Benchmarking and calibrating FVS diameter growth predictions with tree-rings and forest inventory data in Sierra Nevada mixed-conifer forests

Accurate prediction of tree growth is fundamental for evaluating forest dynamics and management outcomes. The Forest Vegetation Simulator (FVS) is among the most widely used growth and yield models in the United States, yet its empirical performance remains uncertain for many western forest types. Here, we integrated annual-resolution tree-ring measurements with long-term inventory data to benchmark the Western Sierra Nevada variant (FVS-WS) across four dominant tree species (Abies concolor, Calocedrus decurrens, Pinus lambertiana, and Pinus ponderosa). We quantified prediction errors in diameter growth and identified the main drivers of model bias. The default model consistently underestimated diameter at breast height (DBH) growth across all species, with bias ranging from −38% to −49% and RMSE ranging from 13.1 to 21.4 cm (64–87% relative RMSE). Prediction errors were primarily structured by tree size and stand development, with variables such as DBH and stand density metrics explaining most of the variation, whereas climate and topographic variables contributed comparatively little. To correct these systematic biases, we implemented a species-specific multiplicative calibration using Random Forest models. Calibration substantially reduced prediction error across all species: overall RMSE decreased from 17.2 cm (66%) to 3.1 cm (12%), mean bias decreased from −40.6% to −3.4%, and the proportion of predictions within ±15% of observed DBH (PA15) increased from 10% to 68%. Species-level RMSE gains ranged from 75% to 89%, with the largest improvements observed in A. concolor and P. ponderosa. These results indicate that model error arises largely from limitations in representing size-dependent growth and structural heterogeneity. Integrating tree-ring-based growth reconstructions with forest inventory data provides a transferable framework for identifying and correcting systematic bias in FVS, enhancing its reliability for applications in forest planning, fuel treatment assessment, and carbon accounting in Sierra Nevada mixed-conifer forests.