PyRiverShift: A Gaussian Transect Decomposition Framework for Quantifying Sub-Pixel Lateral Channel Mobility

<jats:p>Along many rivers, lateral channel mobility is a key process, yet quantification from freely available multi-decadal 30 m Landsat imagery is constrained by low pixel resolution relative to channel width. We introduce PyRiverShift, a sensor-agnostic spectral decomposition framework. PyRiverShift fits a 1-D Gaussian to NDWI profiles along valley-perpendicular transects, recovering sub-pixel channel centre position with propagated uncertainty, FWHM width, and quality-flags, without binary water-masking or centreline extraction. A Bayesian Information Criterion (BIC) -guided two-Gaussian mixture model identifies secondary and/or abandoned channels. Applied to the 170 km upper Zambezi floodplain, PyRiverShift achieved median σ_pos = 6.0 m (IQR 5.0–7.8 m, n = 22,126). Statistically significant lateral mobility was detected (92% of transects), with median migration rate 3.1 m/yr (95% CI 2.8–3.4) from 1986-2025. Median FWHM width was 260 m (IQR 205–344 m). Positional uncertainty remains stable across threshold settings but increases 25% beyond 90 m ground sampling distance.</jats:p>