Study of radiometric variations in Unmanned Aerial Vehicle remote sensing imagery for vegetation mapping

Abstract: Unmanned Aerial Vehicles (UAVs) provide a flexible method for acquiring high-resolution imagery with relative simple operation and cost-effectiveness. This technology emerged 30 years ago and it is widely used by commercial, scientific, and military communities due to its versatility. However, new technology brings new challenges. One of them is the radiometric accuracy of the UAV imagery. Radiometric accuracy is especially important when working with different illumination conditions, dates or sensors. The present study focuses on reducing radiometric errors of UAV images for vegetation mapping. The fieldwork took place from September 2016 to May 2017 in an agricultural area and a mire. The effect in incident light variations was studied flying in different dates and at different times of the day. Sun elevation angle and presence of clouds gave significant variations in UAV imagery. The study of the sun elevation angle showed that suitable hours for UAV surveys at higher latitudes surveys is within 2 hours of solar noon, since the amount of shadows is low because the sun elevation angle is between 20° and 40°. The difference in the type of radiation affected the homogeneity of the UAV imagery and the radiometric correction, making the correction of UAV imagery from days with clear sky more difficult, when the direct radiation is predominant. The BRDF effects were less pronounced under overcast conditions, when the predominant incident radiation is diffuse. Nine correction methods were tested, and their effect on different vegetation indices was compared, showing that the irradiance correction method prior to an empirical line calibration provide less errors than other methods. However, the errors are still high when compared with ground spectral samples (lowest RMSE 38% under overcast conditions for the NDVI). A simple workflow was developed for vegetation mapping purposes for the Micasense Rededge camera. We suggest to use the automatic dark current-corrected and automatic reduced vignetting effect images, plus irradiance compensation and the use of empirical line calibration to obtain reflectance values in single images before generating the orthomosaic. The radiometric correction process should be done for each spectral band, having a new calibration equation per mission due to the change in sky conditions. Unfortunately, this workflow will not provide accurate results in the calculation of vegetation indices that assess small variations like the case of chlorophyll indices or vegetation indices that combine several bands. Further research is needed to improve the accuracy of the correction.