The objective of this work was to obtain regression equations and to indicate the most appropriate from different mathematical models for the estimation of the leaf area of ​​ Allspice (Pimenta dioica) by non - destructive method. 500 leaves of plants located in the municipality of São Mateus, North of Espírito Santo State, Brazil, were collected, 400 of which were used to adjust the equations and 100 for validation. The length (L) along the main midrib, the maximum width (W), the product of the length with the width (LW) and the observed leaf area (OLA) were measured from all leaves. We fitted models of linear equations of first degree, quadratic and power, where OLA was the dependent variable in function of L, W and LW. From the 100 sheets intended for validation, and using the adjusted equations for each mathematical model, the estimated leaf area (ELA) was obtained. Subsequently, a simple linear regression was fitted for each model of the proposed equation in which ELA was the dependent variable and OLA the independent variable. The mean absolute error (MAE), the root mean square error (RMSE) and Willmott's index d also determined. The best fit had as selection criterion the non-significance of the comparative means of ELA and OLA, MAE and RMSE values ​​closer to zero and value of the coefficient of determination coefficient (R2) close to one. Thus, the power model (ELA = 0.7605(LW)0.9926, R2 = 0.9764, MAE = 1.0066, RMSE = 1.7759 and d = 0.9950) based on the product of length and width (LW) is the most appropriate for estimating the leaf area of ​​Pimenta dioica.

ResumoO processo Kraft é um método de preparação de pasta celulósica cuja principal vantagem é a recuperação dos produtos químicos a ele associados. O sistema de caustificação faz parte do ciclo de recuperação química desse processo e, mesmo que esse sistema seja completado nos Caustificadores, grande parte da cal é convertida em CaCO3 no Extintor. O presente trabalho tem por objetivo gerar modelos empíricos com o auxílio da técnica estatística regressão linear múltipla a fim de tentar descrever de maneira mais eficaz a eficiência do Extintor. A partir de dados operacionais fornecidos por uma indústria de celulose e utilizando-se também do software STATISTICA 7 que realizou os tratamentos estatísticos, foram gerados modelos estáticos e dinâmicos, mesmo sabendo que o processo de caustificação ocorre de maneira dinâmica. Nos modelos dinâmicos, a técnica utilizada foi a inserção da variável Eficiência de Caustificação como entrada atrasada, ou seja, adicionou-se valores de EC (lidos anteriormente) aos dados a serem previstos. Os resultados obtidos são apresentados ao final de cada modelo de regressão por meio dos coeficientes de determinação e também por gráficos para os dados de validação. Todos os modelos dinâmicos forneceram resultados superiores quando comparados aos estáticos, descrevendo de maneira satisfatória o Extintor. Palavras chaves: modelagem computacional, extintor da caustificação, regressão linear múltipla AbstractThe Kraft process is a method of preparing pulp whose main advantage is the recovery of chemicals associated therewith. The causticizing system is part of the chemical recovery cycle of this process, and even though this system is completed in Causticizers, most of the lime is converted into CaCO3 in the Slaker. This work aims to generate empirical models with the help of statistical technique multiple linear regression to try to describe more effectively the efficiency of the slaker. From operational data from a cellulose industry and also using up the STATISTICA 7 software that performed the statistical procedures were generated static and dynamic models, even though the causticizing process occurs dynamically. In dynamic models, the technique was used to insert the variable Causticizing Efficiency as late entry. In other words, added values of EC (previously read) in the data to being provided. The results obtained are presented at the end of each regression model by the determining coefficients and also by the graphics to the validation data. All dynamic models provide superior results when compared to static models, describing satisfactorily the slaker.