基于无人机多光谱遥感数据的水稻叶面积指数反演模型研究OA

Taking three indica conventional rice varieties as the research objects,a trial was carried out in Yueyang City,Hunan Province.The multispectral remote sensing images and the data of leaf area index(LAI)at six rice growth stages(14,21,35,42,56,and 73 days after transplanting)were obtained by UAV and manual measurement.Different methods were used to build the intertemporal and non-intertemporal prediction models for rice LAI inversion,and the model prediction performance(fitting effect and prediction accuracy)was evaluated.Through Pearson correlation analysis,five vegetation indices closely related to rice LAI were selected,which were normalized difference red edge index(NDRE),green normalized difference vegetation index(GNDVI),normalized difference vegetation index(NDVI),leaf chlorophyll index(LCI),and optimized soil-adjusted vegetation index(OSAVI).Two multivariate regression methods,stepwise regression and Lasso regression,and two machine learning regression methods,random forest(RF)and support vector machine(SVM),were used to construct intertemporal and non-intertemporal prediction models for rice LAI inversion,respectively.The prediction performance of each model was evaluated based on coefficient of determination(R2)and root mean square error(RMSE).The results showed that among the intertemporal prediction models for rice LAI inversion,the SVM model had the better overall prediction performance,and it had the highest prediction accuracy(R2 of 0.562)on the 56th day after transplanting and the lowest prediction accuracy(R2 of 0.095)on the 14th day on the validation set.Among the non-intertemporal prediction models for rice LAI inversion,the RF model had the better overall prediction performance,with R2 of and above 0.562(up to 0.856)at various growth stages on the validation set,demonstrating stable prediction performance.Overall,the machine learning models had higher prediction performance than the multivariate regression models.Although the intertemporal prediction models for rice LAI inversion showed a certain migration ability,they varied greatly at different growth stages,and it was still necessary to optimize the modeling strategy according to the characteristics of different stages.