<p>The Gutenberg–Richter law describes an exponential relationship between earthquake magnitude and frequency, with the <i>b</i> value as a key parameter for quantifying the relative occurrence of large versus small earthquakes. Accurate estimation of the <i>b</i> value is important for various seismological applications, particularly in identifying asperities, i.e., areas of strong coupling between subducting and overriding plates in the subduction zones. The challenge in mapping spatiotemporal variations of the <i>b</i> value and resolving their spatial heterogeneity is small data sets for which the estimation error is large. To balance the two opposite goals, the smallest estimation error and the largest spatial resolution, we propose the overlapping window algorithm. This method partitions seismic data into overlapping windows, allowing for a more detailed characterization of spatial variations in the <i>b</i> value. We applied the algorithm to the 1998–2024 seismicity in the Japan subduction zone, revealing significant <i>b</i> value patterns that can be interpreted in terms of the locked asperity distributions before and after the largest earthquakes. These patterns suggest that the <i>b</i> value changes reflect plate coupling preceding and following major seismic events. The changing shapes of the small <i>b</i> value patches suggest areas where a strong earthquake is likely to occur. In particular, a future <i>m</i>9 earthquake initiated off the coast of Hokkaido and extending to the area off the coast of Honshu seems likely. Our results demonstrate that this approach, when combined with other methods, provides insight into stress redistribution and locked asperity locations, offering a tool for estimation of location and size of possible future large events.</p>