中文摘要：

定义了土壤变异中的块段效应,提出了考虑块段效应的RBF神经网络空间插值模型（RBFBE法）,以期提高土壤水分和养分的插值精度;并以江苏省扬州市区北部某试区为例,通过A、B两种训练方案,将RBFBE法对土壤含水率与有机质含量的插值结果与常规的RBF神经网络（RBFANN法）空间插值结果及普通克里金法（OK法）插值结果进行了对比分析。结果表明：与OK法相比,RBFBE法能使土壤特性的均方误差Mse缩小19.0%～62.2%,预测吻合度G提高8.9%～28.8%;与输入信息相同的RBFANN法相比,RBFBE法亦能使土壤特性的均方误差Mse缩小10.0%～48.8%,预测吻合度G提高3.4%～22.0%;此外,研究也表明RBFBE法具有较强的泛化能力。考虑块段效应的RBF神经网络方法有利于拓展人工神经网络方法在土壤特性空间插值中的应用范围,具有一定的应用前景。

英文摘要：

Precision farming and decision-making in environmental protection will require a better understanding of the spatial distributions of soil moisture and nutrients.In this study,the block effect due to soil variability is defined and a Radial Basis Function（RBF） artificial neural network（ANN） with Block Effect（BE） is proposed to improve the accuracy of spatial interpolation of soil moisture and nutrients.The new RBFANNBE method is evaluated using soil moistures and organic matters observed from an experimental site in the north Yangzhou region of Jiangsu Province.Two sets of data samples are used in the RBFANNBE training process.The RBFANNBE performance is compared to the conventional RBFANN method and the ordinary Kriging（OK） method using the mean square errors（Mse） and the goodness of prediction fit（G）.The results show that RBFANNBE reduce Mse by 19.0%-62.2% and improve G by 8.9%-28.8% compared to the OK method.In comparison with RBFANN,a 10.0%-48.8% reduction in Mse and a 3.4%-22.0% improvement in G are obtained by RBFANNBE,respectively.The new RBFANNBE method has better generalization capabilities over the RBFANN and OK methods.Thus,RBFANNBE is a promising method for studying the spatial distribution of soil properties.