A Relative Radiometric Normalization Method for Enhancing Radiometric Consistency of Landsat Time-Series Imageries

Radiometric consistency of multi-temporal satellite observations is affected by sensor stability and scene related issues. Relative radiometric normalization (RRN) a widely-used method to reduce these differences, its performance depends on the accurate identification representative pseudo-invariant features (PIFs). However, existing RRN methods are mainly developed for bi-temporal images limited time-series imageries due complexity identifying effective PIFs. In this study, we proposed novel enhance Landsat imageries. This includes (1) trend based PIFs considering land cover changes phenological trends from entire time series; (2) optimization involving an automatic reference selection refining each reference-target image pair; (3) combined modeling using M-estimator sample consensus algorithm robust linear regression. The surface reflectance products were used validate method. experimental results showed that provided consistent all pairs; aided allocation, filtered proper with high spectral spatial similarity pair in monthly stack; achieved good model precision radiance improvement; (4) outperformed seven commonly-used majority stack December. normalized can help generate more comparable analysis reducing uncertainties calibration, atmospheric correction, differences.