A model evaluation framework applied to the Forest Vegetation Simulator (FVS) in Colorado and Wyoming lodgepole pine forests

•We evaluated the Forest Vegetation Simulator (FVS) Central Rockies Variant.•Bakuzis matrix exhibited qualitatively reasonable default model behavior.•Chronosequence validation found default over-prediction of stand biomass.•Equivalence testing found unrealistic growth of non-lodgepole pine species.•Sensitivity analysis identified initial species composition drives output. The Forest Vegetation Simulator (FVS) growth and yield model is widely used throughout the United States, but recent studies have reported unexpectedly large bias for some regional model variants. Here we propose a general framework for model evaluation, designed to highlight model strengths and weaknesses and inform calibration efforts. We apply the framework to evaluate the Lodgepole Pine (LP) model of the FVS Central Rockies Variant (FVS-CR), which has rarely been evaluated in the literature despite its widespread use in the western US. We started with a qualitative verification of the structure and logic of the FVS-CR LP model against a modified Bakuzis matrix, determining that it adequately reproduces known patterns of stand dynamics. We then compared stand-level growth simulations to a chronosequence developed from 554 Forest Inventory and Analysis (FIA) plots measured in Colorado and Wyoming lodgepole pine forests. This quantitative validation exercise revealed that the default model settings substantially over-predict basal area and total stand carbon after 50 years for both pure lodgepole pine stands and mixed lodgepole pine stands containing a minor proportion of Engelmann spruce, subalpine fir, and quaking aspen. Using equivalence testing to validate the large-tree diameter increment model against a separate dataset of 3,790 remeasured trees on 124 lodgepole pine FIA plots, we found that the default FVS–CR LP model adequately predicts lodgepole diameter growth, though we observed large variation in model errors. Equivalence tests also revealed systematic under-estimation of spruce–fir and over-estimation of aspen productivity. Finally, we conducted a sensitivity analysis to identify the most important model parameters and data inputs driving simulated stand structure and carbon accumulation in both the short- (50 years) and long-term (200 years). The model shows the greatest sensitivity to initial species composition; a small proportion of aspen, spruce, or fir seedlings led to long-term stand re-structuring and greatly increased carbon accumulation. Other sensitive parameters