Correction of Landsat thermal images in the study of tundra landscape transformations: the case of the Urengoy oil and gas condensate field

Abstract

A method for the relative radiometric correction of Landsat thermal images to quantify the changes in the surface temperature of tundra landscapes has been developed. A distinctive feature of the methodology is the use of unsupervised classification algorithm to determine pseudo-invariant areas with identical spectral characteristics of the reference and corrected thermal images. The error in temperature contrast correction is minimized by iteratively determining the optimal number of classes and linear regression coefficients using the cross-validation method. The proposed methodology allows to reduce errors by 2–5 times during temperature contrasts correction, which, in general, indicates its effectiveness. Under experimental conditions, the absolute correction error corresponding to the threshold sensitivity of thermal images (0.4 K) can be achieved for temperature contrasts less than 3 oC.