Estimation model of LAI and nitrogen content in tea tree based on hyperspectral image

Abstract: Leaf area index (LAI), the total area of plant leaves on the unit land area, is an important vegetation characteristic of plant canopy, which can reflect the growth status of vegetation. The effective detection of leaf nitrogen, a significant chemical element which could promote the growth of plant leaves, is beneficial to the precision fertilization and nutrient management of tea plantations, while it is also of great importance to the improve of quality and yield of tea leaves. In order to improve the production of Yinhong 9th tea, rational fertilization and protect the tea garden ecological environment, in this study, we used hyperspectral nondestructive testing technology to detect the LAI and nitrogen content with hyperspectral camera. Nitrogen is an important element that makes up tea chlorophyll, its content directly affects the synthesis of organic matter in tea tree, and it can affect leaf area index. Therefore, leaf area index and nitrogen content have certain correlation. At present, information on research of plant leaf area index for hyperspectral detection is limited. By using hyperspectral inversion tea nondestructive detection of two parameters, it can solve the problem of tea plant nutrition diagnosis, which may has positive impact on quality of the Yinhong 9th tea. Although the detection of nitrogen content of leaf is direct and accurate using the traditional chemical method, its complex operation, sample damaging and incapable to detect large area orchard in real-time, fast and nondestructive make it not the best method. To achieve real-time, fast and nondestructive detection of leaf area index (LAI) and nitrogen content of tea leaves, in this paper, a portable hyperspectral imager had been used to gather spectral data. Destructive leaf picking had been used to calculate leaf area index and traditional chemical method had been used to calculate the leaf nitrogen content. Using the Yinghong 9th tea tree as test subject, correlation analysis was done among hyperspectral characteristic variable data, LAI and leaf nitrogen content. Estimated model was built using high relevant hyperspectral characteristic parameter data, LAI and nitrogen content by linear, index, logarithm, parabola and etc. After that, evaluation of model performance and model accuracy test by root mean square error (RMSE) were conducted. After comparing different transformation of spectral data between the correlation of LAI and nitrogen content, the pretreated spectral parameters were used as independent variables to build the regression model of LAI and nitrogen content, respectively. The results showed that the logarithmic optimal fitting model was built between Green peak reflectance Rg and leaf area index, from which the regression coefficient value R2 and test samples RMSE value was 0.9 and 0.087 6, respectively, while the best modeling result was the model built between vegetation index VI4 and the nitrogen content, from which the regression coefficient value R2 and test samples RMSE value was 0.830 3 and 0.102 9, respectively. According to the result above, a better estimated model was developed using LAI detected by traditional chemical method, nitrogen content and hyperspectral data, which produced the theoretical basis for fast and nondestructive detection of LAI and nitrogen content of tea leaves in the possible future. However, sample process and modeling comparison of different growing season had not been done. The further research should be continued to study the relationship between LAI and the nutritional status, growth index and nutrient management of tea tree in different cultivation measures and ecological conditions. These results can provide reference for the nondestructive detection of LAI and the nutrient component of tea tree.