Inversion Of Phosphorus Nutrition Of Rubber Seeding Leaves With Hyperspectal Information

Phosphorus is a vital nutrient for the growth and development of rubber trees and is closely related to many important metabolic activities in rubber trees and rubber synthesis. It is of great significance both in theory and practice to make a rapid and accurate diagnosis with the rubber tree phosphorus nutrition status. This paper provides technical support for the widely application of precise fertilization technique in rubber plantations through rubber leaf phosphorus nutrition quick and non-damaging test technique established by hyperspectral techniques.In order to establish prediction model of phosphorus of rubber tender leaves, this study cultivates rubber seedlings under different levels of phosphorus treatments with sand firstly, collects spectral data of different treatments by using FieldSpec3spectrometer, then, makes wavelet analysis to extract spectral reflectance data after denoising, obtains the first-order and second-order derivative spectrum of spectral reflectance with differential technology, and lastly, takes those data as input variables to establish the prediction model by adopting partial least squares regression and neural net methods, the specific results are as follows:(1) The phosphorus contents in rubber seedlings had significant difference under different phosphorus treatments. As the treatment content increases, phosphorus in rubber seedling leaves increased gradually, and the phosphorus contents between phosphorus treatment and phosphorus-deficiency treatment were significantly different (P<0.05).(2) Wavelet analysis can remove the noise existing in original spectral reflectance curves well. Decomposing and reconstructing the original spectral reflectance curve by using wavelet analysis show that the reconstruction of spectral curve in the5th layer of signal can remove the noise in the original spectrum and meanwhile remained most of its information, whose correlation with the original spectral reflectance reaches0.9997.(3) The spectral reflectance curves of rubber seedlings had significant difference under different phosphorus treatments. The reflectivity with phosphorus treatment was higher than phosphorus-deficiency treatment approaching to550nm and between1500nm to1850nm.(4) There were significant correlations between the phosphorus content of rubber seedling leaves and spectral reflectance with de-noising as well as the first-order and second-order derivative spectrum of spectral reflectance. Phosphorus content in rubber seedling leaves had significant and even extremely significant correlations with many spectral reflectances after de-noising, however, their correlation coefficients were smaller, less than0.6; after de-noising, the correlations between the first-order and second-order derivative spectrum of spectral reflectance and phosphorus content of rubber seedling leaves were significantly stronger, whose coefficients were more than0.7.(5) The model established by partial least-squares regression and neural network methods can predict the phosphorus content of rubber seedling leaves. In partial least-squares regression models, the model that took the characterized bands of second-order derivative spectrum of spectral reflectance with de-noising as input variable present the highest prediction accuracy, in calibration set and validation set, the correlation coefficient between measured values and predicted values of phosphorus content of rubber seedling leaves were0.9317and0.8849respectively. While, in neural network models, the one that took the characterized bands of first-order derivative spectrum of spectral reflectance with de-noising as input variable showed the highest prediction accuracy, in calibration set and validation set, the correlation coefficient between measured values and predicted values of phosphorus content of rubber seedling leaves were0.9127and0.911respectively.In conclusion, in the study of prediction of phosphorus content of rubber seedling leaves, the highest prediction accuracy appears at the neural network model that takes the characterized bands of first-order derivative spectrum of spectral reflectance with de-noising as input variable. This model can be applied in the quick and non-damaging test of phosphorus content of rubber seedling leaves, thus grasping the phosphorus nutrition status of rubber seedlings timely and effectively, which greatly improves the effectiveness of the phosphorus nutrition diagnosis of rubber seedlings phosphorus nutrition diagnosis.