JOURNAL OF NATURAL RESOURCES ›› 2011, Vol. 26 ›› Issue (5): 881-890.doi: 10.11849/zrzyxb.2011.05.015

• Resources Research Methods • Previous Articles     Next Articles

Estimating Soil Total Nitrogen Content Based on Hyperspectral Analysis Technology

ZHANG Juan-juan1,2, TIAN Yong-chao1, YAO Xia1, CAO Wei-xing1, MA Xin-ming2, ZHU Yan1   

  1. 1. National Engineering and Technology Center for Information Agriculture, Jiangsu Key Laboratory for Information Agriculture, Nanjing Agricultural University, Nanjing 210095, China;
    2. College of Information and Management, Henan Agricultural University, Zhengzhou 450002, China
  • Received:2010-09-14 Revised:2010-12-24 Online:2011-05-20 Published:2011-05-20

Abstract: Quantitative relationships between soil total nitrogen content (TN) and hyperspectra in visible and near-infrared region (VIS-NIR) (350-2500 nm) were studied for five soil types (paddy soil, fluvo-aquic soil, salinized fluvo-aquic soil, saline soil, dark soil with lime concretion) collected from central and East China. Based on three different methods of spectral index, partial least square (PLS) and back propagation neural network (BPNN), the models were developed for estimating TN content in soil. The results showed that the newly developed PLS and BPNN models for estimating TN content based on the corrected first derivative spectra of 500-900 nm and 1350-1490 nm regions with Norris smoothing filter performed well, with R2 of calibration as 0.81 and 0.98, respectively. The R2, RMSE and RPD of validation were 0.81, 0.219 g·kg-1 and 2.28 for the method of PLS, and were 0.93, 0.149 g·kg-1 and 3.36 for the method of BPNN, respectively. In addition, DI (NDR872, NDR1482) composed of the corrected first derivative spectra of 872 nm and 1482 nm with Norris smoothing algorithm also had a good correlation with soil TN content. Testing of the estimating model based on DI(NDR872, NDR1482) with independent datasets from different types of soil samples resulted in R2, RMSE and RPD as 0.66, 0.53 g·kg-1 and 1.60, respectively. Comparison of the above three methods, the sequence of prediction accuracy was PLS-BPNN model>PLS>DI(NDR872, NDR1482), which indicated that the newly developed BPNN and PLS models were reliable for estimating soil TN content with high prediction accuracy, and DI(NDR872, NDR1482) maybe a good indicator of soil TN content.

Key words: soil, total nitrogen, hyperspectra, partial least square, BP neural network, spectral index

CLC Number: 

  • S153.6