Remote Estimation of Rice Yield With Unmanned Aerial Vehicle (UAV) Data and Spectral Mixture Analysis

The accurate assessment of rice yield is crucially important for China’s food security and sustainable development. Remote sensing (RS), as an emerging technology, is expected to be useful for rice yield estimation especially at regional scales. With the development of unmanned aerial vehicles (UAVs), a novel approach for remote sensing has been provided, and it is possible to acquire high spatio-temporal resolution imagery on a regional scale. Previous reports have shown that the predictive ability of vegetation index (VI) decreased under the influence of panicle emergence during the later stages of rice growth. In this study, a new approach which integrated UAV-based VI and abundance information obtained from spectral mixture analysis was established to improve the estimation accuracy of rice yield at heading stage. The six-band image of all studied rice plots was collected by a camera system mounted on an UAV at booting stage and heading stage respectively. And the corresponding ground measured data was also acquired at the same time. The relationship of several widely-used VIs and Rice Yield was tested at these two stages and a relatively weaker correlation between VI and yield was found at heading stage. In order to improve the estimation accuracy of rice yield at heading stage, the plot-level abundance of panicle, leaf and soil, indicating the fraction of different components within the plot, was derived from spectral mixture analysis on the six-band image and in-situ endmember spectra collected for different components. The results showed that VI incorporated with abundance information exhibited a better predictive ability for yield than VI alone. And the product of VI and the difference of leaf abundance and panicle abundance was the most accurate index to reliably estimate yield for rice under different nitrogen treatments at heading stage with the coefficient of determination reaching 0.6 and estimation error below 10%.