Abstract:
Abstract: The phosphorus status is an important parameter for evaluating the growth status and predicting the production in apple trees. The objective of the study was to demonstrate the feasibility of remote sensing monitoring the apple leaf total phosphorus content and its expansibility in regional and annual level. Spectral reflectance of leaves and concurrent apple leaf phosphorus content parameters of samples were acquired in Xiazhai Village, Chaoquan District, Feicheng City, Shandong Province, China during the apple growth season from 2012 to 2013, and the optimal weight combination model was built using the Radial Basis Function (RBF) neural network. Leaf spectra and total phosphorus content of apples were measured at the fast-growing period of shoot, the time of blooming of vernal treetop, the fruit expansion period, the fruit maturity stage, and the color changing stage in the leaves. The paper was based on the apple whole stage. The leaf reflectance was measured and then the normalized difference spectral index (NDSI) and ratio spectral index (RSI) that were sensitive to total phosphorus content were built; the optimal weight combination model of the RBF was discussed and the hyperspectral estimation model for total phosphorus content in apple leaves was established. Firstly, We analyzed the correlation between the phosphorus content and the original spectrum, determined R553 and R722 as the diagnostic band of leaf phosphorus content, and constructed the estimation model of total phosphorus content. The coefficient of determination (R2), root mean square error (RMSE) and relative error (RE) were 0.69, 0.07 g/(100g), 0.2% and 0.80, 0.06 g/(100g), 0.2%, respectively; the NDSI and RSI was constructed referred to normalized difference vegetation index (NDVI) and ratio vegetation index (RVI). The sensitivity of hyperspectral vegetation indices NDSI(546, 521), NDSI(553, 518), RSI(543, 525) and RSI(1394, 718) to phosphorus content was determined by the contour map of combination range (400-2500 nm) with the leaf total phosphorus content. The estimated model was built based on the empirical statistical relationships between NDSI(546, 521), NDSI(553, 518), RSI(543, 525), RSI(1394, 718) and total phosphorus content, and the corresponding R2, RMSE and RE were 0.87, 0.05 g/(100g) and 0.3%, 0.86, 0.05 g/(100g) and 0.05%, 0.87, 0.05 g/(100g) and 0.2%, and 0.85, 0.05 g/(100g) and 0.2%, respectively. Lastly, the optimal weight combination model of RBF neural network was constructed; the goal and spread were calculated by iteration until the min (et) was minimal. The R553, R722, NDSI(546, 521), NDSI(553, 518), RSI(543, 525) and RSI(1394, 718) were considered as independent variables and the total phosphorus content was taken as dependent variable in the combination model. Gaussian function was used as radial basis function, which could get the optimal weight for every independent variable. The results indicated that the prediction of the optimal weight combination model of RBF neural network had a higher precision, compared to the mean of the 6 estimated models (traditional empirical statistical models), the R2 was increased from 0.82 to 0.94, and the RMSE was decreased from 0.06 to 0.05 g/(100g). The validation results also indicated that the estimation accuracy of the optimal weight combination model (R2=0.55 and RMSE=0.05 g/(100 g)) was higher than the empirical statistical relations (R2=0.38 and RMSE= 0.06 g/(100g)). The optimal weight combination model of RBF is a new technical method which can provide a rapid and nondestructive diagnosis of the phosphorus status of apple leaves.