Abstract: When operating in downward-looking mode, airborne array radars exhibit strong distance dependence (non-stationarity) of near-field clutter in the azimuth-Doppler domain. This severely compromises the performance of traditional space-time adaptive processing (STAP) methods due to the lack of independently identically distributed (IID) training samples. To address this issue, we propose a STAP method combining geometric mapping-based Doppler compensation with joint dimensionality reduction using sum-difference beams. By establishing a geometric model for radar clutter, deriving the relationship between clutter's spatiotemporal direction vectors and distance thresholds, and constructing a distance-dependent Doppler shift compensation matrix, we develop a 9-dimensional STAP architecture that integrates three-channel sum-difference beam data. This approach maintains sufficient adaptive flexibility while significantly reducing computational complexity and training requirements. Simulation results demonstrate that our method effectively suppresses distance non-stationarity in clutter, achieving over 10 dB improvement in compensation factor performance, and outperforms conventional STAP methods in detecting slow-moving targets.