Radiometric Correction of Digital UAS Multispectral Imagery Using Free and Open Satellite Surface Reflectance Images

Unmanned Aircraft Systems (UAS) have gained popularity for numerous remote sensing applications, including disaster damage assessment, precision agriculture, forest monitoring, and surveillance/reconnaissance missions. UASs can be launched with short notice and flown autonomously to follow specific flight trajectories for high-spatial-resolution multispectral or hyperspectral field observations. The collected multispectral imagery can be used to derive indices such as the Normalized Difference Vegetation Index (NDVI), Soil-adjusted Vegetation Index (SAVI), etc., for precision agriculture and crop monitoring. For accurate multispectral imagery, radiometric calibration from image data to physical units (reflectance) is required, which remains one of the major challenges for UAS-based remote sensing. Notable contributions towards radiometric calibration for UAS imagery include: bi-directional reflectance distribution function on remote sensing imagery from small UASs (Stark, Zhao, and Chen (2016)); semi-automatic model was to calculate reflectance for Red, Green, and NIR bands using white Barium Sulfate (BaSo4) panel and a Halon board (Zaman et al. 2014) and simplified empirical line method of radiometric calibration for sUAS with a modified Digital Single-Lens Reflex (DSLR) camera (Wang and Myint 2015). Most of the current radiometric calibration methods use ground target boards (with Lambertian surfaces) and spectroradiometers. Images of the target boards in Digital Numbers (DNs) are calibrated using the reflectance of the corresponding boards measured by the spectroradiometer. However, these methods may not be convenient and economically feasible due to the following reasons: 1) It may be impossible to create a perfectly Lambertian surface which in turn causes inconsistency in reflectance readings with angle and height of observations, 2) High-quality spectroradiometers can cost $100,000 or more and may not be logistically convenient in survey fields, especially in disaster zones.

The goal of this paper is to propose an alternative radiometric correction procedure for digital UAS multispectral imagery using free and open satellite reflectance images. Our proposed method will be less expensive and more convenient than many existing methods which use low-quality ground calibration boards. Data from KHawk UAS and Landsat 8 (L8) Operational Land Imager (OLI) acquired on 7 June 2017 are used to demonstrate the proposed method. Images of KU Field Station fields from the L8 OLI and the KHawk UAS are compared to obtain a mapping function between the two. L8 OLI images are in units of Surface Reflectance (SR) and KHawk UAS images are uncalibrated in Digital Numbers (DNs). First, we downsample the high-resolution UAS images (0.1 m/pix) to match the L8 OLI image resolution (30 m/pix) followed by a pixel by pixel comparison. Through our analysis, we obtained an exponential function that relates the 8-bit DN images to SR images. This corroborates the findings of Wang and Myint (2015) and Logie and Coburn (2018) who also observed non-linearity between DNs and reflectance for commercial of the shelf (COTS) cameras. Fig.1 shows calibrated NDVI images from the KHawk UAS at 0.1 m/pix ground sampling distance (GSD) and L8 OLI at 30 m/pix GSD.

Next, we plan to compare our satellite corrected UAS reflectance images with traditional approaches using ground calibration board and spectroradiometers. The purpose of this is to validate and quantify the effectiveness of our UAS-satellite-based radiometric correction method. Through this analysis, we will be able to develop a strong relationship between traditional UAS-ground boards and our proposed UAS-satellite relationships which can provide a new and effective overall radiometric correction for low-cost COTS sensors. We have performed some preliminary analysis using the 1^st generation MAPIR ground target board (Fig. 2) as ground truth and applied the resulting correction curves on the same data set mentioned in the previous paragraph (KU Field Station fields). The calibrated NDVI image can be seen in Fig, 3. In the final version, we will include the comparison results using both ground calibration board and spectroradiometer. Finally, we will extend our analysis to further UAS-satellite image comparisons for images acquired on different days and times.

In summary, our final correction methodology will include the following: 1) novel UAS-satellite multi-resolution comparison and data association, 2) comprehensive evaluation and validation of obtained correction curves using ground target boards and spectroradiometers, and 3) implementation of the developed procedure for UAS-satellite images procured on different days and sun conditions.