ABSTRACT Background: The aim of this research is to optimize the routing of Brazil nut harvesting in order to reduce labor costs and promote an equitable redistribution of placements among extractive interests. The study area was located in the Tahuamanu province, designated a Biosphere Reserve by UNESCO, in the Madre de Dios region of Peru. Planning for harvesting activities is vital to support sustainable use of the tropical forest by extractive communities, generating profit while minimizing adverse impacts on the forest. A location-allocation model was used to redistribute extraction placement areas, taking into account the number of trees and productivity for analysis. To optimize the order of nut load pickups at the placement areas, the vehicle routing problem (VRP) was used to minimize real distance and reduce traversal time. Results: The time taken to cover the route without delays was 08:46’, which is superior to the time taken to minimize the distance. However, the total delay time was 05:10’. Conclusion: Network analyses were effective in achieving the reallocation of placement areas and optimizing Brazil nut harvest routes.

ABSTRACT Evaluations of biomass and nutrient contents in teak stands are necessary alternatives for avoiding early cutting to provide supplemental data for the plant module of fertilization recommendation systems. This study aimed to adjust allometric equations to estimate the accumulation of biomass and nutrient contents in teak plantations in the central region of Brazil. Plots in seminal and clonal stands contained 81 trees aged 75 months, that had not received previous fertilization or thinning. Additional clonal stands aged 15, 51 and 63 months and a seminal stand aged 63 months had been previously fertilized during implantation. Allometric equations for biomass and macronutrient contents were obtained for components of the aerial parts of trees as a function of diameter at 1.3 m height (DBH), and the possibility of using an equation or the need for specific equations was evaluated for genetic material (seminal or clonal) and the fertilization scheme (not fertilized or fertilized). The equations adjusted to the four sets of stands, all with elevated predictive capacity, did not differ among each other based on an identity test. Thus, the use of a robust equation adjusted with data from all stands is recommended to obtain estimates with a high degree of accuracy. The biomass and accumulation of macronutrients in components of the aerial parts of teak trees should be considered to obtain the nutrient contents in both the exportable components in thinned trees and the components maintained in the area and available for biogeochemical cycling.

ABSTRACT The objective of this study was to compare different alternatives to estimate the stem volume of individual trees in four different forest formations in the Minas Gerais state, Brazil. The data were obtained in a forest inventory procedure performed by the Minas Gerais Technological Center Foundation. The stem volumes were computed by the Smalian expression up to the outside bark diameter equal to 4 cm. The volume data of outside bark, diameters (DBH) and total heights were used to fit a Schumacher and Hall equation for each forest formation, considering the structures of the linear fixed and mixed models. Next, 100 Multilayer Perceptron artificial neural networks (ANN) were trained in a supervised manner. In addition, we evaluated eight support-vector machine regression (SVMR). The criteria to evaluate the performance of all the alternatives studied were: the correlation between the observed and estimated volumes, the square root of the mean square error and the frequency distribution by percentage relative error class. After the analyzes, all the alternatives were verified to estimate the volume of the individual trees in the different forest formations. Although the alternatives presented close statistics in the validation process, the graphical analysis of the error distribution showed greater precision of the estimates of the mixed linear models for the four formations. Given the results, it is concluded that there is no absolute superiority of one alternative over the others, and that all of them should be evaluated to find the one which best describes or explains the dataset.

Abstract The objective of this study was to assess and compare two methodologies to predict the mortality of individual trees in a seasonal semi-deciduous forest located in the municipality of Viçosa, Minas Gerais, Brazil. The data came from 10 permanent plots and were randomly divided into two groups: fitting (6 plots) and validation (4 plots). When assessing Methodology 1, the mortality was considered a dichotomous variable in a logistic model, assuming a value of 1 for a dead tree and 0 for a living tree. In Methodology 2, the mortality probability was estimated by the group of trees for each plot and for each measurement interval. After the analysis, it was observed that Methodology 2 provided more consistent mortality estimates when compared to the mortality estimated by Methodology 1.

RESUMO A distribuição de diâmetros de uma floresta permite entender seus processos ecológicos e definir ações relacionadas ao manejo florestal visando a conservação da floresta, aumento da biodiversidade, produção e sequestro de carbono, entre outros. Este estudo buscou identificar variações na distribuição diamétrica em uma Mata Atlântica secundária, em função de sítios, ao longo de 24 anos. O diâmetro a 1,3 m de altura (dap) foi medido para cada árvore acima de 5,0 cm dap, em dez locais, de 1992 a 2016. Foi analisado o diâmetro médio quadrático (Dq), o incremento periódico anual em Dq (IPADq), o Quociente de Liocourt (q) e a distribuição de diâmetros com o modelo exponencial de Meyer. O IPADq médio em 24 anos foi de 0,03 cm ano-1, um aumento médio de 0,21%, com variação local de 0,74% a -0,45%. A distribuição de diâmetros da floresta como um todo manteve o formato de J-invertido ao longo de 24 anos. Quando os locais foram analisados em cada ano de inventário, a variação da distribuição de diâmetros foi evidente e houve diferenças na capacidade de recuperação da distribuição para a forma de J-invertido de 1992 a 2016. Os resultados indicam que, sempre que possível, o plano de manejo para recuperar a estrutura dos remanescentes da Mata Atlântica deve ser específica para cada local, particularmente na região montanhosa, onde as condições ambientais são bastante diversas.

ABSTRACT The diameter distribution of a forest makes it possible to understand its ecological processes and to define actions related to forest management aiming at forest conservation, biodiversity increase, production, and carbon sequestration, among others. This study aimed to identify variation in the diameter distribution in a secondary Atlantic Forest, as a function of sites, over 24 years. The diameter at 1.3 m in height (dbh) was measured for each tree over 5.0 cm dbh, in ten sites, from 1992 to 2016. It was analyzed the quadratic mean diameter (Dq), the Dq annual periodic increment (APIDq), the Liocourt’s Quotient (q), and the diameter distribution with Meyer’s exponential model. The average APIDq over 24 years was 0.03 cm year-1, an average increase of 0.21%, with site variation ranging from 0.74% to -0.45%. The diameter distribution of the forest as a whole maintained the reverse J-shaped over 24-years. When the sites in each inventory year were analyzed, the variation in the diameter distribution was evident, and there were differences in the recovery capacity of the reverse J-shaped distribution from 1992 to 2016. The overall results indicate that, whenever possible, the management plan to recover the structure of the Atlantic Forest remnants should be site-specific, particularly in the mountainous region where the environmental conditions are quite diverse.

Resumo A modelagem do recrutamento em florestais tropicais é importante para estudos de sustentabilidade do manejo florestal, por dar subsídio adequado à recuperação do estoque de madeira. O objetivo do trabalho foi estimar o recrutamento após a colheita de madeira, empregando um modelo de rede neural artificial (RNA). A área de estudo está localizada na Floresta Nacional do Tapajós (55°00’ W, 2°45’ S), Pará. Em 64 ha da área de estudo, em 1979, foi realizada colheita intensiva de 72,5 m3 ha-1. Em 1981 foram instaladas, aleatoriamente, 36 parcelas permanentes de 50 m x 50 m. Essas parcelas foram mensuradas em 1982, 1983, 1985, 1987, 1992, 1997, 2007, 2010 e 2012. Para modelar o recrutamento foram consideradas as variáveis da subparcela-alvo e a sua vizinhança. As estimativas obtidas no treino e na generalização da RNA foram avaliadas pelas estatísticas: correlação () e raiz quadrada do erro quadrático médio (RQRQM), sendo obtido RQRQM 35,6% e 0,89. Foi possível modelar a tendência do recrutamento ao longo do tempo em florestas tropicais, após a colheita de madeira.

Abstract Recruitment models in tropical forests are important for studies on forest management sustainability because they provide adequate support to recovery of wood stocks. The objective of this work was to estimate recruitment after wood harvesting by using an artificial neural network (ANN) model. The study area is located at Tapajós National Forest (55° 00’ W, 2° 45’ S), Pará state. In 64 ha of the study area, in 1979, an intensive harvest of 72.5 m3 ha-1 was carried out. In 1981, 36 permanent plots of 50 m x 50 m were randomly installed. These plots were measured in 1982, 1983, 1985, 1987, 1992, 1997, 2007, 2010 and 2012. For recruitment modeling, the variables of the target subplot and its vicinity were considered. The estimates obtained in ANN training and generalization were evaluated by statistics: correlation () and root mean square error (RMSE) were determined: RMSE 35.6% and 0.89. Recruitment tendency could be modeled over time in tropical forests after wood harvesting.

RESUMO Os sistemas agroflorestais (SAF) devem beneficiar todos os componentes do consórcio. O objetivo deste trabalho foi avaliar o efeito da distância entre plantas na linha de plantio sobre o crescimento e a produção do componente arbóreo e da cultura agrícola e, sobre a viabilidade econômica do consórcio. Para isso plantou-se clone de eucalipto nos arranjos de 9,5 x 1,5 m, 9,5 x 2,0 m, 9,5 x 3,0 m e 9,5 x 40 m, em consórcio com soja. O diâmetro (dap) e a altura total (Ht) da árvore e o índice de área foliar (IAF) foram medidos aos 14, 38 e 51 meses. A produção de soja foi avaliada aos 24 meses após o plantio do eucalipto. A avaliação econômica do consórcio foi realizada para um horizonte de planejamento de sete anos. A distância entre as plantas na linha de plantio não afetou a altura das árvores, no entanto, maiores distâncias promoveram maior DAP e volume individual. Valores mais altos de área basal e produção foram alcançados no arranjo de 9,5 x 1,5 m. O IAF foi menor (1,43) no arranjo 9,5 x 4,0 m, aos 38 meses, em comparação aos demais arranjos (média de IAF = 1,66). A produção de soja foi a maior (2.317 kg ha-1) no arranjo de 9,5 x 4,0 m. Na avaliação econômica, a madeira produzida nos dois povoamentos mais densos foi destinada à energia, com baixo valor de mercado, inviabilizando economicamente esses arranjos, embora o rendimento do componente arbóreo tenha sido o mais elevado no 9,5 x 1,5 m. O arranjo espacial de 9,5 x 4,0 m foi o mais viável economicamente, considerando a alocação de 40% da madeira para serraria, e os preços e custos assumidos neste estudo. Neste arranjo de plantio, o rendimento da soja foi o mais elevado e os custos de plantio foram os menores, quando comparados aos demais arranjos.

ABSTRACT Agroforestry systems (AGF) design should benefit all components of the consortium. The objective of this study was to evaluate the effect of the distance between plants within the row on growth and yield of trees and agricultural crop and on the economic viability of the consortium. The eucalypt was planted in 9.5 x 1.5 m, 9.5 x 2.0 m, 9.5 x 3.0 m and 9.5 x 40 m arrangements, in consortium with soybean. The tree diameter (dbh) and total height (Ht), and the leaf area index (LAI) were measured at 14, 38 and 51 months. Soybean yield was evaluated 24 months after planting eucalypt. An economic evaluation of the consortium was carried out for a planning horizon of seven years. The distance between trees within the row did not affect the tree height, however, larger distances promoted a higher dbh and individual volume. Higher values of basal area and yield were achieved in the 9.5 x 1.5 m arrangement. The LAI was smaller (1.43) in the 9.5 x 4.0 m arrangement, at 38 months, compared to the other arrangements (mean LAI = 1.66). Soybean had it's highest yield (2,317 kg ha-1) in the 9.5 x 4.0 m arrangement. In the economic evaluation, the wood produced in the two denser arrangements was destined for energy, with low market value, making these two arrangements unfeasible economically, although the tree component yield was the highest in the arrangement 9.5 x 1.5 m. The 9.5 x 4.0 m spatial arrangement was the most economically viable, considering the allocation of 40% of the wood for sawing, and the prices and costs assumed in this study. In this arrangement, the soybean yield was the highest, and the planting costs were the lowest when compared to the other arrangements.

Resumo: O objetivo deste trabalho foi comparar métodos para obtenção do índice de sítio para povoamentos de eucalipto (Eucalyptus spp.), bem como avaliar seus impactos na estabilidade desse índice em bases de dados com e sem a presença de “outliers”. Foram testados três métodos, com uso de regressão linear, regressão quantílica e rede neural artificial. Foram utilizadas 22 parcelas permanentes de inventário florestal contínuo, medidas em árvores com idade de 23 a 83 meses. Os outliers foram identificados com uso de gráfico de boxplot. A rede neural artificial proporcionou melhores resultados que as regressões linear e quantílica, tanto para as estimativas de altura dominante quanto do índice de sítio. A estabilidade da classificação do índice de sítio obtida pela rede neural artificial também foi melhor que a obtida com os outros métodos, independentemente da presença ou da ausência de outliers na base de dados. Isso indica que a rede neural artificial é uma técnica sólida de modelagem na presença de outliers. Quando a causa da presença de outliers na base de dados não é conhecida, eles podem ser mantidos nela se técnicas como as de redes neurais artificiais ou de regressão quantílica forem utilizadas.

Abstract: The objective of this work was to compare methods of obtaining the site index for eucalyptus (Eucalyptus spp.) stands, as well as to evaluate their impact on the stability of this index in databases with and without outliers. Three methods were tested, using linear regression, quantile regression, and artificial neural network. Twenty-two permanent plots from a continuous forest inventory were used, measured in trees with ages from 23 to 83 months. The outliers were identified using a boxplot graphic. The artificial neural network showed better results than the linear and quantile regressions, both for dominant height and site index estimates. The stability obtained for the site index classification by the artificial neural network was also better than the one obtained by the other methods, regardless of the presence or the absence of outliers in the database. This shows that the artificial neural network is a solid modelling technique in the presence of outliers. When the cause of the presence of outliers in the database is not known, they can be kept in it if techniques as artificial neural networks or quantile regression are used.

Resumo: O objetivo deste trabalho foi avaliar a aplicação do escaneamento a laser aerotransportado (ALS) na realização de inventário de povoamentos de eucalipto, em larga escala, por meio do método de individualização de árvores, bem como propor um novo método para estimativa dos diâmetros das árvores em função das alturas obtidas a partir da nuvem de pontos. O estudo foi conduzido em floresta de 1,681 ha, composta por oito povoamentos de eucalipto com idades entre quatro e sete anos. Após o escaneamento, foram obtidas as alturas das árvores pelo algoritmo de máximos locais, e o estoque total de madeira, pela soma dos volumes individuais. Para determinar os diâmetros das árvores, foram utilizadas regressões ajustadas a partir de dados medidos em parcelas de inventário. Os resultados foram comparados às estimativas obtidas via amostragem de campo. O sistema de equações proposto é adequado para ser aplicado aos dados de altura derivados da nuvem de pontos do ALS. O método de individualização de árvores com o filtro de máximos locais é eficiente para estimar o número de árvores e o estoque de madeira a partir dos dados do ALS, desde que os resultados sejam previamente calibrados com o inventário de campo.

Abstract: The objective of this work was to evaluate the application of airborne laser scanning (ALS) to a large-scale eucalyptus stand inventory by the method of individual trees, as well as to propose a new method to estimate tree diameter as a function of the height obtained from point clouds. The study was carried out in a forest area of 1,681 ha, consisting of eight eucalyptus stands with ages varying from four to seven years. After scanning, tree heights were obtained using the local maxima algorithm, and total wood stock by summing up individual volumes. To determine tree diameters, regressions fit using data measured in the inventory plots were used. The results were compared with the estimates obtained from field sampling. The equation system proposed is adequate to be applied to the tree height data derived from ALS point clouds. The tree individualization approach by local maxima filters is efficient to estimate number of trees and wood stock from ALS data, as long as the results are previously calibrated with field data.

Resumo: O objetivo deste trabalho foi avaliar o crescimento e a produção de povoamentos de teca (Tectona grandis) em diferentes espaçamentos e classes de solo. Foram avaliados 12 espaçamentos em Cambissolo e Latossolo, dispostos em parcelas com área de 1.505 ou 1.548 m2, em delineamento inteiramente casualizado, com nove repetições. As plantas de teca foram medidas aos 26, 42, 50 e 78 meses de idade. A altura total das árvores foi menos influenciada pelos espaçamentos. O diâmetro quadrático médio foi superior nos maiores espaçamentos, enquanto a área basal e o volume total com casca foram superiores nos espaçamentos mais adensados. Observou-se aumento do volume com casca por árvore com o aumento da área útil por planta. Em teca, a estagnação do crescimento ocorre mais cedo, a taxa de crescimento é maior nos espaçamentos mais adensados e as plantas crescem mais em Cambissolo do que em Latossolo.

Abstract: The objective of this work was to evaluate the growth and yield of teak (Tectona grandis) stands at different spacing and in different soil classes. Twelve spacing were evaluated in an Inceptisol and Oxisol, in plots with an area of 1,505 or 1,548 m2, arranged in a completely randomized design with nine replicates. The teak trees were measured at 26, 42, 50, and 78 months of age. Total tree height was less affected by spacing. Mean square diameter was greater in wider spacing, whereas basal area and total volume with bark were greater in closer spacing. An increase in volume with bark per tree was observed with the increase of useful area per plant. For teak trees, growth stagnation happens earlier, the growth rate is higher in closer spacing, and the plants grow more in the Inceptisol than in the Oxisol.

ABSTRACT In forest science it is important evaluate new technologies from computational science. This work aimed to test a different kind of metaheuristic called Variable Neighborhood Search in a forest planning problem. The management total area has 4.210 ha distributed in 120 stands in ages between 1 and 6 years old and site index since 22 m to 31 m. The problem was modelled considering the maximization of the net present value subject to the restrictions: annual cut volume between 140.000 m³ and 160.000 m³, harvester ages equal to 5, 6 or 7 years, and the impossibility of division of the management unity at harvester time. It was evaluated different settings for the Variable Neighborhood Search, varying the quantity of neighbours, the neighbourhood structure and number or generations. 30 repetitions were performed for each setting. The results were compared to the one obtained from integer linear programming and linear programming. The integer linear programming considered the best solution obtained after 1 hour of processing. The best setting to the Variable Neighborhood Search was 100 neighbours, a neighbourhood structure with changes in 1%, 2%, 3% and 4% of prescriptions and 500 iterations. The results shown by the Variable Neighborhood Search was 2,77% worse than one obtained by the integer linear programming with 1 hours of processing, and 2,84% worse than the linear programming. It is possible to conclude that the presented metaheuristic can be used satisfactorily in a resolution of forest scheduling problem when the best parameters are chosen.

Abstract: Timber from the second cutting cycle may make up the majority of future crop volumetric. However, there are few studies of the physical and mechanical properties of this timber, which are important to support the consolidation of new species. This study aimed to use Artificial Neural Networks to estimate the physical and mechanical properties of wood from the Amazon, based on basic density. The properties were: shrinkage (tangential, radial and volumetric), static bending, parallel and perpendicular to the fiber compression, parallel and transverse to the fibers, Janka hardness, traction, splitting and shear. The estimate followed the tendency of the data observed for the tangential, radial and volumetric shrinkage. The network estimated the mechanical properties with significant accuracy. Distribution of errors, static bending, parallel compression and perpendicular to the fiber compression also showed significant accuracy. Artificial Neural Networks can be used to estimate the physical and mechanical properties of wood from Amazon species.

RESUMO Este trabalho teve como objetivo a escolha de um modelo de afilamento para melhor estimar o perfil médio de árvores de diferentes clones de eucalipto, verificando se uma mesma função pode ser empregada para três genótipos distintos de eucalipto. Os tratamentos consistiram em diferentes clones de eucalipto plantados sob o espaçamento de 3 x 3 m. Para cada tratamento foram abatidas cerca de 70 árvores-amostra aos 10 anos de idade. Os diâmetros das seções ao longo do fuste foram medidos a partir da base em 0.5 m; 1.0 m; 1.5 m; 2.0 m e, a partir daí, em intervalos de 2.0 m. Foram testados quatro modelos não-lineares: Garay (1979), Kozak (1988), Max e Burkhart (1976) e Demaerschalk (1972). As estatísticas utilizadas para avaliar os ajustes foram: coeficiente de correlação entre diâmetros observados e estimados, erro-padrão residual, bias (média dos erros), variância do bias, raiz quadrada do erro quadrático médio e a média das diferenças absolutas. O modelo escolhido para representar os tratamentos, pela melhor performance de ajuste, foi o de Kozak (1988). Testes de identidade de modelo foram aplicados para verificar igualdade no formato do fuste entre os tratamentos. Com base nos resultados, a hipótese de igualdade na forma do fuste dos genótipos foi rejeitada, ao nível de 5% de significância. Assim, pôde-se concluir, que existe efeito do genótipo no afilamento do fuste de eucalipto.

ABSTRACT This study aims to evaluate non-linear stem taper models for predicting the pre-commercial diameter of eucalyptus trees and to analyze the effect of genotype on stem taper. The treatments comprise three different genotypes of Eucalyptus sp. in a 3 × 3 m plantation spacing. Seventy sample trees aged 10 years were felled for each treatment. The outside bark diameter measurements were taken at 0.5 m; 1.0 m; 1.5 m; 2.0 m, and then at intervals of 2.0 m till the top of the stem. Four non-linear models were evaluated, namely, the sigmoid model of Garay (1979), the variable exponent model of Kozak (1988), the segmented model of Max and Burkhart (1976), and the compatible model of Demaerschalk (1972). The performance of the models was assessed using the following statistical validation methods: correlation coefficient, standard error of estimate, mean bias, bias variance, root mean squared error, and mean absolute deviation. Graphical analysis of residues was used to evaluate the accuracy and precision of the estimates. Compared with other models, the variable exponent model of Kozak (1988) best described the stem profile, and predicted the total volume of the trees. The identity test showed that the stem profile is affected by the genotype.

RESUMO As plantas invasoras podem modificar profundamente as características físicas e biológicas de seus novos ambientes, especialmente quando esses habitats já estão fragmentados e reduzidos pela pressão antropogênica, como a Mata Atlântica do Brasil. Aqui, nós hipotetizamos que as palmeiras exóticas se estabelecem com sucesso entre as populações naturais de Euterpe edulis através de entrada contínua de propágulos pela avifauna, altas taxas de germinação e rápido crescimento. Como resultado, a palmeira nativa está sofrendo declínio e pode estar ameaçada de extinção. A partir de 2007, realizamos um inventário florestal contínuo (FCI) a cada três anos nos fragmentos florestais primários e secundários de Viçosa, Minas Gerais. Usamos uma matriz de Markov para projetar futuras distribuições de palmeiras. A floresta secundária continha três espécies de palmeiras exóticas: Archontophoenix alexandrae, Livistona chinensis e Arenga caudata. A primeira palmeira é um sério risco para as populações naturais de E. edulis na Mata Atlântica, devido às interações freqüentes com as aves, à rápida germinação e à colonização agressiva nos estratos verticais inferior e médio da floresta. Atualmente, as populações naturais de E. edulis são viáveis e sustentáveis, capazes de regeneração, crescimento e frutificação, pois suas comunidades mantêm o fluxo gênico contínuo, dominando os estratos verticais da floresta em comparação com as palmeiras exóticas. No entanto, palmeiras exóticas devem ser monitoradas e medidas de controle devem ser analisadas, especialmente em áreas com populações de A. alexandrae.

ABSTRACT Invasive plants can profoundly modify physical and biological characteristics of their new environments, especially when such habitats are already fragmented and reduced by anthropogenic pressure, such as the Atlantic Forest of Brazil. Here, we hypothesized that exotic palms successfully establish among the natural Euterpe edulis populations through a continuous propagule input by avifauna, high germination rates, and rapid growth. As a result, the native palm is experiencing decline and may be threatened with extinction. Beginning in 2007, we conducted a continuous forest inventory (FCI) every three years in the primary and secondary forest fragments of Viçosa, Minas Gerais. We use a Markov matrix to project future distributions of palm trees. The secondary forest contained three exotic palm species: Archontophoenix alexandrae, Livistona chinensis, and Arenga caudata. The first palm is a serious risk to natural E. edulis populations in the Atlantic because of frequent interactions with birds, rapid germination, and aggressive colonization in the lower to medium vertical forest strata. Currently, natural E. edulis populations are viable and sustainable, capable of regeneration, growth, and fruiting, their communities maintain continuous gene flow, dominating vertical forest strata compared with exotic palms. However, exotic palms should be monitored and control measures should be analyzed, especially in areas with A. alexandrae populations.

RESUMO Vários métodos têm sido propostos para realizar a classificação de sítio para produção de madeira. No entanto, há necessidade frequente de se obter a capacidade produtiva do local antes mesmo do estabelecimento da floresta. Isto motivou a aplicação das Redes Neurais Artificiais (RNA) para classificação de sítio. Desta forma, o método tradicional da curva guia (CG) foi comparado com a RNA sem medidas do povoamento como variáveis preditivas. Além disso, diferentes configurações de RNA foram testadas. As variáveis utilizadas para treinamento da RNA foram: variáveis climáticas, tipos de solo, espaçamento e material genético. Os resultados obtidos pela RNA e CG foram comparados com as classes de referência, definidas utilizando altura dominantes de povoamentos observadas no sétimo ano de idade. A comparação foi realizada utilizando o coeficiente Kappa (K) e análise descritiva. Os resultados mostraram que a função de custo “entropia cruzada” e a função de ativação da camada de saída “softmax” foram melhores para este propósito. A classificação pela RNA resultou em uma “concordância substancial” com a classificação observada, contra uma “concordância moderada” gerada pela CG. A mudança no padrão de crescimento das árvores durante o tempo dificultou a classificação pela CG. Entretanto, este método mostrou alta eficiência na classificação quando dados em idades próximas à de referência estão disponíveis. Além disso este método pode ser aprimorado se outras técnicas avançadas de regressão forem aplicadas. Porém o método da RNA apresentado aqui não é sensível à instabilidade do crescimento e permite classificar locais sem registro de plantios anteriores.

ABSTRACT Several methods have been proposed to perform site classification for timber production. However, there is frequent need to assess site productive capacity before forest establishment. This has motivated the application of Artificial Neural Networks (ANN) for site classification. Hereby, the traditional guide curve (GC) procedure was compared to the ANN with no stand measures as input. In addition, different ANN settings were tested to assess the best setting. The variables used to train the ANN were: climatic variables, soil types, spacing and genetic material. The results from the ANN and the GC methods were compared to the observed classes, which were defined using the observed dominant high at the age of seven years. The comparison was performed using the Kappa coefficient (K) and descriptive analysis. The results showed that the cost function “Cross Entropy” and the output activation function “Softmax” were the best for this purpose. The ANN classification resulted in substantial agreement with the observed indices against a moderate agreement of the GC procedure. The change in growth patterns throughout the rotation may have hindered the proper classification by the CG method, which does not happen with the ANN. Moreover, the GC method shows efficiency on classification in cases which data from stands at the age close to the reference age are available. Also, it could be possible to improve its accuracy if another advanced regression techniques were applied. However, the ANN method presented here is not sensible to growth instability and allows classifying sites with no plantation history.