Objective This study aims to develop crown radius models for Larix gmelinii based on the theory of generalized additive model (GAM), and compares the predictive accuracy with the aggregation for crown radius and crown width, and providing theoretical foundations and practical guidance for predictions of crown radius and crown width in Larix gmelinii.

Method The research subjects were 3 444 Larix gmelinii trees from 68 natural forest plots in the Greater Khingan Mountains of Heilongjiang Province, northeastern China. From eight crown width-diameter base model fitting results, the model with the smallest Akaike information criterion (AIC) and Bayesian information criterion (BIC) was selected as the base models for each crown radius. Single-tree and stand factors were introduced into the optimal base models to construct the generalized models. Based on the constructed generalized models, the aggregation and GAM theory were used to build a system of compatible models for each crown radius.

Result (1) The base model fitting results indicated that the optimal base models varied for crown radii in different directions. (2) Introducing variables such as height to the crown base, basal area, and quadratic mean DBH into the base models for different crown radius directions all improved the model fitting effects. Subsequently, generalized models containing single-tree size and competition variables were constructed for each crown radius. (3) The comprehensive comparison of compatible models of crown radius and crown width based on the aggregation and GAM methods showed that GAM had better fitting effects and predictive accuracy, and the predictions for both crown radius and crown width was better than that of aggregation method.

Conclusion Each crown radius of Larix gmelinii exhibits different growth trends. In the crown radius model for Larix gmelinii, the predictive accuracy of GAM is superior to that of aggregation. GAM not only does not require strict model assumption but also simplifies the selection process between predictor and response variables. Therefore, from the perspectives of model assumptions and application convenience, GAM is recommended for predicting crown radius and crown width in this region’s Larix gmelinii forests.