Examining strategies to project tree diameter for unobserved species in diverse tropical forests using mixed-effects models

Mixed-effects models with species as a random effect have provided a practical solution to produce reliable predictions of tree growth. Applying them to new datasets can be challenging because species-specific adjustments are not automatically available for “new” species (i.e., species that are not “observed” in model training). In general, there are four strategies used for unobserved species when applying mixed models: (1) generating predictions using only the fixed-effects portion of the model, (2) computing species-specific adjustments post hoc when limited observations for the new species are available, (3) building two separate models with and without species as a random effect, and (4) combining data grouping with mixed modeling. To our knowledge, the relative efficacy of these strategies has not been explicitly examined for diverse, species-rich forests in the Caribbean. Long-term data collected by the US Department of Agriculture (USDA), Forest Service, Forest Inventory and Analysis (FIA) program in Puerto Rico and the U.S. Virgin Islands over the past 20 years were used in this study.Results show that all strategies examined can be applied when predicting future diameters for new species. However, approaches that include adding species information tended to provide more precise predictions. Post-hoc calculation of species-specific adjustments is suggested as the predictions can be calibrated, but this strategy requires at least one repeated measurement for new species. Implementing a hybrid method of mixed modeling and data grouping can also integrate species-level information into prediction. This work provides an example of selecting the proper methodology to handle rare species to facilitate the continuous monitoring of diverse and vulnerable forest resources.