基于典型点的目的性采样设计方法及其在土壤制图中的应用

A Purposive Sampling Design Method Based on Typical Points and Its Application in Soil Mapping

鉴于经典采样和空间采样存在的局限性,提出了一种旨在寻找典型点的目的性采样设计方法。该方法通过分析与目标地理要素空间分布具有协同变化关系的环境因子,提取地理要素空间变化的典型模式,进而在典型模式上布设样点,即可获得典型点,从而减少所需样本量。以位于黑龙江鹤山农场的2个研究区为例,分别选择土壤厚度和表层有机质2个土壤属性,通过对土壤属性空间变化的4个协同环境因子进行模糊c均值聚类,获得对应土壤属性空间变化模式的环境因子组合;根据其模糊隶属度结果设计典型点并进行采样,最后结合典型点的属性值与环境因子组合模糊隶属度结果,采用加权平均模型得到土壤属性空间分布图,反映了土壤属性随地形变化的连续性分布。基于独立野外验证点,选择了4个评价指标对所得属性图进行定量评价。结果表明:2个研究区验证点集的预测值和观测值一致性指数均较高,可见本研究提出的方法是一种有效的样点布设方法。研究还对在每一环境组合类设计不同数量典型点所得土壤属性制图结果的差异进行了讨论,认为典型点增多并不一定能提高土壤属性空间推测的精度。

In consideration of limitations of classical sampling and spatial sampling, this paper proposed a purposive sampling method based on typical points. This method employed environmental factors which co-vary the target geographical element to find typical patterns of the geographical element over space. Field samples were then designed based on the locations of these typical patterns. It is believed that these field samples capture the overall pattern of spatial variation of the target variable well. With this approach, the number of field samples needed to characterize the spatial pattern of the target variable was reduced. The approach was applied in two study areas in Heshan Farm in Heilongjiang Province of China. Two soil properties were chosen, i.e. Ahorizon organic matter and soil thickness （each property for one study area）. Environmental combinations were generated by a fuzzy c-means clustering on four local environmental data layers and typical points were then designed with help of fuzzy membership maps of environmental combinations. Spatial variation of the two soil properties was inferred using a linear weighted average model with typical points＇ soil property values and fuzzy membership maps of environmental combinations. The mapping results reflected continuous changing of soil properties with terrain changing. Four indices were set up for evaluation of mapping results by using independently validation points. The evaluation results showed that, the agreement coefficients between observed values and predicted values of validation points were high for both of the study areas. We then concluded that the proposed sampling design approach was effective. Analysis was also conducted on difference between soil property mapping results by using different amount of typical points when different number of typical points representing each environmental combination. It was showed that using more points would not necessarily result in a more accurate map.