The travel time misfit is an important argument for the existence of
strong temporal variations in structure. These parameters reflect how
well the derived velocity model satisfies the observed data. As discussed
in Section 3, the variance is reduced by just 15% for the simultaneous
inversion of all data—a very low reduction for tomographic inversion.
The full dataset cannot be fit by a single model. In Fig. 2C and D we
present the variance reductions and RMS travel time residuals for
forty inversions of overlapping subsets corresponding to yearly windows
with three months of time steps in a period from 1999.01.01 to
2009.12.31. When inversions were performed in one-year windows,
the variance reductions were generally higher than for the case of the
simultaneous inversion. Note that in the case of the entire dataset, we
selected only the best events with a maximum number of picks, while
for the yearly subsets, we used more data of poorer quality. While this
would normally ensure lower noise and higher variance reduction,
the fact that we observe a larger variance reduction means that there
are considerable time variations in velocity.