Abstract: Self-organizing map (SOM) technique has unique advantage in multi-dimentional soil data analysis, because of its powerful nonlinear fitting ability and visualization.This paper analyzed the topsoil multi-dimentional data by SOM technique and carried out studies on applying artificial neural network (ANN) in environment area. The region of Wusu-Kuitun-Dushanzi-Shawan in Xinjiang was selected as the study area, and then 199 topsoil samples were collected and 22 soil properties were analyzed. The clustering analysis of soil samples, sampling points optimization and correlation of soil properties were studied by SOM technique. The results showed that the soil in study area was significantly affected by human activities. Except pH, the other 21 soil property data obeyed normal distribution or lognormal distribution. The soil samples were divided into 42 groups according to soil internal correlation and similarities of soil properties. According to the results of grouping situation and spatial distribution characteristics of sampling points, 24 sampling points with similar soil properties and spatial adjacent were optimized. Certain correlation existed between soil properties, such as: there was negative correlation between OM and pH, while positive correlation between OM and ωH2O; there was positive correlation between OM and As, Cr, Mn, Cu, Zn, Pb, Ni, Al, Co, Fe, Mo, Ti; but there was negative correlation between OM and K, Na, Sb, and positive correlation between OM and Cr, Mn, Cu, Zn, Co, Fe. There was positive correlation between Cd and Ni; there was higher positive correlation between Al and Ti; but there was no conspicuous correlation between V, Hg, Pb and Sb.

Key words: self-organizing characteristic mapping, multi-dimentional soil data analysis, sampling points optimization, correlation analysis

摘要： 自组织特征映射神经网络技术（self-organizing map，SOM）因其强大的非线性拟合能力和可视化特点在土壤多维数据分析中具有独特的优势。将SOM技术应用于表层土壤多维数据分析，进而开展人工神经网络技术（artificial neural network，ANN）在环境领域的应用研究。以乌苏-奎屯-独山子-沙湾地区为研究区域，共采集199个表层土壤样本，分析得到22个土壤属性。利用SOM技术开展了土壤聚类分析、采样点优化、土壤属性相关性等研究。结果表明：研究区域土壤属性变异系数较高，土壤受人类活动影响明显，除pH外，其余21个土壤属性数据分布服从正态分布或对数正态分布。依据土壤内在的相关性及土壤属性的相似性，将土壤样本分为42组，综合考虑分组情况和采样点的空间分布特点，将土壤属性显著相似且空间相邻的24个采样点进行优化处理。土壤属性之间呈现一定的相关性，如OM与pH呈负相关性，而与ωH2O呈正相关性；OM与As、Cr、Mn、Cu、Zn、Pb、Ni、Al、Co、Fe、Mo、Ti呈正相关性，但与K、Na、Sb呈负相关性；Cr、Mn、Cu、Zn、Co、Fe之间呈正相关性；Cd和Ni呈正相关性；Al和Ti具有较高的正相关性；而V、Hg、Pb、Sb相互间无显著相关性。

关键词: 自组织特征映射, 多维土壤数据分析, 采样点优化, 相关性分析