Machine learning algorithms promote reduction of operating costs in forestry experiments

This study applied artificial neural networks to estimate tree heights in experiments with clones and commercial planting spacing, seeking to replace complete measurements and reduce costs. The experiment included 84 plots with 2,677 trees, whose height and circumference data at breast height were collected. Simulations reduced samples by about 10% at a time, until reaching 14.5% of the measured heights, with the rest estimated by RNA. The trees were stratified by diameter classes at breast height, requiring at least one height observed per class in each plot. The variables used included genetic material, DBH, average and maximum DBH, Dg, average plot height and area per plant. The results showed that more than 97% of the deviations were between -10% and 10%, validating the accuracy of the method. In addition, a reduction of up to 40% in operating costs was observed with the adoption of estimates via RNA.