameter inside and outside bark were measured. Table 1 provides summary statistics for the dataset. Table 1. Summary statistics of the sampled trees Mean deviation Min Max Age (years) 19.90 1.52 16.00 22.00 Height (m) 8.30 2.03 5.20 12.30 DBH outside bark D (cm) DBH inside bark d (cm) Bark thickness BTh (cm) = (D-d)/2 9.72 3.94 3.40 19.50 8.27 3.02 3.25 17.45 0.72 0.55 0.01 2.45 Data suggest that, generally, bark thickness increases as DBH increases, following a trend that could be either linear or curve, as later shown in Fig. 2. That is why we tested the following ten regression models for fitting (Arlinghaus, 1994): 1. Linear 2. Logarithmic 3. Inverse 4. Quadratic 5. Cubic 6. Power 49

7. Compound 8. S-curve 9. Logistic where u = upper boundary value = max BTh rounded up = 3.00 10. Growth where: : estimated bark thickness (cm) b i (i=0 3): regression coefficients D: DBH (cm) As typical in statistical inference, firstly hypotheses testing and statistical criteria were used for model comparison; then, biological rationale was taken into account for the final model selection. Ten criteria were used for models comparison (Kitikidou, 2005) (Table 2). Model selection included several consecutive steps. Firstly, we checked the significance of regression coefficients; then we calculated the comparison criteria; finally, we checked the biological aspects and made a decision. RESULTS Regression coefficients of all models were significant, except for two (p-value >0.05, Table 3). Comparison criteria values for all models, except for the two excluded from the results of Table 3 are given in Table 4 (best values for each criterion are highlighted). Even thought models 1 (linear), 2 (logarithmic) and 3 (inverse) gave good values for the criteria used for models comparison, they gave unrealistic (negative) values of bark thickness in certain DBHs. The S-curve model had clearly better values for all criteria, except for 5 (sum of squared errors) and 9 (variance ratio). The selected bark thickness model for Pinus halepensis was:. The S-curve is illustrated in Fig. 2. 50

Table 2. Comparison criteria N Criterion Formula Optimum value 1 Absolute mean error 0 2 error of the estimate min 3 Coefficient of determination R 2 1 4 Root of the mean squared error min 5 Sum of squared errors 0 6 Sum of relative squared errors 0 7 Relative mean squared error % 0 8 Average deviation 0 9 Variance ratio 1 10 Regression coefficients α, β α=0, β=1 Abbreviations: : observed values of bark thickness (i = 1-40); : estimated values of bark thickness; n - number of observations (trees); p: number of regression coefficients; : average of observed bark thicknesses; : average of estimated bark thicknesses 51

Fig. 2. S-curve fitting for bark thickness estimation. DISCUSSION We developed a model to estimate bark thickness of Aleppo pine in Kassandra peninsula (Northern Greece). Ten regression models were tested and S-curve model was found to be more appropriate, based on ten comparison criteria. This model is able to estimate Aleppo pine bark thickness within a wide range of diameters from 3.25 to 17.45 cm. A comparison between the selected S-curve model of this study, and Rigolot s linear model (Rigolot, 2004), shows that the linear model overestimates bark thickness for D<12 cm, while there is an underestimation for D>12 cm (Fig. 2). The model developed in this study could be used to estimate bark thickness of Aleppo pine at various tree heights and to address wood volume along the stem inside bark. Meyer (1946) used DBH and bark thickness measurements to estimate volume along the stem inside bark of various tree species, and the results showed that the predicted volumes inside bark were close to the observed values. Researchers have attempted to improve bark thickness models by using various transformations of DBH (quadratic, exponential), or adding other independent variables such as tree age, height, and site characteristics (Hale, 1995, Dimitrov, 1976). This study considered DBH as the only necessary predictor, because research on bark thickness of coniferous trees shows that bark thickness is linearly correlated to the DBH and the ratio of diameter inside bark to outside bark is close to constant along the stem (Kozak, Yang, 1981). As a follow up of this study, the selected model s validation is currently under development by the authors. Acknowledgments: We would like to thank Kassandra Forestry Service for their cooperation and Bodossaki Foundation for the financial support of the 4th author, Mr. Stampoulidis. 52

Table 3. Parameter estimates and their significances and standard errors for the compared models Model b 0 error 1. Linear -0.474 2. Logarithmic 3. Inverse 4. Quadratic -1.780 1.699-0.666 5. Cubic 0.015 6. Power 0.022 7. Compound 8. S- curve 9. Logistic 10. Growth 0.209 1.500 9.237-1.565 0.110 0.237 0.135 0.263 (0.016) 0.562 (0.979) 0.009 (0.003, 0.041) 0.042 (0.124, 0.294) 0.158 (1.180, 1.820) 2.736 (3.697, 14.776) 0.200 (-1.971, -1.160) b 1 0.123 1.142-8.022 0.164-0.061 1.515 1.126-17.533 0.809 0.119 error 0.010 0.106 1.015 0.052 (0.003) 0.173 (0.724) 0.161 (1.188, 1.842) 0.015 (1.095, 1.157) 1.955 (-21.491, -13.576) 0.019 (0.770, 0.847) 0.014 (0.091, 1.146) b 2 error -0.002 0.002 (0.426) 0.016 0.020 (0.224) * In brackets: significance level p for models 1 5; confidence interval for models 6 10. b 3 error -64 10-5 47 10-5 (0.180) References Arlinghaus, S.L. 1994. PHB Practical Handbook of Curve Fitting. CRC Press, USA, 249. Daskalakou, E.N., Thanos, C.A. 1996. Aleppo pine (Pinus halepensis) postfire regeneration: the role of canopy and soil seed banks. International Journal of Wildland Fire, 6(2), 59-66. Dimitrov, E.T. 1976. Mathematical models for determining the bark volume of spruce in relation to certain mensurational characteristics. Forestry Abstracts, 37, 62 81. Farr, W.A. 1967. Board-foot tree volume tables and equations for white spruce in interior Alaska. Research Note PNW-59, USDA Forest Service, 4. Hale, J.D. 1955. Thickness and density of bark trends of variation for six pulpwood species. Pulp and Paper Magazine of Canada, 56(13), 113 117. Harlow, W.M., Harrar, E.S., Hardin, J.W., White, F.M. 1996. Textbook of Dendrology. 8th edition. McGraw-Hill, New York, USA, 543. Haygreen, J.G., Bowyer, J.L. 1996. Forest products and wood science. 3rd edition. Iowa State University Press, Ames, Iowa, USA, 484. 53

Table 4. Results via comparison criteria Criterion Optimum Model 1. Linear 2. Logarithmic 3. Inverse 6. Power 7. Compound 8. S- curve 9. Logistic 10. Growth 1 2 3 4 5 6 7 8 9 10 0 min 1 min 0 0 0 0 1 0 1 0.178 0.258 0.784 0.252 2.5 1948.6 4871.5 24.605 0.784 0.000 1.000 0.195 0.277 0.752 0.270 2.9 2393.3 5983.2 26.988 0.752 0.000 1.000 0.238 0.342 0.622 0.333 4.4 4438.0 11094.9 32.922 0.622 0.000 1.000 0.188 0.272 0.760 0.265 2.8 3823.1 9557.9 25.397 0.679-0.061 1.061 0.229 0.315 0.679 0.307 3.8 7411.8 18529.4 30.729 0.562-0.102 1.106 0.180 0.254 0.791 0.247 2.4 1707.1 4267.6 24.783 0.775-0.011 1.011 0.190 0.278 0.751 0.271 2.9 4383.9 10959.6 25.617 0.734-0.031 1.012 0.229 0.315 0.679 0.307 3.8 7411.8 18529.4 30.729 0.562-0.102 1.106 Hutchison, K.O. 1967. Alaska s Forest Resource. Resource Bulletin PNW-19, USDA Forest Service, Juneau, Alaska, USA, 74. Kitikidou, K. 2005. Applied statistics with use of the SPSS statistical package. Tziola Publications, Thessaloniki, Greece (In Greek), 288. Kozak, A., Yang, R.C. 1981. Equations for estimating bark volume and thickness of commercial trees in British Columbia. Forestry Chronicle, 57(3), 112 115. Marden, R.D., Lothner, D.C., Kallio, E. 1975. Wood and bark percentages and moisture content of Minnesota pulpwood species. Research Paper NC-114, USDA Forest Service, 9. Matis, G. 2004. Sampling of natural resources. Pigasos 2000 Publications, Thessaloniki, Greece (in Greek), 525. Meyer, H.A. 1946. Bark volume determination in trees. Journal of Forestry, 44, 1067 1070. Ne eman, G., Goubitz, S., Nathan, R. 2004. Reproductive traits of Pinus halepensis in the light of fire a critical review. Plant Ecology, 171, 69-79. Oliver, C.D., Larson, B.C. 1996. Forest Stand Dynamics. John Wiley & Sons, Inc. New York, 544. Papageorgiou, A. 2011. Social differentiation analysis of trees in P. halepensis stands which were initiated after the fires of 1990 and 1998 in the Peninsula of Kassandra in Chalkidiki, Greece. MSc Thesis, Democritus University, Department of Forestry and Management of the Environment and Natural Resources (In Greek), 66. Rigolot, E. 2004. Predicting postfire mortality of Pinus halepensis Mill. and Pinus pinea L.. Plant Ecology, 171(1-2), 139-151. 54

Small, H.B. 1884. Forest trees, timber, and forest products. Dawson Brothers, Montreal, Canada, 64. Smith, D.M., Larson, B.C., Kelty, M.J., Ashton, P., Mark, S. 1997. The practice of silviculture. Applied Forest Ecology. John Willey & Sons, Inc. New York, 537. Viereck, L.A., Little, E.L. 1972. Alaska trees and shrubs. Agriculture Handbook no. 410, USDA Forest Service, 265. Voulgaridis, E., Grigoriou, A., Passialis, C. 1985. Investigations on bark extractives of Pinus halepensis Mill. Holz als Rob und Werkstoff, 43, 269-272. *Corresponding author: kkitikid@fmenr.duth.gr 55