应用模糊c均值聚类获取土壤制图所需土壤-环境关系知识的方法研究

EXTRACTION OF KNOWLEDGE ABOUT SOIL-ENVIRONMENT RELATIONSHIP FOR SOIL MAPPING USING FUZZYc-MEANS(FCM) CLUSTERING

在没有土壤普查专家及土壤图的地区,获取土壤环境间关系的知识是基于知识进行预测性土壤制图中的关键问题。本文建立了一套应用模糊c均值聚类(Fuzzyc-means,FCM)获取土壤环境间关系知识的方法:得到对土壤形成发展具有重要作用的环境因子,建立环境因子数据库;对环境因子进行模糊聚类,得到环境因子组合隶属度分布图;根据隶属度值确定野外采样点;将环境因子组合与土壤类型对应,进而提取土壤-环境关系知识。为检验该方法的有效性,应用所得知识进行土壤制图,通过独立采样点对土壤图进行精度评价。本文在黑龙江鹤山农场一个研究区的应用结果表明,该方法仅需要少量的野外采样即可获得有效的土壤-环境关系知识,为预测性土壤制图提供必需的依据,同时也显著提高了野外采样的效率。

For areas with no soil survey experts or soil maps available, knowledge about soil-environment relationships is a key to predictive soll mapping using knowledge-based approaches. An approach based on an unsupervised fuzzy clustering method （FCM, fuzzyc-means clustering） is recommended to acquire the knowledge. The method consists of four steps： 1） define those environmental factors which play decisive roles in formation and development of soil, then build up environmental database; 2） identify environmental niches （combination of environment conditions） by running FCM on the environmental database; 3） allocate field efforts to relate unique environmental factor combinations to soil types; 4） establish a soil-landscape model by analyzing the relationship established from the field samples. The model is then used to predict spatial distribution of soils. The method was applied in a study area in Northeast China and a soil map at the subgroup level was created for the area. The derived soil map was accurate up to 72% when verified using an independently collected field sample set. Comparing this method with the conventional soil mapping method （50% -60% ） in USA in accuracy, it could be concluded that the approach is effective for prognostic soil mapping over areas with no local soil experts available. At the same time, the clustering method can improve efficiency of the field sampling.