河套灌区土壤质地克里金插值与平滑效应校正

土壤质地的空间分布对指导土壤改良、灌溉排水管理及水土资源合理利用等具有重要意义，而克里金插值方法尽管应用广泛，但往往会产生平滑效应，即高值低估和低值高估。本研究针对河套灌区土壤颗粒组成，使用不同的对数比转换方法处理其成分数据特性，通过普通克里金法对转换后的数据进行空间插值，同时基于邻近点特征构建插值点残差估算的误差反向传播(BP)神经网络(BPNN)模型，并对插值结果的平滑性进行分析评估和校正探究。河套灌区的应用结果表明，不同对数比转换方法对克里金插值的平滑效应作用程度不同，其中等距对数比转换(ILR)的改善效果最显著，与直接使用普通克里金插值相比，黏粒、粉粒和砂粒的极差平滑率分别减小5.8%、33.8%和45.6%,标准差平滑率分别减小38.6%、53.9%和60.2%。进一步使用BPNN模型校正后的土壤颗粒组分插值平滑性显著降低，其中极差平滑率降为0,标准差平滑率仅为0.03～0.07;同时插值误差有一定减小，其中黏粒、粉粒和砂粒的均方根误差分别减小19.8%、21.0%和14.6%。本研究提出的基于对数比转换(ILR)和误差反向传播(BP)神经网络模型联合处理方法(ILR-BP)...

Ordinary Kriging Interpolation and Smoothing Effect Correction for Soil Texture Mapping in Hetao Irrigation District

Spatial distribution of soil texture is important for soil improvement, irrigation and drainage management, and rational utilization of water and land resources, which is usually obtained by spatial interpolation methods like ordinary Kriging (OK). However, OK generally has the smoothing effect that high values are underestimated and low values are overestimated. For soil particle-size fractions and texture interpolation, four logratio transformation methods, including additive (ALR), centered (CLR), isometric (ILR), and symmetry (SLR) logratio transformations, were firstly used to deal with soil particle-size fractions as compositional data. OK was then used to interpolate the transformed data. Finally, a model based on the back propagation neural network (BPNN) was developed to estimate the residuals between the observations and OK estimations to correct the smoothing effect. Results for the Hetao Irrigation District showed that the four logratio transformations had a discrepancy in correcting the smoothing effect. Generally, ILR had the best performance in reducing the smoothing effect. Compared with OK estimations without logratio transformations, the range smoothing rates (CR) of ILR were decreased by 5.8%, 33.8% and 45.6%, and the standard deviation smoothing rates (CS) were decreased by 38.6%, 53.9% and 60.2% for clay, silt, and sand particle-size fractions, respectively. Furthermore, the smoothing effect was obviously weakened and the interpolation errors were decreased after correction by BPNN model, where CR and CS values were reduced to 0 and 0.03~0.07, and the root mean squared errors were reduced by 19.8%, 21.0% and 14.6%, respectively. Therefore, the present correction method can not only weaken the smoothing effect caused by OK but also improve the interpolation accuracy. The spatial interpolation results obtained after the correction showed there were mainly five soil texture types in the Hetao Irrigation District, including sand, loamy sand, sandy loam, loam, and silt loam. The sandy soil mainly distributed in the western part, loam mainly in the central west, northeast, and southwest parts, and silt loam in the north part and south part close to the Yellow River. In conclusion, the OK interpolation results can better reflect the spatial distribution pattern of soil texture after the correction of smoothing effect by using the logratio transformation and BPNN model, which also provided a reference for reasonable utilization of water and land resources in Hetao Irrigation District.