Resource communication. Individual tree growth model for radiata pine plantations in northwestern Spain

Instituto Nacional de Investigación y Tecnología Agraria y Alimentaria (INIA) Available online at www.inia.es/forestsystems http://dx.doi.org/10.5424/fs/2012213-03109 Forest Systems 2012 21(3), 538-542 ISSN: 2171-5068 eISSN: 2171-9845 Resource communication. Individual-tree growth model for radiata pine plantations in northwestern Spain F. Crecente-Campo1, *, U. Dieguez-Aranda1 and R. Rodriguez-Soalleiro2 1 Departamento de Ingeniería Agroforestal, Universidad de Santiago de Compostela. Escuela Politécnica Superior, Benigno Ledo s/n, Campus universitario, 27002 Lugo, Spain 2 Departamento de Producción Vegetal, Universidad de Santiago de Compostela. Escuela Politécnica Superior, Benigno Ledo s/n, Campus universitario, 27002 Lugo, Spain Abstract Individual-tree basal area and height increment models were developed with data from 130 permanent plots of Pinus radiata D. Don located in Galicia (northwestern Spain). Mixed-models techniques were used for model fitting. Covariates acting at tree and stand level were included as fixed effects. Estimated values of stand variables obtained from aggregation of individual-tree predictions were used in model evaluation. The developed models accounted for 54% of the variability in basal area increment and 36% of the variability in height increment, with mean errors of 16 cm2 and 0.36 m, respectively. These models, along with an existing individual-tree mortality model, constitute a whole individual-tree growth model that can be used to simulate forest management alternatives, helping in forest managers’ decision making. Key words: mixed-effects modeling; Pinus radiata; Galicia. Resumen Comunicación de recurso. Modelo de crecimiento de árbol individual para plantaciones de pino radiata en el noroeste de España En este estudio se han desarrollado modelos de incremento en sección normal y altura de árbol individual, utilizando datos de 130 parcelas de Pinus radiata D. Don localizadas en Galicia. La técnica de modelos mixtos se utilizó en el ajuste de los modelos. Se incluyeron covariables que actúan a nivel de árbol y a nivel de rodal como efectos fijos en los modelos. En la evaluación de los modelos se utilizaron estimaciones de variables de rodal obtenidas por agregación de las predicciones de árbol individual. Los modelos desarrollados explicaron el 54% de la variabilidad en el incremento en sección normal y el 36% de la variabilidad en el incremento en altura, con errores medios de 16 cm2 y 0.36 m, respectivamente. Estos modelos, junto al modelo de mortalidad de árbol individual existente, forman un modelo de crecimiento de árbol individual que puede utilizarse para simular alternativas de gestión forestal, ayudando a la toma de decisiones de los gestores forestales. Palabras clave: modelo de efectos mixtos, Pinus radiata D. Don; Galicia. * Corresponding author: [email protected] Received: 06-05-12. Accepted: 01-08-12. Introduction Radiata pine (Pinus radiata D. Don) is the most commonly used exotic conifer in reforestations in Spain, especially in northern Spain, where it covers approximately 270,000 ha (DGCN, 1998). In Galicia (northwestern Spain) it is estimated to cover 90,000 ha (Xunta de Galicia, 2001). Related studies mainly refer to stand growth and yield and to ecological and silvicultural aspects. Yield tables (Sánchez et al., 2003) and a dynamic stand growth model (Castedo-Dorado et al., 2007) have been developed for Galicia. Individual-tree growth models provide more detailed information than other modeling approaches (Gadow and Hui, 1999; García, 2003) and usually perform better than stand growth models for short term projections (Burkhart, 2003). The objective of the present study was to develop distance-independent individual-tree basal area and height increment models for even-aged radiata pine 539 Individual-tree growth model for radiata pine plantations in northwestern Spain means of visual analysis of studentized residuals. If necessary, weighted regression was used. The weak autocorrelation in the remeasured data was ignored because the impact of variance underestimation is likely to be masked by fitting each individual tree as an independent observation (Temesgen and Gadow, 2004). Model comparison Three statistical criteria were used for model comparison: root mean square error (RMSE), the model efficiency (EF), similar to the coefficient of determination for linear regression, and Akaike’s information criterion (AIC, Akaike, 1974).Plots of observed against predicted increments and plots of residuals against predicted increments were also used for model comparison and selection. Validation was not carried out because an independent dataset was not available, and other validation methods seldom provide any additional information compared with the fitting statistics (Kozak and Kozak 2003). Anyway, the developed equations where used together with the individual-tree mortality equation for the species in the region (Crecente-Campo et al., 2009) to make simulations using the fitting data. As the accuracy of the system (i.e., the individual-tree growth model) is more important than the accuracy of the individual equations, real values of stems per hectare (N) and stand basal area (G, m2ha–1), and the diameter and height distributions in the second and three inventories were compared with the predicted values. The Kolmogorov-Smirnov test was used to compare real and predicted diameter distributions. Different cut-off values were tested to convert mortality estimations to live/ dead trees (Crecente-Campo et al., 2009). Results and discussion The models that performed best were: ig d G t BALMOD BAR= + + α α α α α α 0 3 4 5 1 2 exp ( ) [1] ih h d SI t RBA D = + − α α α α α α 0 4 5 1 2 3 exp( ) [2] where ig is the annual basal area increment (cm2), BALMOD is the competition index developed by Schröder and Gadow (1999), BAR is a basal area ratio (100 · g/G, where g is the basal area of the tree –m2), ih is the annual height increment (m), SI is the site plantations in Galicia (northwestern Spain) to be combined with an existing mortality model and form a whole individual-tree growth model. Methodology Data 130 plots located in Galicia, representing the existing range of ages, stand densities and sites, were used (stand basal area –G– ranging from 6.6 to 70.6 m2ha–1, stems per hectare –N– ranging from 192 to 3,936, dominant height –H– ranging from 6.8 to 35.2 m, and age –t– ranging from 8 to 47 years). Diameter at breast height (d, cm, 1.3 m above ground) was measured in all the trees in each plot and total tree height (h, m) was measured in 30 randomly selected trees and in the dominant trees (the proportion of the 100 thickest trees per ha). 79 plots were measured in two occasions (1996 and 1999) and 51 plots were measured in three occasions (1996, 1999 and 2005). As the re-measurement interval was variable (with either 3 or 6 years between successive inventories), the method of calculating the average annual growth was used to calculate the independent variables (annual basal area and height increments). Models Zeide (1993) showed that many growth equations can be decomposed into two components when they are consider in the differential form, leading to simple basic models. These basic models were proposed for fitting to the database. The inclusion of stand and tree variables and competition indices in these basic forms was tested using linear transformations and OLS techniques. Those alternatives with a variance inflation factor larger than 10 were rejected (Myers, 1990). Because of the specific correlation structure of the experimental data (trees grouped in plots), the basic assumption about non-correlated residuals did not hold. To compensate for this, a nonlinear mixed model —including both fixed and random components— was applied. The first-order method of Beal and Sheiner (1982), available in the NLMIXED procedure of SAS/STAT® (SAS Institute Inc. 2009) was used for model fitting. Different combinations of model parameters were tested to be mixed. Heterocedasticity was analyzed by F. Crecente-Campo et al. / Forest Systems (2012) 21(3), 538-542 540 index (Diéguez-Aranda et al., 2005), RBA – D is a basal area ratio, raised to the power of a diameter ratio ([g/G]d/Dg, where Dg is quadratic mean diameter –cm) and αi are fixed parameters, common to the population. The results for the best mixed-effects models are shown in Table 1. All parameters were significant at the 0.01 level. The plots of residuals against predicted values showed homogeneous variance patterns. An attempt was made to relate the random parameters to stand variables but no significant relationships were found. Wykoff (1990) and Monserud and Sterba (1996) found that most of the variance explained in basal area increment was due to size factors (diameter, crown ratio), and then competition variables, primarily basal area in larger trees (BAL). The total variance explained was between 20% and 63%, from which only 2-6% was explained by site factors. The same was found in this study. The height increment model showed poorer results. Similarly, Hasenauer and Monserud (1997) could only explain 14% of the observed variation. Tanaka (1988) explained so little variation in height increment (R2 < 0.01) that a purely stochastic height increment model was developed instead. Other studies developed non-linear height-diameter models to obtain height estimations (e.g., Mabvurira and Miina, 2002; Palahí et al., 2003), instead of developing a height increment model. Other authors (e.g., Nord-Larsen, 2006; Burkhart et al., 2003) obtained similar results than in this study, explaining 37% and 46% of the observed variation, respectively. The ability of the models developed here to handle thinning effects is associated with the variation in stand variables and competition indices that the removal of some trees causes. Past research has shown that a true thinning effect was rarely significant (e.g., Pukkala, 1989), and was then only a minor effect (Monserud and Sterba, 1996). The fixed parts of Eqs. (1) and (2) are proposed for individual-tree basal area and height annual increment estimations of radiata pine stands in Galicia, with the parameter estimates shown in Table 1. These models plus the existing individual-tree mortality model (Crecente-Campo et al., 2009) conform a whole individual-tree growth model that can be used to forecast forest management alternatives, giving accurate results (Table 2) for the entire rotation age Table 1. Parameter estimates and goodness-of-fit statistics for the individual-tree basal area and height increment models Model Fixed effects Variance components RMSE EF Parm. Estimate Std. Err. Parm. Estimate Std. Err. Basal area α00.2822 0.0288 σ2210.0 2.9 14.82 (16.21*) 0.6196 (0.5448*) α12.845 0.064 σ2 b00.01470 0.00121 α2–0.8608 0.0546 σ2 b40.006326 0.001178 α3–0.05018 0.00257 σ2 b51.676 0.242 α4–0.04286 0.01142 σ2 b0b4–0.004530 0.000950 α5–1.471 0.0170 σ2 b0b5–0.1444 0.0119 σ2 b4b50.03648 0.01112 Height α00.05284 0.01268 σ20.1011 0.0021 0.3123 (0.3575*) 0.5157 (0.3647*) α1–0.5727 0.07344 σ2 b10.2644 0.0041 α20.5482 0.05977 σ2 b20.2198 0.0032 α31.080 0.087 σ2 b1b2–0.2375 0.0017 α4–0.03272 0.00353 α5–50.01 5.46 bi: random parameter related with the fixed αi parameter; σ2: error variance of the model; σ2 bi: variance associated with the random parameter bi; σbibj: covariance between random parameters bi and bj. *These statistics were calculated with only the fix part of the models. Table 2. Statistics obtained with the individual-tree growth model for stand basal area (G) and trees per hectare (N) projection using several cut-off values cut-off n RMSE G RMSE N 0.80 9 2.413 98.21 0.81 9 2.399 95.56 0.82 12 2.387 92.77 0.83 17 2.363 88.62 0.84 17 2.341 85.56 0.85 25 2.337 84.77 0.86 27 2.315 82.07 0.87 33 2.308 82.06 0.88 35 2.313 84.29 0.89 39 2.316 89.15 0.90 43 2.331 97.34 n: number of plots (out of 130) that failed to pass the Kolmogorov-Smirnov test. 541 Individual-tree growth model for radiata pine plantations in northwestern Spain of the species in the region, i.e., 25-45 years depending on site quality. From application, a fixed cut-off value of 0.84 gives a good compromise between G and N estimations and the number of plots that passed the Kolmogorov-Smirnov test (Table 2). Results for 50 years of projection using a cut-off of 0.84 are shown (Fig. 1) in order to show the biological realism of the model. Acknowledgements The present study was financed by the Spanish Ministry of Education and Science; project No AGL-2004- 07976-C02-01. References Akaike H. 1974. A New Look at the Statistical Model Identification. IEEE T Automat Contr AC-19, 716-723. Beal SL, Sheiner LB. 1982. Estimating population kinetics. CRC Crit Rev Biomed Eng 8, 195-222. Burkhart HE. 2003. Suggestions for Choosing an Appropriate Level for Modeling Forest Stands. In: Amaro A, Reed D, Soares P. (Eds.), Modeling Forest Systems. CAB International, Wallingford, UK, pp. 3-10. Burkhart HE, Amateis RL, Westfall JA, Daniels RF. 2003. PTAEDA3: Simulation of Individual Tree Growth, Stand Development and Economic Evaluation In Loblolly Pine Plantations. Report No. 134, College of Natural Resources, Virginia Tech, Blacksburg, Virginia. Castedo-Dorado F, Diéguez-Aranda U, Álvarez-González JG. 2007. A growth model for Pinus radiata D. Don stands in north-western Spain. Ann For Sci 64, 453-465. Crecente-Campo F, Marshall P, Rodríguez-Soalleiro R. 2009. Modeling non-catastrophic individual-tree mortality for Pinus radiata plantations in northwestern Spain. For Ecol Manage 257, 1542-1550. DGCN. 1998. Segundo inventario forestal nacional, 1986- 1996. España. Dirección General de Conservación de la Naturaleza, Ministerio de Medio Ambiente, Madrid. Diéguez-Aranda U, Burkhart HE, Rodríguez-Soalleiro R. 2005. Modelling dominant height growth of radiata pine (Pinus radiata D. Don) plantations in north-western Spain. For Ecol Manage 215, 271-284. Gadow Kv, Hui G. 1999. Modelling Forest Development, Kluwer Academic Publishers. García O. 2003. Dimensionality reduction in growth models: an example. Forest Biometry, Modeling and Information Sciences 1, 1-15. Hasenauer H, Monserud RA. 1997. Biased predictions for tree height increment models developed from smoothed ‘data’. Ecol Model 98, 13-22. Kozak A, Kozak R. 2003. Does cross validation provide additional information in the evaluation of regression models? Can J Forest Res 33, 976-987. Mabvurira D, Miina J. 2002. Individual-tree growth and mortality models for Eucalyptus grandis (Hill) Maiden plantations in Zimbabwe. For Ecol Manage 161, 231-245. Monserud RA, Sterba H. 1996. A basal area increment model for individual trees growing in even- and uneven-aged forest stands in Austria. For Ecol Manage 80(1-3), 57-80. Myers RH. 1990. Classical and Modern Regression with Applications (Second ed.). Duxbury Press, Belmont, CA. Nord-Larsen T. 2006. Modeling individual-tree growth form data with highly irregular measurement intervals. Forest Sci 52(2), 198-208. Palahí M, Pukkala T, Miina J, Montero G. 2003. Individualtree growth and mortality models for Scots pine (Pinus sylvestris L.) in north-east Spain. Ann For Sci 60, 1-10. 0 0 20 40 60 80 100 20 40 60 80 100 Stand basal area (m2ha–1) Age (years) 0 400 800 1,200 1,600 2,000 2,400 2,800 Stand density (trees ha –1 ) 0 20 40 60 80 100 Age (years) Figure 1. Simulation of the changes in stand basal area and tems per hectare by use of the whole individual-tree growth model for radiate pine stands in Galicia. F. Crecente-Campo et al. / Forest Systems (2012) 21(3), 538-542 542 Pukkala T. 1989. Predicting diameter growth in an even-aged Scots pine stand with a spatial and a non spatial model. Silva Fenn 23, 101-116. Sánchez F, Rodríguez R, Rojo A, Álvarez JG, López CA, Gorgoso JJ, et al. 2003. Crecimiento y tablas de producción de Pinus radiata D. Don en Galicia. Inv Agrar. Sist Rec F 12(2), 65-83. SAS Institute Inc. 2009. SAS/STAT® 9.2 User’s Guide (2nd ed.). SAS Institute Inc., Cary, NC. 7869 pp. Schröder J, Gadow Kv. 1999. Testing a new competition index for maritime pine in north-western Spain. Can J Forest Res 29, 280-283. Tanaka K. 1988. A stochastic model of height growth in an even aged pure forest stand: why is the coefficient of variation of the height distribution smaller than that of the diameter distribution? J Jap Forest Soc 70(1), 20-29. Temesgen H, Gadow Kv. 2004. Generalized height-diameter models — an application for major tree species in complex stands of interior British Columbia. Eur J For Res 123, 45-51. Wykoff WR. 1990. A basal area increment model for individual conifers in the northern Rocky Mountains. Forest Sci 36(4), 1077-1104. Xunta de Galicia. 2001. O monte galego en cifras. Dirección Xeral de Montes e Medio Ambiente Natural, Consellería de Medio Ambiente, Santiago de Compostela. Zeide B. 1993. Analysis of growth equations. Forest Sci 39(3), 594-616.