Journal of Ecology and Rural Environment ›› 2021, Vol. 37 ›› Issue (7): 934-942.doi: 10.19741/j.issn.1673-4831.2020.0811

Previous Articles     Next Articles

Accuracy Study of Spatial Predicting in Soil Attributes Based on Interpolations by Artificial Neural Network and Ordinary Kriging

XIE Meng-jiao1,2, WANG Yang2,3, KANG Ying3, WU Zhi-tao4, CHEN Qi-le3, LIU Qi1, WU Chao-yu3, ZHANG Jun-mei2,3   

  1. 1. College of Land and Resources, Hebei Agricultural University, Baoding 071000, China;
    2. State Key Laboratory of North China Crop Improvement and Regulation, Baoding 071000, China;
    3. College of Resources and Environmental Science/Hebei Province Key Laboratory for Farmland Eco-environment, Hebei Agricultural University, Baoding 071000, China;
    4. School of Information and Communication Engineering, Beijing University of Posts and Telecommunications, Beijing 100876, China
  • Received:2020-09-28 Online:2021-07-25 Published:2021-07-23

RichHTML

14

PDF (PC)

164

Abstract: Precise predictions of spatial variability of soil organic matter and total nitrogen are largely depended on methods and spatial scales of the interpolation. In this study, two spatial scales including plot scale (50 m×50 m, 80 soil samples) and farm scale (1 000 m×1 000 m, 100 soil samples) were selected from nearly harvesting field of summer maize in the northern part of Huanghuaihai Plain. And two interpolating methods including ordinary Kriging and artificial neural network based on radial basis function (RBF) were compared to explore more accurate spatial variations of soil organic matter and total nitrogen content in the study area. Results of nugget to sill ratio (0.448-0.637) showed moderate spatial variations of both spatial scales with contents of soil organic matter and total nitrogen ranged from 8.39 to 20.59 g·kg-1 and from 0.31 to 2.90 g·kg-1, respectively. Spatial variations of soil organic matter and total nitrogen presented decreasing trends with higher contents distributed along the diagonal from northwest to southeast and lower contents distributed around the edges at farm scale, in contrast, higher and lower contents respectively located in northeast and southwest parts at plot scale. Concerning root mean square error (RMSE), average absolute error (MAE) and average relative error (MRE) of interpolation by artificial neural network based on RBF, smaller values indicated better predictions of spatial variations at plot scale rather than at farm scale. Without smoothing effect, increased R2 and decreased errors proved better performance of artificial neural network based on RBF rather than ordinary Kriging in modeling spatial variations of soil organic matter and total nitrogen at each spatial scale. It indicates that artificial neural network based on RBF is the better interpolating method for predicting spatial variations of soil organic matter and total nitrogen in this study.

Key words: soil organic carbon, soil total nitrogen, prediction accuracy, ordinary Kriging, RBF neural network

CLC Number: