Abstract
Puerto Rico (PR) and the US Virgin Islands (VI), two archipelagos located in the Caribbean region, possess forest ecosystems with high species diversity, which remain largely underrepresented in forest modeling research compared to temperate forests. Accurately characterizing tree allometry is essential for monitoring forest status over time and predicting tree growth and yield in sustainable forest management. Height-diameter relationship models, denoted as H–D models, are widely used in forestry practice since they effectively reduce the time and cost of field data collection and provide quick estimates of the height of trees based on tree diameter at breast height. For diverse forests, mixed-effects models have often been proposed. In these models, fixed effects provide the predicted H–D curve for the population average (all species) while a calibrated prediction of H–D relationship for each species can be made with the inclusion of both fixed and random effect terms. This study developed species-specific H–D relationship models for different forest types across PR and VI. A total of 25 parametric model candidates published in the literature were examined for each forest type on each archipelago, and the best-fit model in each forest type was selected to build a mixed model for a given forest type and archipelago. The mixed model was then used to generate species-specific predictions of total tree heights. Tree data used were collected from the permanent plots of the USDA Forest Service’s Forest Inventory and Analysis (FIA) program. A total of 419 species with 51,166 observations were collected from eight forest types across the two archipelagos. The results highlight variations in H–D relationships across forest types and archipelagos, emphasizing the need for forest type–specific models to capture their unique ecological traits of species. Results show that three-parameter H–D models provide more accurate predictions than two-parameter ones. Carefully examining all possible combinations of placing the random effects when constructing mixed models is suggested, which can improve model predictability for capturing the H–D relationships for diverse species forests.