Abstract:

Objective: Cunninghamia lanceolata is widely distributed in China, and the growth speed is quite different in different areas. The spatial clustering method was used to group the distribution area of C. lanceolata, and the growth model of C. lanceolata was established in different groups, which provided a method for the prediction of nationwide C. lanceolata high-precision growth. Method: 16 provinces(autonomous regions)where C. lanceolata was distributed were selected as study areas. Based on seventh and eighth continuous inventory of forest resources in China for the retest data of fixed plots of C. lanceolata and the topography, soil, meteorology and other environmental data of the study area, the importance of environmental factors affecting the growth of C. lanceolata were analyzed by the random forest model. Five environmental factors with great influences on the growth of C. lanceolata were selected. Using the group analysis function of ArcGIS 10.2, the C. lanceolata in the study area was grouped according to the environmental similarity and spatial proximity. The accumulation growth rate model of grouped and ungrouped was respectively established. Taking the national ungrouped model as a reference, the five indicators including coefficient of determination(R²), root mean square error(RMSE), mean relative error(MRE), systematic error(SE)and residual standard deviation(S)were used as the evaluation indexes of the model to analyze the modeling results. Result: The top eight environmental factors those have great influences on the growth of C. lanceolata by random forest model analysis were bio4(standard deviation of seasonal variation of temperature), elevation, bio3(isothermality), bio8(the wettest quarterly mean temperature), bio1(annual mean temperature), bio14(the most dry month precipitation), bio12(annual mean precipitation), and bio2(monthly mean diurnal temperature variation). The results of the group analysis showed that when the research area were divided into 7 groups, the internal environment similarity within the group could be maximized, and the environmental similarity between groups was the smallest. As only four of the 7 groups had data of C. lanceolata plot, the growth rate model of C. lanceolata stands was established using two model formulas in these four groups. Compared with the ungrouped models, the determination coefficients of the grouping models were all increased by more than 0.1, indicating that the fitting degree of the grouping model was better. At the same time, the accuracy of grouping modeling was also significantly improved:RMSE was reduced by about 0.5, MRE was reduced by about 6%, SE was reduced by about 3%, and S was reduced by about 1. Conclusion: The growth model based on the grouping of C. lanceolata study areas according to the environmental similarity is a high-precision forecasting method for C. lanceolata growth nationwide. The model could be used to estimate the growth of C. lanceolata stand and to update the data of forest subcompartment. The proposed method might provide a new method for realizing large area growth prediction of main plantation species.

Key words: spatial clustering, random forest model, Cunninghamia lanceolata, growth model