Spatial variation and predictions of soil organic matter and total nitrogen based on VNIR reflectance in a basin of Chinese Loess Plateau

• Autores: Hongfen Zhu, Zhanjunm Xu, Yaodong Jing, Rutian Bi, Wude Yang
• Localización: Journal of soil science and plant nutrition, ISSN-e 0718-9516, ISSN 0718-9508, Vol. 18, Nº. 4, 2018, págs. 1126-1141
• Idioma: inglés
• Enlaces
  • Texto completo
• Resumen

  • Taiyuan basin in Chinese Loess Plateau was characterized as the variety of landform, land use fragmentation, low soil organic matter (SOM) and soil total nitrogen (STN) content; therefore, the predictions of soil nutrients in the area were rather difficult. In this study, three soil sampling transects of cropland soil from northwest to southeast in Taiyuan basin were established and the visible-near infrared reflectance (VNIR) of soil samples were measured. The predicting models for SOM and STN based on VNIR were established, and the predicting accuracies were assessed by traditional evaluating index, wavelet transform, and semivariance structure. The traditional evaluating index showed that the partial least square regression (PLSR) and optimum number of latent variables were suitable for SOM prediction. The accuracies were “good” (RPD ranges from 2.30 to 2.40) for calibration and “moderate” (RPD ranges from 1.80 to 1.95) for validation, whereas the model and parameters of STN were “moderate” (RPD ranges from 1.83 to 1.87) for calibration and “acceptable” (RPD ranges from 1.41 to 1.48) for validation procedure. Based on the wavelet transform, the patterns of global wavelet power spectrum for predicted and measured SOM were closer than that of STN, and their difference in local wavelet spectra could present the predicting errors in the scale and location domain. The nugget effect indicated that the stochastic variability weakened, and the spatial structure of predicted SOM and STN enhanced. The range of predicted SOM and STN were greater than those of measured. Therefore, the predicting models based on independent dataset using PLSR could be used for the prediction of SOM or STN in the un-sampled areas. Wavelet transform and semivariance parameters could be used to guide the utilization of predicted values.