Segmenting and characterising ripple patterns on sand dunes using machine learning and 2D semi-variogram

Sand ripples, shaped by fluid flow like wind or water, are common on dunes on Earth and Mars. Their patterns reveal local transport conditions, offering insights into wind regimes where direct observations are lacking. Since manual mapping is slow and subjective, automated methods are essential for consistent large-scale analysis. This study presents two novel and complementary methods for mapping ripple patterns on Martian dunes using high-resolution imagery: a U-Net model for pattern classification and a 2D semi-variogram for measuring ripple spacing and orientation. Tested on 42 barchan dunes across six Martian regions, the U-Net showed reliable ripple classification (F1-score 79 %), while the variogram method achieved high accuracy for ripple spacing (R 2 = 0.78) and orientation (R 2 = 0.98). Together, these approaches enable efficient, large-scale analysis of ripples for sediment transport on any planetary surface and can be applied to other patterned features. • Ripple patterns on Martian dunes are mapped and characterised. • The U-Net machine learning model efficiently segments ripple textures. • The 2D semi-variogram accurately measures ripple spacing and orientation.