一种用于山区土壤重金属评估的集成RBF空间插值算法(英文)

Spatial interpolation method based on integrated RBF neural networks for estimating heavy metals in soil of a mountain region

提出了一种在山区能够准确、稳定地预测未采样点土壤重金属浓度的集成径向基函数神经网络空间插值方法(IRBFANNs).该方法集成径向基函数神经网络和神经网络集成技术的优点.为了研究所提IRBFANNs方法的性能,进行了3组不同采样密度条件下的实验.通过M n元素插值的均方根误差和分布估计图对IRBFANNs和其他6个插值方法进行了比较.实验结果表明:IRBFANNs方法在精确性和稳定性方面优于其他参评方法,且在采样密度稀疏条件下该方法能够提供细节较丰富的分布估计图.

A novel spatial interpolation method based on integrated radial basis function artificial neural networks （IRBFANNs） is proposed to provide accurate and stable predictions of heavy metals concentrations in soil at un- sampled sites in a mountain region. The IRBFANNs hybridize the advantages of the artificial neural networks and the neural networks integration approach. Three experimental projects under different sampling densities are carried out to study the performance of the proposed IRBFANNs-based interpolation method. This novel method is compared with six peer spatial interpolation methods based on the root mean square error and visual evaluation of the distribution maps of Mn elements. The experimental results show that the proposed method performs better in accuracy and stability. Moreover, the proposed method can provide more details in the spatial distribution maps than the compared interpolation methods in the cases of sparse sampling density.