The scaling of human mobility by taxis is exponential

As a significant factor in urban planning, traffic forecasting and prediction of epidemics, modeling patterns of human mobility draws intensive attention from researchers for decades. Power-law distribution and its variations are observed from quite a few real-world human mobility datasets such as the movements of banking notes, trackings of cell phone users’ locations and trajectories of vehicles. In this paper, we build models for 20 million trajectories with fine granularity collected from more than 10 thousand taxis in Beijing. In contrast to most models observed in human mobility data, the taxis’ traveling displacements in urban areas tend to follow an exponential distribution instead of a power-law. Similarly, the elapsed time can also be well approximated by an exponential distribution. Worth mentioning, analysis of the interevent time indicates the bursty nature of human mobility, similar to many other human activities.

💡 Research Summary

This paper investigates the statistical properties of human mobility using an unprecedentedly large set of high‑resolution GPS traces collected from more than 10,000 taxis operating in Beijing. The authors extracted 12,028,929 trips from October–December 2010 (dataset D1) and 9,942,697 trips from October–November 2008 (dataset D2). Each trip is defined by a taxi identifier, origin location, origin time, destination location, and destination time, obtained from the taxi’s operational status (with or without passengers). Trips with unrealistically short (< 1 min) or excessively long (> 120 min) durations were filtered out, leaving a clean corpus of urban travel events.

The study focuses on three random variables: (1) displacement Δl, the Euclidean distance between origin and destination; (2) elapsed time ΔT, the travel duration; and (3) inter‑event time τ, the idle period between consecutive passenger‑carrying trips. The authors first examine the probability density function (PDF) of Δl. Visual inspection shows a rapid rise up to about 2 km, a peak, and then a sharp decline. Approximately 98 % of trips are shorter than 20 km, confirming that the dataset captures predominantly intra‑urban movement.

To determine the functional form of the displacement distribution, the authors compare two canonical models: a power‑law P(Δl) ∝ Δl^−α and an exponential P(Δl) ∝ e^−λΔl. Parameters are estimated by maximum likelihood (MLE) and model quality is assessed using the Akaike Information Criterion (AIC). The displacement range is split into two parts: (i) 0–20 km (urban core) and (ii) > 20 km (suburban fringe). For both parts and for both datasets, the exponential model yields AIC weights (W_exp) essentially equal to 1, while the power‑law weights (W_pow) are zero. The estimated decay rates are λ ≈ 0.233 km⁻¹ for the short‑range part and λ ≈ 0.170 km⁻¹ for the long‑range part, with R² values exceeding 0.97, indicating an excellent fit. The power‑law exponents (α ≈ 1.49 for short trips, α ≈ 5.10 for long trips) are statistically insignificant because of the overwhelming AIC evidence for the exponential.

The authors also explore geographic heterogeneity by extracting displacement distributions for five representative 1‑km‑radius zones (Beihang University, a subway station, Beijing Capital International Airport, a central business district, and a residential district). All zones, except the airport (which naturally generates longer trips), follow the same exponential pattern, confirming that the observed scaling is not an artifact of spatial inhomogeneity.

Next, the elapsed time ΔT is analyzed. Its PDF also follows an exponential distribution with a decay parameter closely matching that of Δl, reflecting a near‑linear relationship between distance and travel time (Pearson correlation ≈ 0.78). This suggests relatively stable average speeds across the city and reinforces the exponential nature of the underlying process.

The inter‑event time τ, representing periods when taxis are empty, displays a heavy‑tailed (power‑law) distribution, indicative of bursty human activity: many short idle intervals interspersed with occasional long gaps. This pattern aligns with findings from other domains (e‑mail, phone calls) and points to the intrinsic burstiness of demand for transportation services.

Methodologically, the paper’s strengths lie in (i) the massive scale and fine temporal granularity of the dataset, (ii) rigorous model selection using AIC and MLE rather than visual fitting alone, and (iii) simultaneous treatment of distance, time, and demand intervals. Limitations include potential GPS measurement error, the fact that taxi trips may not represent the full spectrum of human mobility (e.g., walking, subway, private car), and the absence of post‑2010 data that could capture recent changes in mobility patterns (e.g., ride‑hailing services).

The findings have practical implications. Traffic forecasting and real‑time dispatch algorithms can adopt exponential travel‑distance kernels, simplifying computations while preserving accuracy. Epidemic models that rely on human movement kernels may also benefit from an exponential assumption, leading to more realistic spread predictions in dense urban settings. Moreover, the bursty inter‑event distribution suggests that demand‑responsive fleet management should anticipate periods of high request intensity followed by lulls.

In contrast to earlier studies based on bank‑note tracking, mobile‑phone location logs, or inter‑city vehicle traces—where power‑law or truncated power‑law behaviors were reported—the present work demonstrates that, when focusing on intra‑urban taxi trips with fine spatial resolution, the displacement and travel‑time statistics are fundamentally exponential. This challenges the prevailing notion that human mobility universally follows Lévy‑flight dynamics and highlights the importance of data granularity and transportation mode in shaping observed scaling laws.

In summary, the paper provides robust empirical evidence that human mobility in Beijing’s urban environment, as captured by taxi GPS trajectories, is governed by exponential distributions for both travel distance and duration, while the demand‑gap (inter‑event) times exhibit bursty, power‑law characteristics. These insights enrich our understanding of urban mobility, refine modeling assumptions for transportation and epidemiological applications, and open avenues for further comparative studies across cities and mobility modalities.