The long-term stability of cratons has been attributed to low temperatures and depletion in
iron and water, which decrease density and increase viscosity. However, steady-state thermal
models based on heat flow and xenolith constraints systematically overpredict the seismic
velocity-depth gradients in cratonic lithospheric mantle. Here we invert for the 1-D thermal
structure and a depth distribution of metasomatic minerals that fit average Rayleigh-wave
dispersion curves for the Archean Kaapvaal, Yilgarn and Slave cratons and the Proterozoic
Baltic Shield below Finland. To match the seismic profiles, we need a significant amount of
hydrous and/or carbonate minerals in the shallow lithospheric mantle, starting between the
Moho and 70 km depth and extending down to at least 100-150 km. The metasomatic
component can consist of 0.5-1 wt% water bound in amphibole, antigorite and chlorite, ~0.2
wt% water plus potassium to form phlogopite, or ~5 wt% CO2 plus Ca for carbonate, or a
combination of these. Lithospheric temperatures that fit the seismic data are consistent with
heat flow constraints, but most are lower than those inferred from xenolith
geothermobarometry. The dispersion data require differences in Moho heat flux between
individual cratons, and sublithospheric mantle temperatures that are 100-200°C less beneath
Yilgarn, Slave and Finland than beneath Kaapvaal. Significant upward-increasing
metasomatism by water and CO2-rich fluids is not only a plausible mechanism to explain the
average seismic structure of cratonic lithosphere but such metasomatism may also lead to the
formation of mid-lithospheric discontinuities and would contribute to the positive chemical
buoyancy of cratonic roots.