Mixed models for biomass prediction in the semiarid zone of Pernambuco State, Brazil

Abstract

Most of the datasets used for biomass modeling in native forests have high variability, containing several sample plots of different sites, trees of different species in the same plot, and several measurements of the same tree along the bole. Assuming that data from uneven-aged forests are highly diversified, the mixed-effects modeling analyzes hierarchically structured data more efficiently than any other approach and increases the prediction accuracy of the equations. The objective of this paper was to develop a mixed model to predict the above-ground biomass for Caatinga species in the municipality of Floresta - PE, Brazil. Biomass data of 100 trees belonging to five species were used for this purpose. The model of Schumacher and Hall was fitted considering the structure of a mixed linear model by the inclusion of intercepts and random coefficients, considering the species as the random effect. The best model was selected based on the root mean square error, bias, mean absolute error, Akaike information criterion, and residual graphic analysis. Species as a random effect contributed to increasing the accuracy of the estimates in a mixed model. The Schumacher and Hall model with random effects on dbh (diameter at breast height) and h (height) was the best way to predict the total biomass of the trees in the Pernambuco semiarid region.