From Cycling Efficiency to Cycling State: An Exploratory Study Decoupling the Environmental Determinants of Cycling Detour and Trajectory Stability

Technical Summary: From Cycling Efficiency to Cycling State

Problem Statement
Current evaluations of urban cycling environments are predominantly constrained by an "efficiency-first" macroscopic paradigm. Traditional metrics, such as the Behavioral Detour Ratio (BDR) and Normalized Degree of Deviation (NDD), quantify physical constraints and route choices based on Origin-Destination (O-D) pairs. While effective for measuring macroscopic path efficiency, these indicators create a critical evaluation blind spot: they fail to capture the dynamic, microscopic "process state" of cycling. Specifically, they overlook the high-frequency spatial friction and "micro-perturbations" cyclists experience in dense street networks due to tidal pedestrian flows, mixed-traffic conflicts, and functional agglomeration. Furthermore, existing research relies heavily on linear statistical models (e.g., Structural Equation Modeling) which assume constant marginal effects, rendering them mathematically incapable of identifying systemic spatial overloads, non-linear mutation thresholds, or the "cliff-like collapse" of riding quality that occurs under extreme environmental conditions.

Methodology
The study proposes a "three-in-one" multi-scale analytical framework to decouple macroscopic path efficiency from microscopic process states, utilizing crowdsourced GPS trajectory data from Longhua District, Shenzhen. The methodology integrates three distinct analytical stages:

1. Dual Structural Equation Modeling (SEM): Two baseline SEMs were constructed to establish global linear causal transmission chains. One model links road network structure and urban functions to macroscopic detour rates (BDR/NDD), while the second links these variables to the newly proposed "Trajectory Stability" metric. This step validates the distinct causal mechanisms driving macro-efficiency versus micro-state.
2. Explainable Machine Learning (XGBoost & SHAP): To overcome the limitations of linear assumptions, the study employs Extreme Gradient Boosting (XGBoost) to capture non-linear constraints and multicollinearity in high-dimensional spatial data. Shapley Additive Explanations (SHAP) and Two-Dimensional Partial Dependence Plots (2D-PDP) are used to deconstruct "asymmetric long-tail penalty effects" and visualize interactive gaming between spatial variables. This allows for the identification of critical mutation thresholds where environmental factors trigger abrupt changes in cycling behavior.
3. Hierarchical Cluster Analysis (HCA): The continuous non-linear behavioral features identified by machine learning are translated into operational spatial morphological classifications. HCA is applied to BDR, NDD, and Trajectory Stability to verify their statistical independence and generate a typology of cycling environments.

Key Contributions

• Introduction of "Trajectory Stability": The paper introduces a novel geometric proxy derived from crowdsourced GPS data. Defined as the arithmetic mean of heading deviation angles at offset points, this metric quantifies the "instantaneous decision pressure" and micro-state of cyclist-environment interactions. It distinguishes between smooth, coherent riding (high stability) and turbulent, stress-reactive riding (low stability), effectively capturing "micro-perturbations" that traditional detour rates miss.
• Decoupling of Scales and Logic: The study empirically demonstrates that macroscopic detour rates and microscopic trajectory stability are driven by distinct mechanisms. Macro-efficiency is primarily dictated by large-scale network integration and destination attractiveness (Environmental Possibilism), whereas micro-stability is governed by local functional density and physical friction (Environmental Determinism).
• Identification of Non-Linear Thresholds: By utilizing explainable AI, the research identifies critical mutation thresholds that linear models obscure. Notably, it finds that an Office Functional Mixing Degree exceeding 0.63 triggers a "CBD space black hole" effect, causing a precipitous collapse in trajectory stability. Similarly, it identifies a "Golden Corridor" for micro-network local integration (NAIN R1000) between 1.11 and 1.27, where cycling experience is optimal.
• Parametric Governance Typology: The study classifies cycling environments into four distinct spatial morphologies: "Micro Active Travel Safe Haven," "Multi-scalar Topological Overload Zone," "Oasis of Topological Isolation," and "CBD Spatial Black Hole." This classification moves beyond static descriptions to a parametric understanding of spatial pressure fields.

Results

• Dual Causal Mechanisms: SEM results confirm that while road network structure guides the spatial distribution of urban functions to drive detour choices, high-density functional agglomeration exerts a negative physical coercion on trajectory stability.
• Feature Importance: SHAP analysis reveals a "Scale Decoupling": global integration (NAIN R5000) dominates trajectory stability (acting as a "state destroyer" via macro traffic flows), whereas local integration (NAIN R1000) dominates detour rates (acting as an "efficiency safety net").
• Non-Linear Effects: The Office Functional Mixing Degree exhibits a strong "long-tail penalty effect." Below the 0.63 threshold, high mixing can coexist with good stability; above this threshold, the "black hole effect" causes exponential deterioration in both stability and efficiency due to gated boundaries and tidal crowds.
• Visual Verification: Case studies mapped onto the 3D quadrant framework (Detour Rate vs. Trajectory Stability vs. Mileage) visually confirm that high-efficiency routes in high-density areas often suffer from low stability due to dense, low-angle offsets caused by micro-environmental friction.

Significance
The paper claims a paradigm shift in active mobility governance, moving from "static spatial allocation" (based on ideal cross-sectional widths) to "threshold-based parametric interventions." By identifying "highly sensitive mutation points" (e.g., the 0.63 office mixing threshold), the study offers a robust framework for diagnosing "spatial overload zones" where conventional road adjustments are ineffective. It argues that planners must transition from being "space allocators" to "regulators of spatial pressure fields," utilizing precise, non-linear thresholds to prevent systemic overload and ensure resilient active mobility systems in high-density metropolitan areas. The study emphasizes that understanding the "scale interface"—where macro-network potentials overlap with micro-local activities—is essential for accurately diagnosing the quality of the cycling experience.