Rice biomass estimation based on multispectral imagery from unmanned aerial vehicles

Rice is one of the major food crops in China. It is important to timely, accurately, efficiently, and rapidly monitor the rice biomass. In this study, ten multispectral vegetation indices were extracted from Unmanned Aerial Vehicles (UAV) multispectral images. The data source was obtained for the correlation analysis with rice biomass. The results showed that the correlation between biomass and vegetation index was higher for the full reproductive period than for a single period. The correlation between biomass and vegetation index was also improved to divide the data for the full reproductive period into a nutritive and a reproductive period, according to the vegetation cover up to 100%. All ten vegetation indices were significantly correlated throughout the reproductive and nutrient growth period, with the highest DVI and RE-RVI correlation of 0.689 and 0.894, respectively; Except the DVI, the vegetation indices were significantly correlated in the productive growth period, with the highest RE-NDVI correlation of -0.794 for the rice spike portion and 0.629 for the stem and leaf portion. The biomass estimation models were then constructed using vegetation indices for the nutritive and reproductive growth periods. Quadratic regression and exponential models were also obtained in the nutritive growth period. The optimal multispectral index was achieved in the RERVI, with the validation accuracies of R²=0.90, and RMSE=119.36 g/m²; In the reproductive growth period, the rice spike was a quadratic regression model with an optimal multispectral index of RVI and validation accuracies of R² = 0.78 and RMSE = 124.98 g/m², respectively. Overall, the accuracy of the model was improved in the biomass estimation with the full-birth period data using the vegetation cover close to 100%. The accuracy of biomass estimation was also improved to separate the rice spike from the stalk and leaf during the reproductive growth period. Relatively less saturation was suffered in the early growth period of rice, whereas, the accuracy was slightly higher than that in the late growth period; Moreover, the experimental data significantly improved the estimation precision and accuracy of the model. The rice biomass was segmented for modeling and accuracy validation during the whole life cycle, in terms of plant nutritional and reproductive growth. The finding can also provide effective information and references for field crop growth monitoring and decision-making on farmland production.