Prediction and mapping of tobacco yield with fresh leaf mass using hyperspectral sensing data

Tobacco yield is important for agricultural management and policy making. Hyperspectral remote sensing has advantages in predicting and mapping crop yield, providing both yield values and the spatial distribution of yield. Proximal hyperspectral sensing data and unmanned aerial vehicle (UAV) hyperspectral remote sensing images were used to predict and map tobacco yield. To improve the application of hyperspectral data in crop yield prediction, the potential of informative spectral regions for prediction of tobacco yield was explored. The study was conducted at two growth stages in the growth season in 2021 in Yunnan Province, southwestern China. Genetic algorithm and partial least squares regression were used for model calibration. In prediction of tobacco yield using proximal hyperspectral data, compared with the prediction using the full spectral range of raw reflectance spectra, the prediction was improved by using continuum removed spectra and informative spectral subsets. Mean relative error (MRE) of the prediction was improved from 16.91 % to 12.53 % by using the full spectral range of continuum removed spectra at the growth stage in early July and from 16.92 % to 15.38 % by using the green and red edge spectral subset of continuum removed spectra at the growth stage in late July. The prediction model developed using proximal hyperspectral data was applied to UAV hyperspectral remote sensing image to map tobacco yield. Root mean square error (RMSE) and MRE values of the generated tobacco yield map were 446.27 kg ha -1 and 16.68 %, achieving tobacco yield forecasting one month before harvest. The results show informative spectral regions are potential for tobacco yield prediction and UAV hyperspectral remote sensing image can map the spatial distribution of tobacco yield with accuracy. The study provides an alternative to predict and map tobacco yield using hyperspectral remote sensing imagery.