Identifying a Saudi Desert Fossil with PCA and a Shell Image Dataset

A hobbyist found a shell-shaped rock at the base of a cliff in Alghat, Saudi Arabia — 500 km from the nearest coast — and set out to identify it without any paleontology background. The region was submerged in the late Jurassic (~150 million years ago), which explains how marine fossils ended up in the desert, but the curiosity was figuring out what species the fossil might belong to using only its morphology.

The approach was pure shape analysis on the Zhang et al. shell dataset of 7,894 species and 59,244 images. After normalizing for translation, scale, pitch, yaw, and roll (using the longest radius as a rotation anchor), each shell was reduced to a 256-point contour. PCA collapsed those 256 dimensions into two that preserved 67% of variance: PC1 captured ‘pointiness’ and PC2 captured vertical symmetry. Plotting the latent space revealed that round shells cluster tightly and symmetrically, while pointy shells are more varied and rougher.

The nearest neighbor to the Alghat fossil turned out to be Sphincterochila candidissima — a species whose earliest known fossil is only 38 million years old, far younger than the Jurassic strata where the rock was found. The author concedes morphology alone can’t establish lineage, but suggests the resemblance hints at convergent evolution under similar environmental pressures.