Objective Chinese fir (Cunninghamia lanceolata) is a key timber species in Fujian Province, China. To address the issue of low prediction accuracy for individual tree mortality at the Jiangle State-owned Forest Farm, this study constructs a deep learning-based diagnostic model, providing a reference for the scientific tending of Chinese fir.

Method Using data from 79 permanent sample plots from 2011 to 2024, SMOTE, MAHAKIL, and hybrid oversampling strategies were employed to address class imbalance issues. A deep learning-based individual tree mortality model was established. Model performance was evaluated using multidimensional metrics such as F1 score, AIC, and BIC, and further compared with traditional logistic regression models to assess performance and stability.

Result The hybrid sampling strategies significantly improved model performance. The MAHAKIL hybrid oversampling model achieved mean AIC (874) and BIC (949) that were superior to those of the SMOTE hybrid oversampling model (AIC=1 038, BIC=1 117). The optimal strategy was the MAHAKIL-R300 hybrid sampling, with an F1-score of 0.959 and a critical threshold set at 0.646. The performance of the deep learning individual tree mortality model was significantly better than that of the traditional logistic regression model. (2) Among the eight influencing factors, the top three factors contributing the most to individual tree mortality of Chinese fir were, in order, diameter at breast height (DBH), Gini coefficient of DBH, and the total basal area of trees with DBH greater than that of the subject tree (C1). The Gini coefficient of DBH and C1 were positively correlated with mortality probability, while DBH was negatively correlated with mortality probability, indicating that tree competition and structural imbalance significantly increased the risk of individual tree mortality.

Conclusion The individual tree mortality model for Chinese fir constructed in this study, based on oversampling techniques and deep learning, demonstrated higher prediction accuracy and stability, effectively identifying key influencing factors of mortality. This model can provide a reliable decision support tool for the precision nurturing and scientific management of Chinese fir forests.