Whole-stand Models (WSM) have always been fitted with permanent plot data organised in a sequential age-matched database, i.e., i and i+1, where i = 1, 2, ... N plot measurements. The objectives of this study were (1) to evaluate the statistical efficiency of a monthly distributed data structure by fitting the models of Clutter (1963), Buckman (1962) in the version modified by A. L. da Silva et al. (2006), and deep learning, and (2) to evaluate the possibility of gaining accuracy in yield projections made from an early age to harvest age of eucalypt stands. Three alternatives for organizing the data were analyzed. The first is with data paired in sequential measurement ages, i.e., i and i+1, where i = 1, 2, ... N plot measurements. In the second, all possible measurement intervals for each plot were considered, i.e., ii+1; i, i+2; ...; iN; i+1, i+2; ..., N-1, N. The third has data paired by month (j), always with an interval of one month, i.e., j, j+1; j+1, j+2; j+M-1, M, where M is the stand age of the plot measurement in months. This study shows that the accuracy and consistency of the projections depend on the organization of the monthly distributed data, except for the Clutter model. A better alternative to increasing the statistical assumptions of the forecast from early to harvest age is based on a monthly distributed data structure using a deep learning method.