Variability in vegetation indices as a function of unmanned aerial vehicle flight altitudes and other factors during crop monitoring applications

Abstract

Unmanned aerial vehicles (UAV) integrated with multispectral sensors has been widely used for phenotyping in plant breeding programs and other agricultural applications. However, the extracted vegetation indices can be affected by radiometric correction, altitude of image acquisition, and orthomosaic generation. In this study, we used wheat and pea trials to evaluate the effect of using different reference reflectance panels for the radiometric correction as a function of flight altitude on commonly used vegetation indices. Pea and wheat data were collected at the initial ripening or flowering stages, respectively. In both trials, single multispectral images were collected at 25m, 35m, 45m, 55m, 65m, and 75m flying altitudes. In addition, UAV was used to collect multispectral images at 25m, 4m, and 75m flying altitude and stitched with multiple photogrammetry software (DroneDeploy, ImageBreed, Agisoft, and Pix4D). For radiometric correction, four Lambertian reference panels were used (10%, 18%, 50%, and 99%). The main results indicated that panels with reflectance of 10% and 20% were suitable for radiometric corrections, since the digital number from these panels did not saturate in most cases. In addition, the correction using the 18% reflectance panel demonstrated consistent reflectance response across flight altitudes. Nevertheless, differences in responses were observed as a function of the flight altitudes after orthomosaic generation, especially with DroneDeploy, ImageBreed, and Pix4D. Vegetation index values, such as normalized difference vegetation index, were highly affected by photogrammetry software. Such information is critical to characterize the multispectral data acquired from UAV and other platforms to extract precise and stable data for agricultural applications.