Researchers have built a computational model that recreates how people move across a city without needing expensive travel surveys. The framework combines a synthetic population, the built environment, and transport network data to generate daily trips and assign each trip the mode of travel—walking, car, or public transit—using a calibrated multinomial logit model that applies travel‑time coefficients and alternative‑specific constants.

The study applied the model to about 475,000 synthetic residents of Niterói, Brazil. When the model’s predicted neighborhood‑level public‑transit activity was compared with an empirical activity proxy derived from image data, the two patterns matched closely. The Pearson correlation coefficient of 0.885 and Spearman rank correlation of 0.873 show a strong linear association between simulated and observed transit use across the 52 neighborhoods.

The analyses also revealed spatial differences in mode choice. Walking trips are mostly short (average distance ≈ 2.5 km) and common in dense central areas. Car use grows toward the outskirts, while public‑transit use remains high along major bus corridors (average distance ≈ 6.9 km for transit vs. ≈ 9.8 km for car). A hypothetical 30 % reduction in walking time to bus stops would raise the modeled public‑transit‑to‑car ratio by 11.3 %, particularly in peripheral neighborhoods.

These findings underscore the potential of synthetic‑population modeling to inform transit planning, but the authors caution that the model’s assumptions limit validation. Simplified building classifications, lack of household constraints, vehicle ownership, monetary costs, and individual preference differences mean that the simulated activity should be interpreted as a useful proxy rather than a perfect replication of real behavior. The calibration relies on regional modal shares and assumes constant speeds and simplified transit routing, so the intervention outcomes are hypothetical.