Abstract Local calibration methods were evaluated for diameter at breast height (dbh) growth predictions in 11 tree species in Virginia, USA, using a model form based on the Forest Vegetation Simulator Southern Variant (FVS-Sn) large tree dbh regression model. Data from 1090 remeasured forest inventory plots from the USDA Forest Service’s Forest Inventory and Analysis (FIA) database were used to calibrate FVS dbh growth predictions to local conditions and evaluate four calibration methods based on the following information: 1) median prediction errors calculated from locally observed dbh pairs before and after a five-year remeasurement period; 2) a random intercept estimated from locally observed dbh using mixed-effects regression; 3) a simple linear regression (SLR) model fitted to observed and predicted dbh at the local scale; and 4) an SLR model with regression through the origin. Calibration methods were assessed using leave-one-out cross-validation, comparing model predictions to observed dbh growth from withheld trees. Equivalence testing indicated median or regression-based local calibration methods achieved prediction-error tolerances over 5–7 year growth intervals as small as 0.11 cm (0.03 cm for two regression-based methods) for all species, whereas the random-intercept approach only achieved a minimum tolerance of 0.2 cm. Compared to uncalibrated models, local calibration substantially reduced prediction errors, demonstrating efficacy in increasing prediction accuracy, even with sparse FIA dbh growth data used for local-calibration.
Çankaya et al. (Mon,) studied this question.