论文部分内容阅读
基于全国第二次土壤普查得到的6 241个典型土壤剖面数据,采用主成分分析方法和径向基函数神经网络模型建立不同植被类型—土纲单元内土壤有机质与气候、地形和植被等环境因子间的非线性关系,模拟全国表层土壤有机质的空间分析格局。结果表明,该模型具有较准确的预测能力,性能指数达到1.94。与普通克里格法、反比距离法和多元回归模型相比,神经网络模型对621个验证点模拟结果与实测值的相关系数为0.799,分别提高了0.265、0.181和0.120,平均绝对误差分别降低了4.25、4.43和2.34 g/kg,平均相对误差分别降低了30.16%、32.66%和5.93%,均方根误差则分别降低了8.61、8.24和6.24 g/kg;从模拟结果图来看,神经网络模型能够提供更多的细节信息。该方法为大尺度土壤性质空间分布模拟提供了有益的参考。
Based on the data of 6 241 typical soil profiles obtained from the Second National Soil Survey, principal component analysis and radial basis function neural network model were used to establish the relationship between soil organic matter and climate factors such as climate, topography and vegetation in different vegetation types Between the nonlinear relationship between the simulation of the surface soil organic matter spatial analysis of the pattern. The results show that the model has a more accurate prediction ability, the performance index reached 1.94. Compared with ordinary kriging method, inverse distance method and multivariate regression model, the correlation coefficient between the simulated result and the measured value of 621 verification points in the neural network model is 0.799, increased by 0.265, 0.181 and 0.120 respectively, and the average absolute error decreased 4.25, 4.43 and 2.34 g / kg, respectively. The average relative errors were reduced by 30.16%, 32.66% and 5.93%, respectively, and the root mean square errors were decreased by 8.61, 8.24 and 6.24 g / kg, respectively. The network model can provide more details. This method provides a useful reference for the simulation of the spatial distribution of large-scale soil properties.