Use of artificial neural networks to assess yield projection and average production of eucalyptus stands

Eucalyptus stands growth depends on genotype, age, quality of the local soil and silvicultural treatment. Environmental factors, mainly the water availability to plants throughout the years, temperature and solar radiation are relevant to production capacity. The models used in Brazil to stimulate the future production of forestry stands are those that estimate growth and/or production according to age, basal area and local index. One of the possible approaches to do so is the use of procedural models (ecophysiological) such as the 3PG and the artificial neural network. The current study has the aim to construct, validate and apply an artificial neural model to predict the production and growth of eucalyptus stands in Minas Gerais, Brazil. The herein used data resulted from continuous forestall inventory plots conducted in eucalyptus stands in the North, Center and South of the state. The edaphic and climatic information added to the IFC data were used to train neural nets on predicting growth and production in the state. A neural network, lacking inventory variables, was also trained to extrapolate the mean productivity in the entire state of Minas Gerais due to the physiographic, edaphic and climatic conditions. The neural network efficiency was attested by the great accuracy of productivity forecasts. The generated productivity maps are indicated for studies on the expansion of eucalyptus cultivation in the state. The applied methodology is simple and efficiently inapplicable to different forestry cultures in other states or countries. Key words: Eucalyptus, water availability, forestry stands, neural network, productivity.