This document by the same authors of the CERES dataset
www-geodyn.mit.edu/konopliv_dawn_ssr11.pdf states, by the end of the page that has Fig. 6, that the coefficients are chosen such that the integral of the squared harmonics equals the surface of a unit sphere (4pi), so I'm assuming that the same normalization for Legendre functions used with the GGM0X models should be used with this dataset ...?