Investigating impacts of calibration methodology and irradiance variations on lightweight drone-based sensor derived surface reflectance products

Abstract

The miniaturisation of multispectral sensors in recent years have resulted in a proliferation of applications particularly in vegetation-focused studies using lightweight drones. Multi-camera arrays (MCAs), capable of capturing information over different wavelength intervals using separate cameras with specific band-pass filters, are now commonplace in this field. However, data from MCAs require a considerable amount of geometric and radiometric corrections if high quality reflectance products are to be delivered. Some aspects of this workflow can be handled by commercial software packages (e.g. Pix4D and Agisoft Metashape), using black box algorithms, however radiometric uncertainties within products are not reported to the end-user by the software. We present the results of two experiments using a low-cost MCA complete with irradiance sensor (Parrot Sequoia), which set out to assess the accuracy and consistency of hemispherical-conical surface reflectance factors from MCA data. Using reference panels in the field, we found that the empirical line method (ELM) generated the smallest RMSEs (0.0037) when compared to simplified single-panel based workflows; while for the latter there was little difference between using a calibrated Spectralon® panel or grey card imaged prior to the flight (0.0215 vs 0.0154 average over the four bands). Errors for a vegetated target within the survey flight were larger and comparable for all cases. Furthermore, a study on median vegetation index values for single vegetation canopies showed that illumination correction using irradiance data still yields significant differences in resulting values between two acquisitions during changing direct and diffuse irradiance conditions. We therefore highlight the importance of critical assessment prior to integrating drone derived MCA-measured reflectance factors into further geospatial workflows.