Effects of stand factors on tree growth of Chinese fir in the subtropics of China depends on climate conditions from predictions of a deep learning algorithm: A long-term spacing trial

Stand and climate related variables are the main driving forces controlling individual tree growth. Two machine learning algorithms called deep learning and random forest were used to explore how annual diameter growth varied with stand and climatic variables. Data was obtained from a long-term spacing trail of Chinese fir (Cunninghamia lanceolata (Lamb.) Hook.) plantations in four provinces of southern China. Results from model comparisons showed the deep learning model with 8 hidden layers and 90 neurons in each hidden layer achieved the best performance, and the RF model ranked 4th among 9 selected models. In addition, sensitivity analysis showed that individual tree growth increased with an increase in Gini coefficient, while growth decreased with an increase in stand age (A) and the basal area of larger trees (BAL). The relationships between diameter growth and summer mean maximum temperature (SMMT), as well as winter mean minimum temperature (WMMT) and annual precipitation (AP) were not constant, which depended on the range of values of each climate factor. BAL had the greatest influence on diameter growth among all the variables. From an interaction analysis, we found that climate factors exacerbated the negative effects of competition on growth. Climate change promoted the growth of younger trees but restrained the growth of older trees. With climate variables considered, tree growth under high and middle stand structural heterogeneity were similar, and observably higher than that with low stand structural heterogeneity. Positive influences of climate tended to promote tree growth under lower competition and older individuals were more vulnerable to WMMT changes. Our findings enhance our understanding of the mechanisms driving individual Chinese fir growth in southern China in the face of future climate uncertainty.