The algorithm of remote sensing thermal imagery calibration dedicated for UAV-based hydrological studies.

Remote sensing measurements of land surface temperature play a key role in the estimation of evapotranspiration. Thermal cameras used in Unmanned Aerial Vehicles are prone to errors manifested by fluctuations in temperature readings of the same object on different thermal images, vignette effect, and bias against the actual temperature. These problems were addressed with the calibration algorithm. It consists of two steps: i) georeferencing of images using EXIF data, key points matching, and global optimization of relative image positions with the gradient descent method; ii) calibration of temperature offsets occurring between images by correcting sequentially for differences between values on overlapping areas of adjacent images starting from single reference image. The calibration principle is based mainly on the observation that the temperature readings from two overlapping thermal images are shifted by an offset that is approximately constant for the entire overlapping area of the images. Thanks to the algorithm used, it was possible to increase the precision of the georeferencing of aerial images to a level that allows the direct creation of a mosaic of images without the photogrammetric software and reduce the standard deviation of the water surface and vegetation temperature measurements.

Research was partially supported by the National Science Centre, Poland, project WATERLINE (2020/02/Y/ST10/00065), under the CHISTERA IV programme of the EU Horizon 2020 (Grant no 857925) and the "Excellence Initiative, Research University" program at the AGH University of Science and Technology.