Estimating Model for Leaf Area Index of Tobacco via Hyperspectral Reflectance at Different Planting Densities

Leaf area index (LAI) is a key parameter for evaluating tobacco growth status and forecasting its yield and quality. Hyperspectral remote sensing can rapidly and nondestructively acquire LAI. By integrating traditional field monitoring and hyperspectral remote sensing, the primary objective of this study was to explore the best spectral indices and monitoring model for tobacco LAI. On the basis of different planting densities, this study extracted and analyzed 10 spectral parameters. The quadratic polynomial model, logarithmic model, stepwise multiple linear regression (SMLR) and BP neural network model were used to construct the prediction models for tobacco LAI. The results showed that the correlation between the tobacco LAI and NDVI, RVI, MCARI, GM1, GNDVI2 and PSSRb all reached extremely significant correlation (p<0.01), and the correlation coefficients were all higher than 0.80. The tobacco LAI prediction models of quadratic polynomial model, logarithmic model, SMLR and BP neural network model had the R² value of 0.69, 0.57, 0.89 and 0.90, respectively. The validation RMSE of the four models was 0.69, 0.87, 0.62 and 0.44, respectively. Both SMLR and BP neural network models achieved good results, and the BP neural network model is the best model for inversion the tobacco LAI with the biggest accuracy and the minimum error. These results provide technical support and regional reference for accurate monitoring tobacco LAI under different planting densities.