Abstract
Tree height and crown area are important predictors of aboveground biomass but difficult to measure on the ground. Numerous allometric models have been established to predict tree height from diameter (H–D) and crown area from diameter (CA–D). A major challenge is to select the most precise and accurate allometric model among existing ones, depending on the species composition and forest type where the model is to be applied. To propose a principle to select tree H–D and tree CA–D allometric models, we build a method based on k -fold cross-validation using a large dataset spanning six forest types from central Africa. We then compared the errors and biases using 22 previously established H–D and CA–D allometric model forms via three inter-comparable scenarios: locally derived for the forest type vs. regional vs. pantropical; regional (encompassing the forest type) vs . pantropical; regional (not encompassing the forest type) vs. pantropical model. H–D allometries were more variable across forest types in central Africa than CA–D allometries: (i) forest type explained 6% of the variance in H–D allometry and 2% of the variance in CA–D allometry, while species explained 9% and 2% of the variance in H–D allometry and CA–D allometry, respectively; (ii) for H–D allometry, the six forest types resulted in five best-fit models whereas, for CA–D allometry, four models provided the best fit for the six forest types. We recommend using allometric models specific to the forest type, preferentially to regional ones. Regional models should in turn be preferred to pantropical allometric models.