랜덤포레스트 분류 모델 기반 벼 잎 분광반사율을 이용한 질소 시비량 추정 연구 = Estimating Nitrogen Fertilization Levels Based on the Spectral Reflectance of Rice Leaves Using a Random Forest Classification Model

다국어 초록 (Multilingual Abstract)

ABSTRACT Although nitrogen is essential for chlorophyll production and crop growth, the excessive use of nitrogen-based fertilizers is a source of environmental concern. In this study, we sought to assess the optimal levels of nitrogen fertilization for the cultivation of rice based on based on measurements of leaf spectral reflectance and the use of a random forest classification model.
On the basis of the standard fertilization rate, we applied four levels of nitrogen fertilization (0%, 50%, 100%, and 200%) and measured plant height, tiller number, SPAD values, chlorophyll concentrations, and spectral reflectance of rice leaves at different stages of growth. With an increase in rate of nitrogen fertilization, we recorded increases in plant height, the number of tillers, SPAD values, and leaf chlorophyll concentrations. In contrast, we recorded reductions in the spectral reflectance of leaves in the 550 and 710 nm wavelength regions in response to an increase nitrogen fertilization. On the basis of these findings, wavelengths highly correlated with nitrogen fertilization levels were extracted using a PSL model and used as input variables for constructing a random forest classification model. The performance of the model was found to vary throughout the growth period, with a classification accuracy ranging from 76.2% to 91.2%. We speculate that this variability in classification accuracy can be attributed to the timing of nitrogen fertilization, which determines nitrogen availability, and contributes to changes in chlorophyll levels during the heading and mature stages of growth. Thus, nitrogen fertilization levels can effectively be assessed using spectral reflectance in the late growth stages, during which the changes in chlorophyll concentrations are more pronounced. In the future, we will investigate whether obtaining sufficient spectral reflectance data for different treatment at specific growth stages of growth can contribute to enhancing the accuracy of the model developed herein, thereby enabling a more precise classification of nitrogen fertilization levels at the different stages of rice growth.