Rapeseed Growth Monitoring Based On Multi-sources Remote Sensing Data

Rape as the widely cultivated oil crop in China,real-time,accurate and rapid monitoring rape growth is of great significance for rape growth diagnosis, management and yield prediction. With the development of remote sensing technology, more and more abundant remote sensing data source makes real-time monitoring crop growth in larger scale possible,large numbers of improved spectral characteristic parameters and vegetation index are applied to agronomic parameters inversion for precision agriculture. Especially hyperspectral remote sensing has high spectral resolution,strong waveband continuity and a great of spectral information, so the hyperspectral remote sensing technology has became a necessary tool of precision agriculture and provided effective information for real-time rapid and large-scale crop growth.Making full use of multi-source remote sensing data to monitor the growth of crucial growing period and established more accurate monitoring model of rape growth becomes a trend of monitoring rape growth in larger scale.This study optimizes the Length-Width function models to estimate leaf area(LA) and fresh weight(FW) under different conditions(such as different sowing ways, fertilization level, etc.), the hyperspectral remote sensing technology has been applied to inverse leaf area index(LAI) of rape experiment plots in different periods,and estimate accurate model of rape yield using leaf area index. By extracting remote sensing parameters inverse leaf area index of rape to estimate rape yield in larger scale.Based on the above research, the dissertation draws the following conclusions:1. The Levenberg-Marquardt and Universal Global Optimization methods are used to optimize the Length-Width function model to estimate leaf area(LA) and fresh weight(FW), The Length-Width Power function models minimally interfered by the external disturbance(such as different sowing ways, fertilization level, growth stage etc.) can be further used to estimate leaf area and fresh weight and are more accurate than the conventional linear models.2. The quadratic polynomial inversion models can be perfect of estimating LAI of rape using hyperspectral techniques,and the spectral red edge parameters can estimate accurately LAI at seedling stage,the predicted models based on NLI index produces better estimation for LAI at late stages of rape.To simplify the prediction process by integrating all the data at different stage, the unified validation of models shows that there is low prediction precision with the unified spectral variables or vegetation indices monitoring LAI, and the significant difference of plant architecture and coverage in rapeseed at different stages,the different spectral response characteristics among flower, pod and leaf resulted in the low prediction precision of validated models,there is high prediction accuracy of monitoring model based on the appropriate spectral variables and vegetation indices to estimate LAI of the winter rapeseed.at different growth stages.3. Taking LAI of rape experiment plots in different periods as independent variables and the final yield of rape experiment plots as dependent variable, the final yield of the experimental plot has a significant positive correlation with LAI in each period. Especially the LAI of rape at pod stage showes a highly significant positive correlation with the final yield. The yield prediction model based on LAI of rape at ten-leaf stage、flowers-stage and pod stage via a stepwise regression analysis has a high precision,after testing, the prediction model has a better predictive effect.4. The extracted nine commonly used vegetation index from the remote sensing images for retrieval of LAI of rape at ten-leaf stage、flowers-stage and pod stage in Wuxue city via a stepwise regression analysis, and estimating the rape yield based on the yield prediction model of rape experiment plots.The results show that the parameters of remote sensing images can be used to estimate the leaf area index and rape yield in larger scale.