Interferometric virtual source (VS) redatuming converts surface-triggered source records into the equivalent records as if they originated from buried receiver locations by crosscorrelating downgoing waves with the corresponding upgoing waves. The theory suggests that when the receivers are surrounded by an enclosing boundary of sources, then the VS has an isotropic radiation pattern and yields an accurate response. The resultant records should determine improvement in the seismic repeatability and image quality compared with non-VS. However, in the presence of a complex near surface, an intricate shallow structure and highly variable weathering layers can severely distort the raypath, such that it produces uneven angle coverage to the buried VS. In addition, near-surface reverberations, surface multiples, and other mode-converted waves may leak into the time-gated early arrivals and further corrupt the direct wavefields. The above-mentioned issues can result in distorted radiation patterns and contaminated responses of the VS. We address these issues explicitly by spatially filtering the potentially contaminated direct wavefields using a zero-phase matched filter, such that the filtered wavefield is consistent with a model-based ideal direct P-wavefield observed at common receiver locations. This ideal reference response is computed from a homogeneous approximation to the local near-surface overburden on top of each VS. The phases of the original direct arrivals are preserved. Components associated with the reverberations and other noises can be effectively suppressed as their spatial radiation patterns deviate from that of the ideal single P-wave mode. Toward an isotropic radiation pattern by the iterative matched filter, we reduce the unbalanced illumination arising from imperfect source coverage and near-surface complexity. Compared with previous methods, the new VS approach provides significantly improved image quality and repeatability based on a pilot field of 13 time-lapse surveys, which solved a significant repeatability problem across a 17 month survey gap.