关键词: 杉木, 林分进界, 零膨胀模型, Hurdle模型, 混合效应模型

Abstract:

Objective: As one of the processes governing the dynamic changes in forests, tree recruitment is the basis to ensure the long-term maintenance of the structure and function of forest ecosystems. The recruitment model can predict the development dynamics of forest and quantify the future health and productivity of forest ecosystems which can provide the fundamental theoretical framework and data support for the conservation and restoration of natural forests. Method: Based on the 8th (2009), and 9th (2014) national forest inventory in Hunan Province, 784 natural Cunninghamia lanceolata forest sample plots were used to develop a tree recruitment model for natural secondary forests, incorporating stand factors, site factors, and climate factors. The basic model selected is the negative binomial (NB) model, taking into account the overdispersion in the recruitment data, zero-inflation model and Hurdle model are introduced to construct the zero-inflation Poisson (ZIP) model, zero-inflation negative binomial (ZINB) model, Hurdle-Poisson (HP) model and Hurdle-NB (HNB) model. Furthermore, to address potential autocorrelation and heteroscedasticity issues in the data due to repeated measurements and stratification, we introduced the county where the plots were located as a random effect in the five aforementioned models, constructing corresponding mixed-effects models. Finally, model validation was conducted using ten-fold cross-validation. Result: The basal area of the tree species (Bai), soil thickness (ST), elevation (EL) and mean annual precipitation (MAP) significantly influence the recruitment of C. lanceolata forests. The performance of the negative binomial composite models (NB, ZINB, and HNB) in simulating natural C. lanceolata forests recruitment is notably superior to that of the Poisson composite models (ZIP and HP). The fitting performance of ZINB and HNB outperforms NB, with ZINB showing a slight advantage over HNB. The introduction of county-level random effects resulted in likelihood ratio test outcomes that indicate improved goodness of fit for NB, ZIP, ZINB, and HP models compared to the base model, except for the HNB model. Among these, the ZINB mixed-effects model exhibited the best fitting performance, a conclusion supported by the results of ten-fold cross-validation. Conclusion: The climate-sensitive natural C. lanceolata forests recruitment model constructed in this research holds biological significance and statistical reliability. It can provide scientific guidance for the implementation of ecological restoration measures and silvicultural practices in the region’s natural secondary forests, thereby enhancing stand quality and contributing to the timely achievement of“carbon neutrality and peak”strategic goals.

Key words: Cunninghamia lanceolata, tree recruitment, zero-inflated model, Hurdle model, mixed-effects model