Characterizing earthquake ground motions through 3D simulations is becoming standard practice for seismic hazard assessment in urbanized regions. However, accurate ground‐motion predictions require shear‐wave velocity () data at depths that capture the extent of the sedimentary column (usually greater than 30 m), which can be difficult to obtain. We acquired microtremor array data at 11 sites in the Seattle basin, Washington, and applied the wavenumber‐normalized spatial autocorrelation (SPAC) method (krSPAC) to obtain at depths as great as 2200 m. In a traditional SPAC approach, modeling high wavenumbers within the SPAC spectrum requires array symmetry. By contrast, in the krSPAC approach we transform observed coherency versus frequency spectra to coherency versus (in which and are wavenumber and station separation, respectively) prior to modeling. Through this transformation, the requirement for array symmetry is eased. We deployed seven‐sensor nested irregular triangular arrays, with nominal interstation spacings that varied from about 300 to 2000 m. Comparison of derived from krSPAC to a previous interpretation from ambient‐noise tomography studies suggests a broadly comparable structure in the 250–1000 m depth range with improved resolution at shallower depth. At each site, we interpret a high‐velocity Quaternary boundary in which increases above . Using this boundary as the reference horizon, we calculate ground‐motion amplification of a factor of up to 2 from the overlying Quaternary sediments between 0.3 and 7 Hz, assuming vertically propagating S waves.