Inversion of rare earth element (REE) abundances in MgO-rich (MgO grea
ter-than-or-equal-to 6 wt.%) igneous rocks, described initially by McK
enzie & O'Nions (1991), provides a useful means of constraining the di
stribution of melt fraction with depth, X(z), in the mantle. Here, we
investigate the sensitivity of the REE inversion procedure to noise in
measured REE abundances and to variation in parameters which must be
specified a priori. Inverted melt distributions are insensitive to rea
listic amounts of random noise but are sensitive to systematic noise.
Melt distributions are also sensitive to input parameters such as dept
h and range of melting, source composition, and mantle mineralogy. To
determine unique melt distributions, additional independent constraint
s on at least two of the following input parameters are therefore requ
ired: depth and range of melting (i.e., lithospheric thickness); sourc
e composition; source mineralogy (dependent on mantle temperature). Th
e successful replication of melt compositions where these three parame
ters are well constrained and the agreement between inverted melt dist
ributions and melt distributions calculated independently assuming ise
ntropic melting during adiabatic upwelling indicate that the scheme is
reliable. REE inversion therefore provides a means for both routinely
constraining X(z) and for determining unknown input parameters.