engleski

Modelling the damage status of silver fir trees (Abies alba Mill.) on the basis of stand, geomorphological and climatic factors

Introduction The paper investigates the possibility of assessing damage (health status) of silver fir (Abies alba Mill.) using multivariate regression models in the function of stand and ecological factors. Research was carried out on limestone-dolomite substrate in the Dinaric part of the silver fir range in the Republic of Croatia. Damage was assessed on 151 plots. A total of 531 trees of silver fir (Abies alba Mill.) were sampled. To perform the assessment, tree crowns were compared with the existing photointerpretation key on the damage scale of 5%. Of stand factors, the following independent variables were used in modelling: 1) breast diameter of a tree as a variable representing competing ability and position of tree in the stand, 2) tree basal area in relation to (divided by) plot basal area, as an indicator of competitive pressure on the sampled tree, and 3) tree age. Other independent variables included: 1) geomorphological (DEM-based) variables (terrain slope, altitude, terrain orientation with regard to the northness and eastness, flow accumulation, sinkhole depth, terrain curvatures, terrain exposure to the horizontal wind flux, latitude and longitude) and 2) climatic variables (monthly mean air temperature, monthly precipitation, monthly mean global solar irradiation on horizontal surface at ground, monthly potential evapotranspiration on horizontal surface). The basic set of 48 independent macroclimatic estimators (4 variables by 12 months) was reduced to 5 composite estimators (non-linear analogues of principal components) using the five-layered autoassociative NN, which have 48 neurons in the first and last layer (48 basic estimators), 15 neurons in the second and the fourth layer and 5 neurons in the central layer. Logistic function was used as activation function. Using this NN architecture, 99.79% of total macroclimatic variability was explained. A general linear modelling method was used, where square terms and interaction terms (multiplication products) of original variables were treated as independent linear predictors (Ott, 1993). The 'Backward Stepwise' method was used for model optimisation (selection of a subset of linear predictors entering the final model). A 27 separate models were developed regarding the different subsets of input data (all trees, trees with diameter over as well as bellow 40 cm), different subsets of independent variables (limited number of DEM-based variables, all DEM-based variables, all variables without age, all variables) and different model design (with and without square terms). Results and discussion Both particular models developed from separate data sets with regard to diameter of 40 cm (presumed boundary between dominant trees and suppressed or young trees) explained significantly larger part of total variability (in all combinations of other mentioned model characteristics) in relation to the respective models developed for all trees together. Consequently, models developed for all trees were not further examined. All remaining models were preliminary used for construction of hypothetical spatial distributions of fir damage for entire area of fir in Croatia. Criteria for the final choosing models potentially applicable for spatial prediction of fir damage were: 1) total variability explained by the model and 2) portion of predicted values within the range of input data about fir damage (0-100 %). Six finally selected models (all of them with reduced set of DEM-based variables and square terms) were tested on the independent data set collected in the frame of International Co-operative Programme on Assessment and Monitoring of Air Pollution Effects on Forests (ICP ; Group of authors, 1989). Significant correlation between model results were yielded only for trees thinner than 40 cm, which can be explained by the impact of forest management which prefer cutting of damaged mature trees. Conclusions Yielded results could be preliminary used for spatial predicting and mapping of fir damage in a frame of raster-GIS, for entire area of fir in Croatia. Future research have to be leaded in two major directions: 1) completing of larger field sample aiming at using of more plastic prediction models (e.g. developed by neural networks), 2) integration of spatial forest health prediction models with the spatial models of aeropollutants imissions (Antonić and Legović, 1999). Acknowledgements This work was supported by Ministry of Science and Technology of the Republic of Croatia, and by OIKON Ltd. Institute for Applied Ecology. References Antonić, O., Legović, T. Estimating the direction of an unknown air pollution source using a digital elevation model and a sample of deposition. Ecological Modelling 124, 85-95, 1999. Ott, R.L. An Introduction to Statistical Methods and Data Analysis. Duxbury Press, Belmont, 1993. Group of authors. Manual on methodologies and criteria for harmonized sampling, assessment, monitoring and analysis of the effects of air pollution on forest, Hamburg/Geneve: Programme coordinating centers, UN/EC 1986 (revised 1989).

digital elevation model; fir health status; general linear modelling; macroclimate; raster-GIS

nije evidentirano