<p>Geographically weighted regression (GWR) is a widely used spatial statistical modeling technique that assigns greater weights to more similar samples during local modeling. However, contemporary GWR primarily relies on prior knowledge to determine the optimal distance metric, an approach that is inherently subjective and incapable of quantifying the influence of different distance metrics on spatial process variation. To address this limitation, this study proposes the multi-weight coupled geographically weighted regression (MCGWR) method, which introduces a novel perspective for spatial statistical modeling by quantifying the impact of different metric spaces on sample similarity. MCGWR first integrates multiple distance metrics, incorporating them into the total weight function as additive components. Subsequently, MCGWR employs specially designed influencing factors (hyperparameters) to quantify the contribution of each distance metric to the variation in spatial processes. This paper derives the gradient of the MCGWR objective function and estimates the model parameters using a gradient descent algorithm. Simulations and real-data experiments demonstrate that MCGWR outperforms ordinary least squares, GWR, and geographical and temporal density regression in terms of accuracy and local fitting. More importantly, the influencing factor parameters constructed by MCGWR correctly reflect the intensity of different distance metrics’ effects on spatial heterogeneity. This approach provides a new lens for exploring the fundamental question—“What determines similarity between samples?”—and yields interpretable insights from within the GWR framework.</p>