基于范畴型变量和贝叶斯最大熵的土壤有机质空间预测

Prediction of Spatial Distribution of Soil Organic Matter Based On Categorization Variables and BME Methods

选取湖北省沙洋县为研究区域,以土壤质地与土壤有机质定量关系为辅助信息,利用贝叶斯最大熵(BME)方法对沙洋县土壤有机质含量进行空间预测,并与以土壤质地和土壤全氮为辅助变量的协同克里格方法预测结果精度作对照,探讨两种方法的预测效果。结果表明,协同克里格方法和BME方法均能较好反映研究区有机质空间分布特征。在辅助变量与土壤养分存在显著相关性条件下,BME方法能更好地利用范畴型变量等多种类型辅助信息。比较极值误差范围、平均绝对误差、均方根误差等方面,BME方法在土壤属性空间预测方面具有更高精度,且能有效降低数据获取成本和难度,在县域尺度土壤属性空间预测上具有更大优势。

In this paper, spatial dis tribution of soil organic matter(SOM) was predicted based on the Bayesian Maximum Entropy method(BME) with the auxiliary information generated by quantitative relationship between soil texture and SOM. The results were compared with those obtained by Cokriging methods(CK) using soil texture and soil total nitrogen as auxiliary variables. The prediction accuracy of BME and CK was analyzed by mean absolute error(MAE) and root-mean-square error(RMSE). Results indicated that BME could describe the spatial distribution characteristics of SOM of the study area as well as CK. In the condition of significant correlation between auxiliary variables and SOM, BME could effectively narrow the range of errors, reduce the MAE and RMSE, improve the prediction accuracy and reduce the cost and difficulty of data acquisition. Therefore, BME is a better method in the spatial prediction of soil properties in the county scale.