| Trees form the main natural element of the urban landscape. However, the urban environment induces severe ecological conditions that impair tree growth and survival. This situation is of major concern to municipal administrations that devote large budgets to arboricultural programmes. To adequately preserve arboreal heritage, three main issues must be resolved. First, biotic and abiotic inventory parameters that can express the complexity of street tree growth have to be assessed. Second, for an enhanced understanding of tree health and related environmental conditions, an analytical methodology should be defined to cluster street trees with similar growth patterns. Third, optimized tree-inventory procedures ought to be determined.;To develop the classification methodology, a two-step procedure was chosen. First, intermediate linkage clustering and correspondence analysis were used to ascertain groups with dissimilar growth rates. On the species level, the partitioning of dendrograms resulted in three or four clusters. A closer examination of the results showed that clusters with small trees were composed not only of young trees but also of older stressed individuals, in need of enhanced arboricultural care. Attempts to define an all-species model were not conclusive as cluster analysis appeared sensitive to outliers. Accordingly, it was proposed that classification models could be defined on an individual species basis. Second, the clustering knowledge was used to train artificial neural networks (ANNs) to recognize tree growth patterns and predict cluster affiliation. Global cluster classification was estimated by computing the proportion of correctly classified trees over the total number of observations for all train/test cross-validation files. The average value for all species taken together was 89%. To assess the classification success into groups within species, specific correct/incorrect cluster predictions were computed. The classification performance was high. For most species, unseen test files prediction accuracy ranged from 80% to almost 100%.;To optimize inventory procedures, multilayer perceptron (MLP) networks were used to predict the value of essential tree morphological parameters with surrogate variables. Two research objectives were initially decided upon: first, to automate the prediction of DBH, annual DBH increment and crown volume when input data are obtained with aerial LIDAR (Light Detection And Ranging) laser; and secondly to predict height, crown volume and their respective annual increments using less field-work-intensive variables. The prediction performance was assessed with the Pearson r correlation coefficient, computed between the measured and estimated output values for each cross-validation test files per species and model. Pearson r coefficients were greater than 70% for all models except for two predictions, where MLPs appeared sensitive to strongly non-Gaussian distributions. Overall, the average coefficient value for all scenarios was 91%. Despite different age-class distribution of trees, dissimilar morphological characteristics, and uneven species partition within urban ecological zones, ANNs demonstrated considerable robustness and predictive capability.;By taking into account the limitations of various mathematical approaches and exploiting their complementary advantages, an intelligent modeling system was developed to assess urban tree growth. It is suggested that multivariate statistics and artificial intelligence algorithms can become important analytical components of efficient street-tree management plans.;To fulfill these objectives, an experiment was designed using multiple variables. Chosen parameters were measured on 1532 trees and associated sites in Montreal, in five different urban ecological zones. Seven species representing 75% of the total street tree population were sampled: Acer platanoides L. (Norway maple), Acer saccharinum L. (Silver maple), Celtis occidentalis L. (Common hackberry), Fraxinus pennsylvanica Marsh. (Green/Red ash), Gleditsia triacanthos L. (Honeylocust), Tilia cordata Mill. (Littleleaf linden), and Ulmus pumila L. (Siberian elm). To identify key inventory parameters, two approaches were used: multivariate statistics (principal coordinate and correspondence analyses) determined biotic variables, and contingency analysis investigated environmental variables. Results from multivariate analysis revealed that qualitative biotic parameters are of low explanatory importance and might be excluded from inventory protocols. Next, it was discovered that modeling with the synergistic combination of diameter at breast height (DBH), annual DBH increment, crown diameter, crown diameter increment, height and height increment, crown diameter/DBH, crown volume/DBH, height/DBH, crown volume, and crown volume increment gave an adequate portrayal of all tree physiological stages. Contingency analysis unveiled links between some environmental factors and tree growth. Overall, the urban zone type, geomorphologic surficial deposit, presence/absence of aerial obstacles, solar irradiation levels, street width, distance from tree to curb and to adjacent building, tree pit soil volume, and soil penetration resistance were central parameters for some or all species. |
