基于样本分级的土壤属性自适应回归拟合方法

A Novel Self-adaptive Regression Model for Prediction of Soil Attributes Based on Sample Classification

精细土壤属性信息在诸多领域均具有广泛的应用,历来倍受关注。现有土壤属性预测方法具有适用性不强或需要大量人工经验和专家知识等缺点,限制了这些方法在实际应用中的推广。本文提出了一种土壤属性自适应预测方法,可分为4步:①对采样点进行分组处理;②利用回归模型构建各分组内土壤属性与主导环境因子之间的典型关系;③对分组方案进行自动优化;④利用各组对应的土壤–环境因子典型关系对研究区进行优化拟合预测。为了验证方法的有效性,本文在我国东北典型黑土区以土壤有机质含量为例进行了应用研究,结果表明:所提方法可对环境因子做出自动选择,并可通过优化拟合对土壤属性空间分布进行自适应预测,预测精度较高。本方法初步实现了土壤属性预测的自动化,具有较好的适用性。

Detailed and accurate soil attribute information has received growing attention and has increasingly been applied in various related fields. The majority of existing digital soil mapping approaches is effective only in specific condition while others require much expert knowledge and manual intervention. A new self-adaptive method for soil attribute mapping was presented in this paper： firstly, the samples were partitioned into several groups by their properties; secondly, the typical relation between soil attribute and key environmental factors in each class was derived through regression model, and then clustering method was optimized according to the residual information above, finally the non-sampled area was predicted by a weighted fitting of all the typical relations. A case study of soil organic matter mapping was taken in order to exam the performance of the approach. The result showed that the method was of wide suitability and could predict the soil attribute information with a high accuracy by choosing and fitting the key factors automatically.