ARTIFICIAL NEURAL NETWORK PREDICTION OF STREET TREE GROWTH PATTERNS

作者：Jutras P^*; Prasher S O; Mehuys G R

来源：Transactions of the ASABE, 2010, 53(3): 983-992.

• Artificial neural networks
• Classification prediction
• Decision support systems
• Multidimensional correspondence analysis
• Multivariate linkage clustering
• Radial basis functions
• Street trees
• Urban forestry
• Urban tree growth modeling

Municipal administrations annually allocate a large budget to preserve their urban trees. However, survival and growth rates of street trees vary drastically as they are strongly influenced by adverse environmental conditions. Consequently, arboricultural programs must be locally adapted to provide care to stressed trees. Artificial neural networks were used to identify poorly growing trees by learning from growth patterns detected by multivariate statistical analyses. Seven species that are representative of 75% of the Montreal street tree population were sampled: Acer platanoides L., Acer saccharinum. L., Celtis occidentalis L., Fraxinus pennsylvanica Marsh., Gleditsia triancanthos L., Tilia cordata Mill., and Ulmus pumila L. Individuals were of different age, dissimilar morphological characteristics, and had variable distribution among urban ecological zones. Radial basis function networks (RBFs) were selected as the model type. To assess RBFs robustness and predictive capability on unknown data, global and specific cluster classification was used. When global classification was estimated, the lowest accurate prediction value was 83% and the highest 93%. The average, value for all species taken together was 89%. Similarly, the classification success within groups per species was adequate. For most species, test files prediction accuracy ranged from 80% to almost 100%. This indicates that RBFs are well suited for classification decisions. These results have an impact on the management of street trees. Given the present findings, integrating robust predictive algorithms into data banks as a decision-support system is a conceivable avenue. Artificial-intelligence-based models might probably become important elements of efficient street tree management plans.

• 出版日期2010-6