山区土壤重金属空间插值性能的改进

The Research on the Spatial Interpolation of Heavy Metals in Soil by Using an Improved Neural Networks

山区土壤重金属监测中,密集均匀的布点采样困难大,成本高。根据稀疏非均匀样本数据准确预测山区土壤重金属空间分布,对可视化描述检测元素的分布及趋势具有重要意义。基于云南省楚雄市南部山区表层土壤中重金属锰、钒的采样数据,对集成径向基函数神经网络(IRBFANNs)和传统插值方法:反距离(IDW)、普通克里格(OK)、径向基函数神经网络(RBFANNs),进行了不同等级采样密度下的插值性能比较。结果表明,样本点数量减少时,传统预测方法的插值精度都有所下降,而IRBFANNs算法在样本点较少情况下能够保持最好的插值精确度和稳定性,有效改进了空间插值性能。

Dense and regular sampling is usually impractical and expensive for soil heavy metal detection in the mountain region.To improve the quality of visual description for the distribution and the trend of the investigated elements and increase the accuracy of prediction for heavy metals distribution based on sparse sampling data,a spatial interpolation methods based on the integrated radial basis function neural networks（ IRBFANNs） is compared with traditional methods including inverse distance（ IDW）,ordinary Kriging（ OK） and radial basis function neural network（ RBFANNs） are carried out under different level sampling density based on the sampling data of soil heavy metal Mn and V in a mountain region of Chuxiong city. The results show that the interpolation accuracy decreases as the number of sample points decreases,however,the integration of radial basis function（ RBF） neural network algorithm has the ability to keep the accuracy and the stability of prediction under sparse sampling density condition,and provides the improved spatial interpolation performance.