A Network‐Constrained Spatiotemporal Scan Statistic for Identifying Geographic Flow Hotspots

Geographic flows represent the movement of geographic objects between locations at different times. Flow hotspots reflect aggregated movements within specific spatiotemporal ranges. Accurately identifying these hotspots helps reveal their underlying causes and provides targeted insights. Among existing methods, scan statistics are widely used due to their flexible windows and unified significance testing. However, traditional Euclidean cylindrical windows are inadequate for road networks. Spatiotemporal scanning windows of network‐constrained flows must account for network topology and origin–destination temporal precedence. To address this issue, this study proposes a novel spatiotemporal scanning window that integrates network topology and temporal constraints and develops the scan statistic under completely spatiotemporal randomness (CSTR) and global time permutation (GTP) null models. Constructed via the Cartesian product of origin and destination network‐constrained prisms, the window adaptively extends along road networks to precisely capture flow distribution. Synthetic experiments demonstrate that our method outperforms baseline methods in identifying hotspots within complex scenarios. Case studies show that the CSTR model effectively identifies dominant macroscopic tidal travel patterns (such as commuting hotspots), whereas the GTP model successfully overcomes the interference from inhomogeneous backgrounds to accurately capture fine‐grained flow hotspots during off‐peak hours and between specific transit hubs.