Considering that the forest ecosystem is facing severe challenges with the intensification of global climate change, this study takes climate and soil factors as the main factors affecting the probability of tree death, establishes a model of the probability of tree death of Larix principis-rupprechtii, and explores the interaction between climate factors and soil factors and the probability of tree death.

Taking the Larix principis-rupprechtii forest in Boqiang forest farm, Wutai mountain, Shanxi Province as the research object, the long-term climatic factors data, the sample plot per tree gauge data and the data of related soil factors were collected. After the equilibrium dataset was optimized by SMOTE (Synthetic Minority Over-sampling Technique) sampling, The stepwise regression method was used to screen out the key predictors of variance expansion factor (VIF) less than 5, and the prediction model of the death probability of Larix principis-rupprechtii was established by Logistic regression and Bayesian Logistic regression methods.

The results showed that the four environmental factors of crown width, soil pH value, soil NH₄⁺-N concentration and MCMT (mean coldest month temperature) contributed significantly to the mortality of Larix principis-rupprechtii. From the ranking of factors influencing variable importance, there is no significant difference between Logistic model and Bayesian Logistic regression model, among which the importance factors of the variables were the highest in soil pH and crown width, and the importance of MCMT and soil NH₄⁺-N concentration was low. The crown width and MCMT factors were negatively correlated with the mortality probability model, and with the growth of crown amplitude and the increase of monthly mean coldest temperature, the probability of tree death decreased. Soil pH value and NH₄⁺-N were positively correlated with the withering probability model, and the probability of tree death increased with the increase of the disease. In model performance evaluation, the logistic regression model demonstrated better overall performance, with accuracy, sensitivity, F₁-score, and Kappa coefficient values of 0.885, 0.797, 0.875, and 0.771, respectively. The Bayesian logistic regression model performed better in terms of precision and specificity, achieving 0.979 and 0.983, respectively. Moreover, both models achieved AUC-ROC values above 0.9, indicating strong predictive accuracy.

Climate and soil factors have significant contributions to the probability of tree death, and the resulting model has good prediction accuracy, so it is of great significance to consider climate and soil factors in the probability model of tree death. The model can provide a scientific basis for forest management and ecological protection in north China, help formulate more reasonable and perfect forest health monitoring and emergency response measures, and provide a reference for the sustainable management of Larix principis-rupprechtii forests in north China.

Elevational gradient is a critical environmental factor affecting forest carbon storage distribution in mountainous ecosystems, yet the nonlinear mechanisms and optimal elevation zones remain unclear. This study aims to reveal the distribution patterns of forest carbon storage along elevational gradients in Larix principis-rupprechtii forests, compare the predictive performance of different modeling approaches, and to explore the application potential of LiDAR technology in forest carbon storage assessment.

Based on field survey data from 35 plots and LiDAR data, this study applied linear regression, polynomial regression, mixed-effects models, and generalized additive models (GAM) to model and compare the relationship between carbon storage and elevation. Leave-one-out cross-validation was used to evaluate model predictive performance, and standard evaluation metrics such as the coefficient of determination (R²), standard error of estimation (SEE), and mean percentage error (MPE) were calculated.

Forest carbon storage of Larix principis-rupprechtii exhibited a significant inverted U-shaped distribution pattern along the elevational gradients (1 900-2 500 m), with an optimal elevation zone of 2 000-2 200 m and a theoretical optimal elevation of 2 165 m. Elevation was the dominant factor explaining carbon storage variation, accounting for 89.5% of the variance (η²=0.895, P<0.001). Among the four modeling approaches, GAM performed best (fitted R²=0.907, cross-validation R²=0.888, prediction standard error= 10.15 t/hm²) with only 1.86% overfitting. LiDAR-derived canopy density (Intensity_CV) emerged as an important predictor, significantly improving model prediction accuracy.

This study confirmed the nonlinear effects of elevational gradient on forest carbon storage, with GAM demonstrating clear advantages in modeling complex ecological relationships. LiDAR technology shows great potential in forest carbon storage assessment, providing technical support for establishing mountain forest carbon sink monitoring systems. The research findings will provide scientific basis for adaptive forest management under climate change and precision forest management based on elevational gradients.