土壤有机质,一阶导数光谱,Norris平滑滤波,差值光谱指数DI(D554,D1398),BP神经网络
," /> 土壤有机质,一阶导数光谱,Norris平滑滤波,差值光谱指数DI(D554,D1398),BP神经网络
,"/> soil organic matter,derivative spectra,Norris smoothing filter,difference spectral index DI(D554, D1398),BP neural network
,"/> <font face="Verdana">Spectral Characteristics and Estimation of Organic Matter Contents of Different Soil Types#br# </font>

Scientia Agricultura Sinica ›› 2009, Vol. 42 ›› Issue (9): 3154-3163 .doi: 10.3864/j.issn.0578-1752.2009.09.017

• SOIL & FERTILIZER·WATER-SAVING IRRIGATION·AGROECOLOGY & ENVIRONMENT • Previous Articles     Next Articles

Spectral Characteristics and Estimation of Organic Matter Contents of Different Soil Types#br#

ZHANG Juan-juan, TIAN Yong-chao, ZHU Yan, YAO Xia, CAO Wei-xing#br#   

  1. (南京农业大学农学院/江苏省信息农业高技术研究重点实验室)
  • Received:2008-08-18 Revised:2008-10-08 Online:2009-09-10 Published:2009-09-10
  • Contact: CAO Wei-xing

Abstract:

【Objective】 The objectives of the present study were to determine the key spectral parameters and models for estimating SOM content. 【Method】 The dried sample of five different soil types in China were analyzed for SOM content and hyperspectral reflectance within 350-2 500 nm, quantitative models of SOM using spectral index and BP neural network were established, respectively. 【Result】The results showed that correlation between spectral indices which composed of first derivative and SOM content were obviously stronger than those composed of original reflectance, especially derivative with Norris smoothing filter. The correlation sequence of SOM to different index types was DI>RI>ND which composed of spectral reflectance or the first derivative spectra. DI composed of first derivative of 554 nm and 1 398 nm gave a better prediction performance, with equation as y=184.2×exp[-1297×DI(D554, D1398)], coefficient of determination was 0.90. Testing of the monitoring models with independent data from different soil types indicated that R2, RMSE and RPD of validation were 0.84, 3.64 and 2.98, respectively. In addition, the scores computed by PLS were applied as input of BP neural network developed with over 99.56% of cumulative proportion of correlation matrix. R2 of calibration model was 0.98, and R2, RMSE, RPD of validation were 0.96, 2.24 and 4.83, respectively. Compared with BP neural network model, DI(D554, D1398) had a little lower prediction precision, but it could meet need of estimating of SOM content. 【Conclusion】 It is concluded that both of methods based on DI(D554, D1398) and BP neural network can estimate SOM content accurately.

Key words: soil organic matter')">soil organic matter, derivative spectra, Norris smoothing filter, difference spectral index DI(D554, D1398), BP neural network

No related articles found!
Viewed
Full text


Abstract

Cited

  Shared   
  Discussed   
No Suggested Reading articles found!