Best I could come up with was differential readings between static ports L/R, but...
Yer, that could work, and/or dynamic pressure from the pitot probes perhaps? - just to give an estimation.
Which in turn might explain why it is suppressed until 100 kts - when the airflow over the fuselage has stabilised?
Edit:
Checked my old paper FCOM and it just says "wind speed and direction arrow [on the ND] is provided by the ADIRS", but it doesn't say which elements of air data are used.