We present a method for inverting surface magnetic data to recover 3-D
susceptibility models. To allow the maximum flexibility for the model
to represent geologically realistic structures, we discretize the 3-D
model region into a set of rectangular cells, each having a constant
susceptibility. The number of cells is generally far greater than the
number of the data available, and thus we solve an underdetermined pro
blem. Solutions are obtained by minimizing a global objective function
composed of the model objective function and data misfit. The algorit
hm can incorporate a priori information into the model objective funct
ion by using one or more appropriate weighting functions. The model fo
r inversion can be either susceptibility or its logarithm. If suscepti
bility is chosen, a positivity constraint is imposed to reduce the non
uniqueness and to maintain physical realizability. Our algorithm assum
es that there is no remanent magnetization and that the magnetic data
are produced by induced magnetization only. All minimizations are carr
ied out with a subspace approach where only a small number of search v
ectors is used at each iteration. This obviates the need to solve a la
rge system of equations directly, and hence earth models with many cel
ls can be solved on a deskside workstation. The algorithm is tested on
synthetic tramples and on a field data set.