Abstract: Modeling mixed forest growth is challenging due to multiple tree species, uneven-aged structures, and complex environmental interactions. Traditional transition matrix models, widely used for predicting ingrowth, upgrowth, and mortality in mixed forests, rely on statistical approaches but have limited capacity to handle high-dimensional data. Machine learning has the potential, but there is limited application and a lack of comparison, especially in terms of model accuracy and biological plausibility. This study developed a climate-sensitive transition matrix growth model with the integration of machine learning (ML-Matrix) for mixed conifer-broad-leaved forests in northeastern China. Forest data were obtained from permanent plots in the 5th–9th National Forest Inventories. We evaluated five machine learning algorithms—random forest (RF), boosted regression trees (BRT), artificial neural network (ANN), support vector machine (SVM), and k‑nearest neighbor (KNN)—to model upgrowth, two-stage mortality, and recruitment. Results showed BRT excelled in modeling upgrowth of Pinus koraiensis and the first-stage mortality, while ANN excelled for upgrowth of other tree species. For the second-stage mortality, ANN performed best for two broad-leaved groups, while SVM outperformed for three coniferous groups. Optimal recruitment algorithms (SVM, BRT, or ANN) varied by species. Stand variables were the most important predictors with the relative important values of 85.1%–99.8% for upgrowth, 49.0%–88.4% for mortality, and 46.4%–84.3% for recruitment. Compared to the traditional statistical transition matrix (TS-Matrix), ML-Matrix achieved higher test-set accuracy for four tree species groups, excluding Pinus koraiensis. However, long-term simulations revealed TS-Matrix produced more biologically reasonable prediction of stand density, basal area and volume. This study highlighted the potential of ML-Matrix and the need for caution with its long-term projections for mixed forests under climate change.

Key words: Forest dynamics, Machine learning, Upgrowth, Ingrowth, Mortality, Climate change