I showed that the forward-scattered and back-scattered wavefields have overlap in their estimated slowness models. The synthetic example showed that supplying the back-scattered information can improve the tomographic inversion results in terms of convergence rate and accuracy. The overlap is strongest in the low-frequency part of the data. Therefore, the data must contain low frequencies for the combined gradient to give the best results. Moreover, I used a scalar weight that does not vary with iterations to calculate the combined gradient. Varying the weights with iterations could improve the results even more.