庞泉沟自然保护区华北落叶松林的自组织特征映射网络分类与排序

自组织特征映射网络（SOM）是新近引入植物生态学的分析方法，对复杂问题和非线性问题具有较强的分析和求解功能。本研究应用SOM分类和排序研究了庞泉沟自然保护区华北落叶松林。研究结果表明，SOM将120个样方分为7个植物群落类型，分类结果具有明确的生态意义；样方和物种在SOM训练图上呈现一定规律的分布；7个群落类型各有其分布范围和界限，揭示了群落间的生态关系。在此基础上，通过引入一种在SOM训练图上可视化环境因子梯度的方法，能够较好地完成样方、物种和环境因子相互关系的分析，揭示了海拔是影响该区华北落叶松林生长和分布的最主要因子。生态分析表明SOM分类和排序是一种有效的梯度分析方法，适用于表征生态特征和探索群落和环境相互关系的研究。

Self-organizing feature map classification and ordination of Larix principis- rupprechtii forest in Pangquangou Nature Reserve

Understanding plant communities with respect to environmental features is a fundamental basis for vegetation ecosystem management. But ecological data are always bulky, non-linear and complex, showing noise, redundancy, internal relations and outliers. There are also wide variability in variables and complex interactions between explanatory and response variables. Namely, non-linear analyzing methods should be preferred for the study of plant community. Self-organizing feature map (SOM) is a comparatively new tool for data analysis, and it could be effectively applicable to classification and association. Theoretically, it can describe natural phenomena and rules better. It can distribute information within the whole network with variation of weights and problems for some units cannot affect the network function. Therefore, it is suitable for analysis of vegetation ecosystem which has attracted much attention from ecologists. In the present work, SOM was applied to study Larix principis-rupprechtii forest in Pangquangou Nature Reserve through clustering and ordination. Pangquangou Nature Reserve is located in the midst of Luliang Mountain range, at 37°45'-37°55' N, 111°22'-111°33'E. It was established for the conservation of the first-class nationally protected bird, Crossoptilon mantchuricum, and the cold-temperate coniferous forests (Larix principis-rupprechtii forest and Picea forest). The diversity of plant communities is the basis for conservation of endangered animals and plants. In this field research, 120 samples (quadrats) of 10 m×10 m for Larix principis-rupprechtii forest along with relevant environmental factors were set up, and species data was recorded in each sample. 105 major species and 6 environmental variables (elevation, slope, position of slope, aspect of slope, soil thickness and litter layer thickness) were used to implement the method. Classification and ordination was carried out by using SOM toolbox in MATLAB (7.0). As a result, the SOM showed a high performance for visualization and abstraction of ecological data. The trained SOM efficiently classified 120 samples into 7 groups representing 7 types of plant communities according to a gradient of species Importance Value, and displayed a distribution of each sample and species. The result of SOM classification shows significant ecological meanings, and the characteristics of each community were described. The species component planes helped to interpret the contribution of each species to the classification. The obtained groups showed clear boundaries on the trained SOM, indicating SOM ordination could reveal the ecological relationship between them. Additionally, by proposing a method to visualize environmental variables not used in its training phase, the SOM showed high performance in analyzing the relationships among samples, biological variables and environmental variables, and revealed that elevation is the most important factor affecting the growth and distribution of Larix principis-rupprechtii forest. This method could be used as a tool to extract relationships between sampling sites, communities, and environmental variables, although the algorithm is theoretically an indirect gradient analysis. However, it remains necessary to quantify the relationships among variables. Finally, approach using SOM classification and ordination showed that it could take into account the variability of ecological data efficiently. Therefore, SOM is a very effective quantitative technique in plant ecology research, and this procedure could be preferred when ecological modeling is applied to understand non-linear and complex ecological data.