Exploring the Relationship Between the Built Environment and Spatiotemporal Heterogeneity of Urban Traffic Congestion During Tourism Peaks: A Case Study of Harbin, China

Abstract

Understanding the spatial heterogeneity of traffic congestion drivers is crucial for data-informed urban planning in tourist cities. This study investigates the spatiotemporal relationship between built environment characteristics and traffic congestion in the central urban area of a major northern Chinese tourist city. We apply a Multiscale Geographically Weighted Regression (MGWR) model to geospatial data across four typical peak periods and benchmark the results against Ordinary Least Squares (OLS) and Geographically Weighted Regression (GWR). The MGWR model demonstrates superior capability in capturing spatial non-stationarity and multiscale effects. The results reveal strong spatiotemporal heterogeneity in the effects of built environment factors on congestion. Intersection density demonstrates a stronger mitigating effect during weekday evening peaks. Catering facilities significantly exacerbate congestion in tourist hotspots. Tourism-related facilities such as hotels and attractions intensify congestion during weekend peaks. Parking availability shows dual impacts, with peripheral parking reducing pressure and central clustering worsening congestion. Our geospatially disaggregated results provide empirical evidence for location-sensitive and temporally adaptive traffic management and urban design strategies. This study highlights the value of MGWR-based spatial modeling in supporting geoinformation-driven urban mobility planning.

1. Introduction

Urban traffic congestion has emerged as a critical issue faced by cities worldwide, driven by a combination of factors such as high population density, rapid growth in private vehicle ownership, and the surge in ride-hailing and delivery services [1]. Congestion not only leads to significant travel delays and reduced economic efficiency, but also contributes to excessive energy consumption and environmental degradation, posing serious challenges to urban sustainability. With the accelerating pace of motorization, peak-hour traffic congestion has become increasingly severe in many large Chinese cities. This issue is particularly pronounced during tourism seasons, when popular destinations such as Harbin, Xi’an, and Hangzhou experience substantial surges in traffic volume, placing considerable strain on urban transport systems. These circumstances call for an urgent and in-depth investigation into effective mitigation strategies.

During peak tourism periods, traffic congestion near major attractions often surpasses that of typical commuting hours [2]. The temporal overlap between tourist travel demand and local commuting flows places considerable strain on urban transportation networks, particularly in city centers. The increase in urban tourist volume is significantly associated with heightened levels of traffic congestion [3,4,5]. In this context, examining how the built environment influences traffic congestion is both timely and essential. Such an investigation not only helps to reveal the spatial and temporal patterns of congestion during tourism peaks but also offers actionable insights for urban planning and traffic management. By identifying context-specific drivers of congestion, city planners and policymakers can develop more adaptive, data-driven strategies to mitigate pressure on the transport system and enhance mobility in tourism-intensive areas.

Numerous studies have shown that the formation of traffic congestion is closely related to the urban built environment. Elements such as land use layout, road network structure, and the provision of transport infrastructure significantly influence traffic volume and roadway service levels [6]. Improving traffic conditions through the optimization of the built environment is considered one of the key approaches to alleviating congestion. For example, the rational planning of parking facilities is regarded as a critical strategy for mitigating urban congestion. Enhancing public transit accessibility, increasing road density, and promoting mixed land use have also been proven effective in reducing peak-hour traffic congestion [7]. However, the effects of built environment factors vary across spatial contexts. Their influence may differ considerably across regions and spatial scales. Therefore, it is necessary to adopt models capable of capturing spatial heterogeneity to analyze the relationship between built environment features and traffic congestion in greater depth. This study conducts an empirical analysis based in Harbin to explore the relationship between traffic congestion and the built environment during peak tourism periods. Using the Multiscale Geographically Weighted Regression (MGWR) model, the study investigates the spatiotemporal variations in the impacts of built environment factors on congestion and explores the spatial distribution of regression coefficients. The results are further compared with those of Ordinary Least Squares (OLS) and Geographically Weighted Regression (GWR) models to validate the superior performance of MGWR in revealing spatial heterogeneity.

2. Literature Review

To provide a comprehensive foundation for this study, the following section is structured into three thematic parts:

(1)

Research progress on traffic congestion

The causes and management of traffic congestion have long been central topics in the field of transportation research. Early studies approached the problem from traffic engineering and economic perspectives, focusing on the underlying mechanisms of congestion formation. For instance, Downs [8] proposed the “fundamental law of peak-hour congestion,” which posits that increasing supply alone cannot fully eliminate congestion, thereby shifting attention toward demand management strategies. In empirical work, Turner [9] analyzed traffic data from 50 large and medium-sized cities in the United States to evaluate the relationship between various congestion indicators and actual congestion levels. Over the past decade, the rapid advancement of big data technologies has enabled researchers to use real-time traffic data to better characterize urban congestion patterns. For example, Zhao and Hu [10] conducted a long-term spatiotemporal analysis using big data techniques and identified four typical congestion patterns in Beijing. Priambodo et al. [11] utilized sensor data to predict the impact of nearby road congestion on overall traffic conditions. Li et al. [12] employed mobile signaling data to trace the root causes of traffic congestion. Additionally, some scholars have developed Macroscopic Fundamental Diagrams (MFDs) to link city-wide traffic conditions with regional road volumes, offering a quantitative description of how congestion spreads across urban areas [13]. Overall, existing studies have established a multi-level framework, ranging from theoretical exploration to model-based analysis, providing valuable insights into diagnosing the causes of congestion and developing effective mitigation strategies.

(2)

Relationship between the built environment and traffic congestion

Traditional research on traffic congestion has primarily emphasized expanding road capacity and improving traffic engineering measures. Drawing upon the long-established interdependence between transportation and land use, recent studies highlight the critical role of built environment attributes—such as road network configuration, functional density, and public transit facilities—in shaping traffic patterns and congestion levels. Xiao et al. [13] constructed a path mediation model and suggested that road connectivity can indirectly alleviate congestion by enhancing regional traffic capacity. Wu, Jia, Feng, Li, and Kuang [14], using spatiotemporal statistical models, found that road density and intersection design exert significant delayed regulatory effects in high-density urban areas, acting as either amplifiers or buffers of congestion. Wang et al. [15] highlighted that mismatches between land use and transportation infrastructure are a direct cause of traffic congestion following urban redevelopment. Using taxi trajectory data from Shanghai, Liu et al. [16] revealed strong correlations between traffic flow and different land use types. Nguyen-Phuoc et al. [17] assessed the influence of public transit systems on congestion and advocated for incorporating travel preference models that prioritize public transit in highly congested areas, along with the development of grid-level congestion evaluation indicators. Ding et al. [18], using clustering models, found that each additional kilometer of metro line density per square kilometer could reduce the congestion delay index by approximately 1%. In summary, built environment factors influence traffic congestion through diverse mechanisms. Importantly, the strength and direction of these effects vary across space and time. Therefore, it is essential to integrate spatial analytical methods to identify key drivers of congestion in different regions and time periods.

(3)

Spatial heterogeneity modeling methods

GWR is a widely used method for addressing spatial non-stationarity. It was first introduced by Charlton et al. [19] and later refined and extended by Fotheringham et al. [20], who established a comprehensive theoretical framework and promoted its application across various fields. In the study of urban traffic congestion, Pan et al. [6] demonstrated the effectiveness of GWR in capturing the spatial heterogeneity of traffic conditions and visualized the spatial distribution of parameter estimates. To incorporate temporal non-stationarity and explore spatiotemporal interactions, Liu et al. [21] applied the Geographically and Temporally Weighted Regression (GTWR) model to reveal how the built environment influences traffic congestion over both space and time. However, both GWR and GTWR assume that all explanatory variables operate at the same spatial scale, which is often unrealistic in real-world settings. To address this limitation, the MGWR model was developed. Proposed by Fotheringham et al. [22], MGWR allows each explanatory variable to operate at its own optimal bandwidth, thus accommodating variable-specific spatial scales. Recent empirical studies have shown that MGWR typically achieves better model performance than GWR, often producing higher R² values and lower corrected Akaike Information Criterion (AICc) scores. For example, comparative studies have demonstrated that MGWR provides a more detailed representation of spatial heterogeneity and yields more robust parameter estimates [23,24,25]. Zhou et al. [26] applied MGWR to uncover localized effects of the built environment on bicycle usage. Xie et al. [27] used MGWR to model the spatial effects of factors influencing roadside parking duration, offering data-driven support for urban parking policy enhancement. In summary, MGWR provides a powerful multiscale analytical for exploring the relationship between the built environment and traffic congestion.

As the above literature indicates, existing studies predominantly focus on regular commuting periods, with limited attention paid to traffic patterns during tourism peak seasons. In addition, research on the built environment often relies on a narrow set of indicators and lacks a comprehensive, multidimensional perspective. Furthermore, traditional models such as OLS and GWR are limited in their ability to capture differences in spatial scales across variables, resulting in inadequate identification of spatial non-stationarity. To address these gaps, this study centers on the tourism peak period, constructs a multidimensional indicator system of the built environment, and introduces the MGWR model to enhance the explanatory power for the spatial mechanisms of traffic congestion.

3. Materials and Methods

3.1. Study Area

Harbin serves as a major regional center in Northeast China and is widely recognized for its distinctive historical and cultural heritage, as well as its internationally renowned ice and snow tourism. According to official reports, Harbin received over 3 million tourist visits during the 2023–2024 ice and snow season, setting a new record in tourism revenue. The study area covers the central urban districts located within the Third Ring Road of Harbin, including Daoli, Daowai, Nangang, Xiangfang, and Songbei (Figure 1). Harbin’s central urban area comprises several major commercial centers and tourism clusters, including the Central Street business district, Qiulin commercial zone, and well-known tourist destinations such as Ice and Snow World Scenic Area. These areas represent key urban activity nodes and tourism hotspots within the study area.

Since the implementation of the reform and opening-up policy, Harbin has experienced rapid urbanization and significant growth in its tourism industry. According to the Harbin Statistical Yearbook, the city’s permanent population exceeded 9.77 million in 2023, with the number of registered motor vehicles reaching 4.5 million. As a result, issues such as traffic congestion and carbon emissions have become increasingly severe, particularly during the peak winter tourism season. Therefore, it is crucial to explore the spatial characteristics and influencing factors of traffic congestion in the central urban area during tourism peak periods, in order to promote sustainable urban transport development.

3.2. Data Resources

The data used in this study mainly include four components: real-time traffic condition data, point of interest (POI) data, road network data, and public transportation facility data. The AutoNavi real-time road traffic dataset is produced by fusing GPS traces uploaded by map service users with roadside sensors such as video cameras and induction coils on AutoNavi’s computing platform. Traffic conditions are coded on a five-level scale, where 0 denotes unknown, 1 denotes smooth, 2 denotes slow, 3 denotes mild congestion, and 4 denotes heavy congestion. Each traffic status record is georeferenced to the corresponding road segment coordinates, allowing spatial aggregation and analysis by time period.

To obtain real-time traffic conditions, we accessed the AutoNavi Map API. The AutoNavi real-time road traffic dataset is produced by fusing GPS traces uploaded by map service users with roadside sensors such as video cameras and induction coils on AutoNavi’s computing platform. Each traffic status record is georeferenced to the corresponding road segment coordinates, allowing spatial aggregation and analysis by time period.

Considering the marked weekday–weekend differences in traffic conditions during Harbin’s tourism peak and their cyclical recurrence, we set the observation window to 6–12 January 2025. Harbin’s peak tourist season typically spans late December to mid-February; within this interval we deliberately selected a weather-normal week without extreme events or special traffic controls to minimize confounding from non-recurrent disruptions and to capture representative peak-season patterns. For comparison with non-tourism conditions, we additionally selected an off-peak control week—19–25 May 2025. Real-time traffic information was collected daily between 5:00 and 24:00 at 20-min intervals. Using the ArcGIS 10.7 platform, the collected time-sliced data were georeferenced and converted into raster format.

POI data were obtained from the Baidu Map platform, covering a variety of functional types such as tourist attractions, commercial areas, residential zones, educational institutions, medical facilities, and recreational spaces. The distribution of POIs reflects land use characteristics associated with individual activity locations [28]. In this study, the density of each POI type within the study area was calculated to assess land use intensity and functional diversity, thereby characterizing features of the built environment. Road network data were derived from OpenStreetMap, including information on road hierarchy, total road length, road density, and the number of intersections, which reflect the supply and connectivity of the road infrastructure. Public transportation data were collected through the Baidu Map API, encompassing the locations and routes of bus and subway stations. Station density and route coverage were computed to represent the level of public transport accessibility across different areas.

All datasets underwent preprocessing procedures including projection transformation, spatial clipping, and spatial association to ensure consistency in coordinate systems and alignment with the study area. The resulting integrated database satisfies the requirements of the MGWR model and provides a solid foundation for analyzing the spatiotemporal effects of the built environment on traffic congestion during peak tourism periods.

3.3. Grid-Based Partitioning of the Study Area

To more precisely capture spatial variations between built environment factors and traffic congestion, the study area was divided using a fishnet grid approach, a method commonly applied in existing research [29,30,31]. Following prior work, we adopt a 500 m × 500 m grid to balance neighborhood detail and data sufficiency. Modifiable areal unit problem (MAUP) checks at 300 m and 1 km are broadly consistent, and results remain stable under alternative zoning schemes. We adopt an equal-area fishnet grid because it reduces shape heterogeneity across zones and avoids boundary misalignment with traffic corridors. Compared with administrative units or Traffic Analysis Zones (TAZs), the grid better captures corridor-level, localized congestion patterns and facilitates transparent MAUP checks across scales. Consistent with this choice and prior studies [32], the area within the Third Ring Road was partitioned into 1135 base cells; after excluding 151 boundary cells that were incomplete or lacked valid data, 984 valid grids were retained for analysis. To minimize boundary effects and improve the accuracy and continuity of spatial metric calculations, each grid cell was buffered by 250 m and used as the fundamental spatial analysis unit.

3.4. Dependent and Explanatory Variables

As noted above, traffic status was originally coded on a five-level scale. Records with a score of 0 accounted for a negligible proportion of observations; these were excluded to avoid noise, and their removal did not materially affect the empirical results. We then dichotomized the remaining records: segments with a score of 1 were classified as non-congested, whereas segments with scores of 2–4 were classified as congested. The average length-weighted proportion of congested roads within Harbin’s Third Ring Road area was calculated separately for weekdays and weekends (Figure 2a,b). To investigate temporal heterogeneity in congestion, four distinct time periods were defined as dependent variables. The contiguous weekday intervals where the congestion share exceeded 10% were labeled Weekday Morning Peak (W-AM) and Weekday Evening Peak (W-PM). The remaining intervals with shares above 5% on weekdays and weekends were labeled Weekday Off-Peak (W-OFF) and Holiday/Weekend Peak (H-PK). The two cutoffs were chosen in a data-driven manner to separate spike conditions from the elevated baseline. Robustness checks that applied alternative thresholds of 8% and 12% for the peaks and 4% and 6% for the elevated regime, and that shifted window boundaries by ±30 min, led to the same substantive conclusions, with only minor changes in AICc and adjusted R² and no meaningful differences in residual Moran’s I.

Guided by existing literature and model performance [6,26,33,34], this study identifies a total of 16 explanatory variables from five dimensions: road characteristics, land use, tourism-related factors, daily service factors, and parking and public transport infrastructure (

Table 1. Candidate explanatory variables.

Category	Name	Description
Urban road features	Road density	Density of road networks within each grid cell
	Intersection density	Density of road intersections within each grid cell
Land use features	Land use mix	Shannon entropy index within each grid cell
Tourism-related factors	Catering	Mean value of kernel density raster cells for POIs such as Chinese restaurants, fast food, hotpot, bakeries, specialty restaurants, cold drinks, international cuisine, cafés, teahouses, dessert shops, etc., within each grid cell
	Cultural and recreational	Mean value of kernel density raster cells for POIs such as libraries, museums, science and technology centers, art galleries, planetariums, cultural centers, theaters, and amusement parks within each grid cell
	Accommodation	Mean value of kernel density raster cells for POIs such as hotels and hostels within each grid cell
	Tourist attraction	Mean value of kernel density raster cells for POIs such as attractions, parks, zoos, botanical gardens, aquariums, temples, city squares, memorials, churches, etc., within each grid cell
	Tourism-related retail	Mean value of kernel density raster cells for POIs such as duty-free stores, shopping streets, specialty markets, department stores, and malls within each grid cell
Daily life-related factors	Residence	Mean value of kernel density raster cells for POIs such as residential complexes and mixed-use buildings within each grid cell
	Education	Mean value of kernel density raster cells for POIs such as schools, research institutes, and driving schools within each grid cell
	Healthcare	Mean value of kernel density raster cells for POIs such as hospitals, clinics, and pharmacies within each grid cell
	Daily retail	Mean value of kernel density raster cells for POIs such as department stores, electronics stores, home furnishing markets, stationery shops, pet markets, supermarkets, convenience stores, and marketplaces within each grid cell
	Enterprises	Mean value of kernel density raster cells for POIs such as industrial parks, companies, factories, and agricultural enterprises within each grid cell
Parking and transportation facilities	Parking facilities	Density of parking facilities within each grid cell
	Public transit stations	Density of bus and subway stations within each grid cell
	Bus route density	Density of bus routes within each grid cell

). These variables are used to systematically investigate the mechanisms through which the built environment affects traffic congestion. Road density and intersection density are core indicators of network accessibility and connectivity. While a denser road network can improve travel flexibility, it may also lead to increased traffic conflicts and thus congestion [33]. Land use mix reflects the intensity and complexity of regional functions and is often highly correlated with travel demand [35]. During tourism peak periods, the clustering of tourism facilities such as restaurants, hotels, and scenic spots significantly increases short-term traffic pressure [2]. Daily service facilities, although primarily serving routine needs, can generate concentrated travel demand during certain hours and impose pressure on the surrounding road network. Hence, they should be included as essential factors influencing congestion [6]. Furthermore, the reasonable allocation of parking resources directly affects road efficiency. Insufficient or uneven parking may intensify illegal parking and circling behavior [36]. Lastly, the density of bus and subway stations and the coverage of transit lines reflect the level of public transportation service. Studies suggest that higher transit density contributes significantly to easing urban traffic congestion [28]. Constructing a multi-dimensional explanatory variable system of the built environment helps reveal the spatial factors and mechanisms underlying urban traffic congestion.

3.5. Definition and Quantification of Variables

In this study, urban road features are measured by two variables: road density $D_r$ and intersection density $D_S$. The calculation formulas are as follows:

$$\begin{array}{c}Dr={\displaystyle \frac{{\sum }_{i\mathsf{\in }M} {\alpha }^{i}le{n}_{Rd}^i(g)}{\mathit{cellArea}}}\end{array}$$

(1)

$$D_S={\displaystyle \frac{N_{\mathit{sec}}}{\mathit{cellArea}}}$$

(2)

where $le{n}_{\mathit{Rd}}^i(g)$ is the total length of roads of class $i$ within grid cell $g$, and ${\alpha }^i$ is the class-specific weight reflecting effective traffic capacity. We use $M=\{\mathrm{motorway}, \mathrm{trunk}, \mathrm{primary}, \mathrm{secondary}, \mathrm{tertiary}\}$ and set ${\alpha }^i=\{4, 3, 3, 2, 1\}$ for these classes respectively, approximating relative effective capacity. This capacity-weighted scheme avoids treating major arterials and minor roads as equivalent when lane-level data are unavailable. $D_S$ is the intersection density, and $N_{\mathit{sec}}$ is the number of intersections within the grid cell.

Land use mix for each grid cell was calculated based on POI data. Major POI categories with clearly defined functions and distinct classifications were selected as the basis for computation. The Shannon entropy index was adopted to quantify land use mix, and the formula is as follows.

$$\begin{array}{c}LUM=-{\sum }_{i=1}^k p_i·{\displaystyle \frac{\mathit{ln}\left(p_i\right)}{\mathit{ln}\left(k\right)}}\end{array}$$

(3)

where $p_i$ represents the proportion of POIs of category i within the buffer zone relative to all POIs, and k denotes the total number of POI functional categories included in the calculation.

Tourism-related features and daily-service features in this study were derived from various categories of POI data. Kernel density estimation was first performed, followed by the extraction of average values of raster cells within each analysis grid. To capture the spatial clustering of different POI types within each grid cell, the Kernel Density tool in ArcGIS was applied using the number of POIs per category as input. The calculation formula is as follows:

$$\begin{array}{c}f\left(x\right)={\displaystyle \frac{1}{n{h}^2}}{\sum }_{i=1}^n K\left({\displaystyle \frac{x-x_i}{h}}\right) \end{array}$$

(4)

For all $n_j$ raster cell centers within the analysis unit $U_j$, the arithmetic mean of their kernel density estimates is calculated, yielding the POI density indicator for the unit:

$$D_j={\displaystyle \frac{1}{n_j}}{\sum }_{k=1}^{n_j} \widehat{f}\left(x_k\right)={\displaystyle \frac{1}{n_j}}{\sum }_{k=1}^{n_j} \left[{\displaystyle \frac{1}{n{h}^2}}{\sum }_{i=1}^n K\left({\displaystyle \frac{x_k-x_i}{h}}\right)\right]$$

(5)

where f(x) denotes the kernel density estimate at location x; n is the number of POIs of a given type within the analysis unit; h is the search radius; K is the spatial weighting function; $x_i$ represents the location of the i-th POI; $n_j$ is the number of raster cell centers within unit $U_j$; and $x_k$ indicates the center coordinate of the k-th raster cell within unit $U_j$.

The characteristics of parking and transit infrastructure are represented by three variables: the density of parking facilities $D_P$, the density of metro stations $D_m$, and the density of bus routes $D_b$. The calculation formulas are as follows:

$$D_P={\displaystyle \frac{P_{\mathit{park}}}{\mathit{cellArea}}}$$

(6)

$$D_m={\displaystyle \frac{N_{\mathit{bus}}+N_{\mathit{metro}}}{\mathit{cellArea}}}$$

(7)

$$D_b={\displaystyle \frac{{\sum }_{i=1}^m L_i}{\mathit{cellArea}}}$$

(8)

where $D_P$ denotes the density of parking facilities, and $P_{\mathit{park}}$ represents the number of parking facilities within the grid cell. $D_m$ is the density of metro stations, and $N_{\mathit{bus}}$ refers to the number of bus stops within the grid cell. $N_{\mathit{metro}}$ represents the number of metro stations within the grid cell. $D_b$ denotes the density of bus routes, where $L_i$ is the length of the i-th bus route segment within the grid cell, and m is the total number of bus routes traversing the grid cell.

The average traffic condition is used in this study as an indicator of traffic congestion. Specifically, the mean value of congestion levels observed at all monitoring points within each grid cell is calculated to represent the overall traffic condition of that cell.

3.6. Regression Models

While the traditional OLS model can evaluate the overall impact of the built environment on congestion, it assumes spatial homogeneity and thus fails to capture geographic variations. The GWR model addresses this limitation by employing a spatial weighting function to estimate local regression parameters for each spatial unit, thereby revealing the spatial variability of variable effects. Its basic form is as follows:

$$y_i={\beta }_0\left(u_i,v_i\right)+\sum _k {\beta }_k\left(u_i,v_i\right)x_{\mathit{ik}}+{\epsilon }_i$$

(9)

where $y_i$ is the dependent variable for spatial unit i; $x_{\mathit{ik}}$ is the value of the k-th explanatory variable for spatial unit i; $\left(u_i,v_i\right)$ denotes the geographic coordinates (longitude and latitude) of spatial unit i; ${\beta }_0\left(u_i,v_i\right)$ and ${\beta }_k\left(u_i,v_i\right)$ represent the intercept and the regression coefficients at location $\left(u_i,v_i\right)$, respectively; ${\epsilon }_i$ is the residual term.

MGWR is a multiscale extension of the GWR model, allowing each explanatory variable to be estimated at its optimal spatial scales [22]. To uncover the temporal heterogeneity in how built environment factors influence congestion, this study conducts time-segmented modeling for weekdays and weekends. The model is specified as follows:

$$y_i={\beta }_{{\mathit{bw}}_0}\left(u_i,v_i\right)+\sum _k {\beta }_{{\mathit{bw}}_k}\left(u_i,v_i\right)x_{\mathit{ik}}+{\epsilon }_i$$

(10)

where $y_i$, $x_{\mathit{ik}}$, $\left(u_i,v_i\right)$, and ${\epsilon }_i$ have the same meanings as in the previous equation. ${\beta }_{{\mathit{bw}}_k}\left(u_i,v_i\right)$ denotes the local regression coefficient for the k-th variable at location $\left(u_i,v_i\right)$, estimated based on its own optimal bandwidth ${\mathit{bw}}_k$. ${\beta }_{{\mathit{bw}}_0}\left(u_i,v_i\right)$ represents the local intercept term, corresponding to an independent bandwidth ${\mathit{bw}}_0$.

Prior to the main analysis, an OLS model was estimated, using traffic congestion levels during different time periods as dependent variables and the 16 built environment variables as independent variables for preliminary fitting. The VIF test results showed that all variance inflation factors were below 5, indicating no severe multicollinearity issues and thus no need to exclude any variables from the model. These explanatory variables were then incorporated into the subsequent modeling analysis.

4. Results and Analyses

4.1. Temporal Variation in Traffic Congestion

This study first provides an overall statistical analysis of the proportion of congested roads for the tourism peak and the tourism off-peak (Figure 2a,b). The results show that, except for the morning peak where the congestion share is comparable across periods, congestion is consistently higher during the tourism peak than during the tourism off-peak at other times of day. This pattern indicates an additional congestion burden attributable to tourist demand. Accordingly, the empirical analysis that follows focuses primarily on the tourism-peak period. During the tourism peak (Figure 2a), the results reveal distinct temporal patterns between weekdays and weekends, reflecting typical commuting and leisure travel behaviors. On weekdays, traffic congestion exhibits a clear “double-peak” pattern, closely aligned with urban commuting activities. The morning peak occurs between 7:20 and 9:00, with the highest level of congestion observed around 8:20, when approximately 13% of roads are congested—marking a pronounced travel peak. The evening peak appears between 16:40 and 18:40, with the highest congestion at around 17:40, when the proportion of congested roads exceeds 23%, significantly higher than at other times. This indicates that traffic pressure is especially intense during the evening commute, with road infrastructure under the greatest strain. This “dual-peak congestion pattern” reflects a typical rigid commuting behavior.

In contrast, congestion intensity on weekends is generally lower, and the congestion curve exhibits a smoother trend. Noticeable congestion does not emerge until around 10:00 a.m., and travel during peak hours is likely dominated by non-commuting activities, reflecting the “non-commuting” nature of urban travel during peak tourist periods. Although the peak congestion intensity throughout the day is lower than that of weekday rush hours, its duration is longer, indicating a certain cumulative effect.

Further observation reveals that between 10:20 and 16:20, the proportion of congested roads on weekends generally exceeds that on weekdays. This may be attributed to large numbers of tourists and local residents engaging in midday and afternoon travel, often for short-distance trips, exhibiting a clear tourism-driven pattern. This phenomenon reflects a shift in the use of urban traffic space during holidays, changing from a commute-oriented model to one oriented toward consumption and leisure activities.

4.2. Spatial Variation in Traffic Congestion

Within Harbin’s inner ring road, traffic congestion exhibits a spatial pattern characterized by central concentration and temporal differentiation. The distribution of high-congestion areas varies significantly across different time periods, reflecting dynamic changes in travel patterns and mobility structures (Figure 3).

During the W-AM period, traffic congestion is primarily concentrated along commuting corridors connecting southern residential areas with the central core of the city. Key routes include Xuefu Road and Hexing Road in Nangang District, as well as roads in the Haxi area and the southeastern part of the city, forming a distinct “commuting radial belt” pattern. During the W-OFF period, overall congestion levels decrease significantly, with a more scattered spatial distribution and spot-like clustering. High-congestion areas are mainly located around the Central Avenue commercial district in Daoli District and the Haxi commercial center. In the W-PM period, congestion reaches its highest intensity of the day. A radial “tidal expansion” pattern emerges, extending from the city center towards the east, west, and south. Multiple local congestion hotspots gradually connect, reflecting a compound congestion effect driven by reverse commuting from workplaces to residences. During the H-PK period, congestion is slightly alleviated compared to the weekday peaks. However, its spatial distribution shows a “discrete hotspot plus scenic aggregation” pattern. High-congestion points are relatively scattered and primarily located around Central Avenue, Ice and Snow World, the Haxi commercial district, and the cultural-tourism corridor in Nangang District. This indicates that traffic pressure during non-working days is largely driven by tourism and commercial activities rather than commuting.

4.3. Modeling Results

The comparative results of the OLS, GWR, and MGWR models are presented in

Table 2. Comparison of model fit among OLS, GWR, and MGWR.

	W-AM			W-OFF
	AICc	Adj R2	Moran’s I (Residuals)	AICc	Adj R2	Moran’s I (Residuals)
OLS	2181.804	0.188	0.475658 (0.00) *	1931.002	0.402	0.414445 (0.00) *
GWR	1969.471	0.508	0.193620 (0.07)	1645.127	0.649	0.191340 (0.06)
MGWR	1622.269	0.666	0.094464 (0.12)	1389.102	0.744	0.093356 (0.12)
	W-PM			H-PK
	AICc	Adj R2	Moran’s I (residuals)	AICc	Adj R2	Moran’s I (residuals)
OLS	1874.552	0.442	0.41912 (0.00) *	1845.067	0.461	0.374854 (0.00) *
GWR	1606.649	0.698	0.189743 (0.08)	1504.035	0.727	0.157528 (0.10)
MGWR	1286.554	0.785	0.121773 (0.14)	1266.302	0.798	0.087753 (0.19)

Notes. “*” denotes p ≤ 0.05.

and

Table 3. Summary of regression results from the MGWR model.

Independent Variable	W-AM			W-OFF			W-PM			H-PK
	BW	Mean	EP	BW	Mean	EP	BW	Mean	EP	BW	Mean	EP
Intercept	44	−0.060	62.90	44	−0.038	55.35	44	0.031	69.83	44	0.025	46.72
Road density	782	0.107	64.48	820	−0.087	66.30	52	0.076	38.20	46	−0.086	31.75
Intersection density	818	−0.117	100.00	44	−0.139	23.72	252	−0.210	80.66	344	−0.171	91.73
Land use mix index	44	0.114	13.14	820	−0.003	0.00	44	0.143	29.56	44	−0.051	21.29
Catering	820	0.151	100.00	812	0.222	100.00	811	0.164	100.00	820	0.102	100.00
Cultural and recreational	455	−0.009	0.00	820	−0.120	100.00	820	−0.034	0.00	820	−0.065	100.00
Accommodations	820	−0.187	100	547	−0.089	45.86	820	0.026	0.00	364	0.113	54.62
Tourist attractions	820	0.008	0.00	445	0.162	93.67	470	0.108	82.85	440	0.150	96.23
Tourism-related retail	820	0.009	0.00	816	−0.135	100.00	820	−0.065	63.38	63	0.010	40.63
Residence	820	0.046	0.00	820	−0.011	0.00	820	−0.001	0.00	820	−0.037	0.00
Education	820	−0.012	0.00	816	0.091	100.00	183	0.091	28.83	181	0.097	55.23
Healthcare	818	−0.077	0.00	820	−0.163	100.00	820	−0.109	100.00	820	−0.131	100.00
Daily retail	83	−0.087	13.02	820	0.059	0.00	820	0.049	0.00	820	0.057	0.00
Enterprises	820	0.049	0.00	820	0.085	0.00	820	0.035	0.00	820	−0.013	0.00
Parking facilities	820	0.001	0.00	185	0.083	52.68	820	−0.004	0.00	330	0.115	63.38
Public transport stations	820	−0.062	12.65	820	−0.046	0.00	612	−0.083	67.39	820	−0.061	60.46
Bus route density	44	0.266	36.62	44	0.448	52.07	46	0.437	68.49	44	0.528	75.06

Notes. BW represents the bandwidth. EP refers to Explanatory Power, representing the percentage of grid cells in which the explanatory variable has a statistically significant coefficient (p ≤ 0.05) relative to the total sample size.

. The findings indicate that the MGWR model outperforms both OLS and GWR in terms of AICc and adjusted R². For example, during the W-AM period, the AICc of the OLS model is 2181.804, while that of the GWR model is reduced to 1969.471. The MGWR model further improves the fit with a significantly lower AICc of 1622.269. Likewise, the adjusted R² increases from 0.188 in the OLS model and 0.508 in the GWR model to 0.666 in the MGWR model. Similar patterns are observed across other time periods. Additionally, we conducted spatial autocorrelation tests on the residuals of the three models. Moran’s I decreases markedly from OLS and is not statistically significant for either GWR or MGWR, with MGWR exhibiting the smallest residual spatial autocorrelation. Therefore, the remainder of this section focuses on the results derived from the MGWR model.

4.3.1. Temporal Variations in the Impact of the Built Environment on Traffic Congestion

To further examine the temporal differences in the impact of built environment factors on traffic congestion, the regression coefficients from the MGWR model were summarized and visualized using boxplots across the four defined time periods (Figure 4). Overall, the coefficients of various types of variables exhibit noticeable fluctuations over time, indicating significant temporal heterogeneity.

Regarding road network characteristics, intersection density generally exhibits a negative correlation with traffic congestion across all time periods, indicating that areas with denser intersections tend to possess stronger traffic dispersion capacity. As shown in Figure 4 and Table 3, during the weekday evening peak, intersection density has the largest absolute regression coefficient and an explanatory power of 80.66%, highlighting its particularly prominent role in regulating traffic conditions during peak commuting hours. This may be attributed to the high concentration of commuter flows, where the flexibility offered by dense intersections becomes more effective. Intersection density has a larger optimal bandwidth in W-AM because directional, synchronized inbound commuting generates queue spillovers and shockwave propagation along corridors, enlarging the operative scale. At other times, demand is less synchronized and queues remain localized, so MGWR selects smaller bandwidths. In contrast, the effect of road network density on traffic congestion varies across time periods. During W-AM and W-PM, areas with higher road network density, tend to experience more severe congestion, suggesting a potential structural mismatch between road supply and traffic organization.

Due to the relatively balanced land use mix within the Third Ring area, the impact of this indicator is not significant during off-peak periods. Land-use mix shows a larger bandwidth only in W-OFF because dispersed non-commuting trips leverage proximity and mode substitution across a wider neighborhood–district catchment. In W-AM and W-PM, corridor commuting dominates, and in H-PK, tourism clustering dominates; capacity constraints and destination focusing keep impacts localized, yielding smaller bandwidths.

Regarding tourism-related features, catering facilities consistently exhibit a significant positive impact on traffic congestion across all four time periods, with overall high and fluctuating coefficients. This indicates that, during peak tourism seasons, dining establishments exert a strong and stable influence on urban travel patterns. Notably, the regression coefficient peaks during W-OFF, which may be attributed to tourists’ concentrated lunchtime dining behavior and their tendency to travel outside traditional commuting hours. In contrast, cultural and recreational facilities show no significant effect during typical commuting periods. Their impact is more evident during non-commuting hours, when tourist activity is higher, and their influence tends to be negative. These facilities typically involve longer stays and lower trip frequency, imposing less short-term demand on the transport system. As such, they help disperse travel demand, demonstrating a “soft clustering” effect that alleviates congestion.

Hotel accommodation exerts the strongest positive influence on traffic congestion during H-PK, with the highest coefficient. This indicates that during peak tourism periods, the concentration of tourists staying and traveling from hotels turns hotel-dense zones—such as Central Street, Harbin Station, and Harbin West Station—into major pressure points in the transportation system (Figure 5). These areas often serve as key tourist hubs during holidays and are prone to temporary vehicle stops, such as tour buses and shuttle vans, which further intensify the traffic load on adjacent road segments.

Tourist attractions also show a strong positive correlation with traffic congestion during W-OFF, W-PM, and H-PK, with explanatory power exceeding 80% in all cases. This pattern is particularly evident during non-commuting periods, when a large number of tourists travel to scenic spots via taxis, public transport, or tour groups, leading to localized traffic bottlenecks. The spatial clustering of attractions and the temporal concentration of tourist trips together contribute to uneven traffic loads in specific areas. In contrast, tourism-related retail facilities exhibit a slightly negative association with congestion during W-OFF and H-PK, with modest explanatory power. This may be attributed to the fact that shopping areas in Harbin are typically equipped with clear traffic guidance systems, dedicated parking lots, or shuttle stops.

During the study period, most primary and secondary schools had already concluded their final exams and entered the winter holiday. As a result, education did not show significant effects on traffic congestion during W-AM. In contrast, they exhibited a more notable positive influence during W-OFF, W-PM and H-PK, suggesting that weekend training sessions and enrichment classes may create localized traffic pressure. Healthcare facilities showed an explanatory power of 100% during W-OFF, W-PM, and H-PK, with consistently negative regression coefficients. This indicates that traffic management around healthcare facilities in Harbin’s central urban area is well-organized, and their relatively dispersed spatial distribution (Figure 5) prevents the formation of concentrated congestion zones. Daily retail facilities demonstrated very low explanatory power and consistently small coefficients across all time periods, implying that although these areas may attract traffic, they are not decisive contributors to congestion. Residential and enterprise facilities exhibited no significant explanatory power in any of the four time periods, with all explanatory values at 0%. One possible reason is their widespread and spatially uniform distribution across the study area (Figure 5), which limits the density variation between grid cells and thus the model’s ability to capture their local traffic impacts. Another explanation may be that commuting between residential and workplace locations often involves cross-regional travel, dispersing traffic impacts along the commuting chain, making them difficult to detect through static built environment variables at the spatial unit level.

During the tourism peak period, parking facility density exhibited a particularly significant impact on traffic congestion during W-OFF and H-PK. This may be attributed, on the one hand, clusters of parking facilities attract frequent vehicle ingress and egress, thereby increasing traffic pressure; on the other hand, insufficient supply leads to vehicles cruising for available spaces, which further exacerbates congestion. The density of public transit stations had the strongest impact during W-PM, with an explanatory power of 67.39%—the highest across all periods—and a negative regression coefficient with the largest absolute value. This indicates that areas with dense transit stations had the most substantial alleviating effect on traffic congestion during this time. The result is closely related to the surge in travel demand during the evening peak. High station density areas can effectively divert private car usage and serve as buffers that absorb traffic flow. Bus-route density strongly explains congestion across all periods, yet coefficients are uniformly positive. This does not mean transit causes congestion; rather, it reflects co-location and operational frictions along high-demand corridors. Overlapping routes and compressed headways increase dwell time and boarding and alighting near major stops; the lack of exclusive lanes introduces mixed-traffic and curb conflicts with taxis, ride-hailing, bicycles, and e-mopeds; intensified transfers for commuting and sightseeing raise curb pressure. Consequently, dense transit areas exhibit bottlenecks and higher saturation, especially around attraction belts and commercial centers during holidays.

4.3.2. Spatial Variation in the Impact of the Built Environment on Traffic Congestion

To explore the spatial mechanisms by which built environment factors influence traffic congestion, this study conducted a spatial visualization analysis of the regression coefficients for selected explanatory variables. Using ArcGIS 10.8, the spatial distribution patterns of representative variables from the MGWR model across the four key time periods were illustrated. As shown in Figure 6 below and Figure 7 later in this paper, several variables exhibit clear spatial heterogeneity, highlighting significant spatial differences in their effects.

Figure 6 shows that during W-PM, the significant influence of intersection density on traffic congestion is predominantly observed in the central urban areas, whereas parts of the southeastern edge exhibit non-significant associations. By H-PK, the spatial extent of the significant influence exhibits a slight expansion. However, the absolute values of the regression coefficients are generally lower than those during W-PM, suggesting a diminished effect in relieving traffic pressure. These findings suggest that in periods of intense and repetitive commuting flows—such as W-PM—dense intersections can significantly improve traffic efficiency and mitigate bottleneck effects.

Hotel accommodations show a negative effect on traffic congestion during W-AM in traditional hotel clusters within Nangang and Xiangfang, indicating that these areas may have relatively mature traffic management systems (Figure 6). However, during H-PK, as tourism activities intensify, the influence of hotel accommodations on congestion shifts to positive. This effect is particularly evident in tourist hotspots and emerging commercial areas such as Central Avenue, Daowai District, and Songbei New Area, where the regression coefficients increase significantly. These findings indicate that concentrated tourist lodging and travel activities during holidays exert stronger attraction and pressure on the surrounding transportation system.

According to the regression coefficient distribution of tourist attractions, their impact on traffic congestion is most pronounced during W-OFF, exhibiting a “core intensification and peripheral diffusion” pattern (Figure 6). High coefficient values are concentrated in the historical city center—such as Central Avenue and Saint Sophia Cathedral—and gradually extend toward peripheral areas like Songbei and Daowai. In H-PK, the high-value zones expand westward. In addition to the central area, attractions in peripheral districts such as Daowai, Qunli, and the West Railway Station also show relatively high coefficients, indicating a trend of congestion pressure spilling over from holiday tourism.

Parking facility density exerts a particularly strong influence on traffic congestion in the traditional commercial and tourism hubs at the junctions of Nangang, Daoli, and Daowai districts during the W-OFF and H-PK periods of the tourism peak season (Figure 7). This is largely attributed to the pronounced mismatch between parking supply and demand, coupled with complex traffic organization. In addition, negative regression coefficients observed in parts of Daoli and Nangang during W-OFF suggest that inadequate parking provision may lead to vehicles circling for spaces or parking illegally, further intensifying local congestion.

During the W-PM period, the negative impact of public transit station density on traffic congestion is primarily concentrated in the central and western parts of the main urban area, especially in the Songbei and Qunli districts, where the coefficients exhibit relatively large absolute values (Figure 7). This indicates that high-density transit stations in these areas help alleviate traffic pressure during peak commuting hours. However, during the H-PK period, the absolute values of the coefficients generally decrease across the study area, with a sustained negative relationship only observed in certain segments of the western district. This suggests that residents’ travel patterns are more dispersed during holidays, and the congestion-mitigating effect of public transit facilities is correspondingly weakened.

At W-PM, bus route density shows large positive coefficients concentrated along the southwestern arterials and within the urban core, aligning with zones of dense commercial activity and intense commuter flows (Figure 7). In these corridors, bus route density interacts with overlapping services and short stop spacing, compounding evening-peak saturation. By contrast, the eastern and northeastern sectors display mostly nonsignificant or small effects of bus route density, consistent with lower network intensity and fewer curbside conflicts. During H-PK, coefficient magnitudes rise and significant clusters become more compact: the high-value belt in the southwest extends toward major tourist destinations and adjacent residential neighborhoods, indicating that bus route density amplifies transfer demand and curb pressure under mixed traffic. Overall, the pattern points to corridor-specific operational bottlenecks—especially linked to stop spacing and curb management—rather than a uniform citywide effect.

5. Discussion and Policy Implications

Set against the backdrop of peak tourism periods, this study investigates how various built environment factors influence urban traffic congestion across different spatial and temporal scales. Using the MGWR model, the analysis reveals pronounced spatiotemporal heterogeneity in these relationships. The findings not only validate existing theoretical frameworks on the impact of the built environment on congestion but also contribute new empirical evidence specific to tourism-intensive contexts. These results offer valuable implications for tailoring urban planning and traffic management strategies in cities facing seasonal fluctuations in travel demand.

5.1. Urban Roads

Intersection density demonstrates a congestion-mitigating effect across all time periods, which aligns with previous studies suggesting that high intersection density significantly enhances traffic dispersion capacity [37]. This effect is most pronounced during the weekday evening peak, indicating that the dispersal function of intersection design exhibits temporal sensitivity. Therefore, in road network optimization, policymakers should leverage the dispersal advantages of high intersection density during weekday evening peaks. At other times, greater emphasis should be placed on signal control and the alleviation of bottleneck points at intersections to reduce localized pressure. On the other hand, road density consistently shows a relatively low level of influence. This suggests that increasing road density alone is not an effective strategy for congestion mitigation. Instead, efforts should focus on optimizing the existing road network and traffic organization—such as refining one-way street systems, improving road hierarchy functions, and enhancing connectivity between arterial and local roads—to alleviate peak-hour congestion more effectively.

5.2. Land Use

Appropriate adjustments to urban land use patterns play a vital role in alleviating traffic congestion. Land use mix shows a significant positive correlation with congestion during weekday peak periods, indicating that areas with high land-use mix tend to generate concentrated travel demand. This finding is consistent with previous studies suggesting that mixed-use zones are prone to traffic congestion [38]. Therefore, it is necessary to moderately reduce development intensity in urban core areas and strictly control the addition of large commercial complexes to avoid excessive functional clustering. In contrast, newly developed peripheral zones should integrate balanced mixed-use planning from the early stages, particularly by promoting jobs–housing proximity.

5.3. Tourism-Related Factors

Previous studies have confirmed that during peak tourism periods, the concentrated travel demand of visitors can significantly exceed the original capacity of urban road networks, rapidly intensifying traffic congestion [2]. In such contexts, tourism-related facilities often act as “amplifiers” of congestion. Some scholars have identified that accommodation land use has the most significant impact on traffic during morning peak hours in both tourism and non-tourism seasons [39]. However, this study finds that catering facilities consistently exhibit the most pronounced positive impact on congestion across all time periods, which is closely linked to Harbin’s identity as a tourist city and the highly clustered distribution of its food services. In response, it is recommended that city management agencies dynamically monitor pedestrian flows in catering hotspots during peak tourism periods and implement time-restricted pedestrian zones, vehicle diversion routes, or temporary parking areas to relieve pressure. Cultural and leisure facilities show a “soft dispersal” effect during weekday off-peak periods; thus, policies such as extending opening hours and coordinating with late-night public transport services may expand the temporal dispersal window. Scenic attractions and hotels show a sharp increase in their positive impact on congestion during weekends, with congestion hotspots extending from the old town to emerging areas such as Qunli and Songbei. Measures such as timed reservations and capped ticket sales are suggested to smooth peak demand. Additionally, park-and-ride (P + R) facilities combined with direct shuttle services should be placed at highway exits and terminal metro stations to connect major scenic areas. In the longer term, encouraging the development of new suburban winter-tourism destinations can help spatially redistribute visitor flows and reduce pressure along single corridors. These findings reflect strong spatial associations between tourism facilities and traffic congestion rather than proven causal effects; part of the relationship may stem from the co-location of tourism clusters with highly accessible areas, a common feature of urban morphology.

5.4. Daily Life-Related Factors

The results reveal a high degree of heterogeneity in the impact of daily-function-related factors on traffic congestion during peak tourism periods. Residential and enterprise densities exhibit limited spatial gradients and show no significant correlations with congestion across all four time periods, suggesting that commuting pressure is predominantly shaped by cross-district, chain-like travel flows. Therefore, new residential areas should be planned with a jobs–housing balance in mind—embedding employment opportunities while improving local services—whereas existing neighborhoods should focus on enhancing connections with public transportation and non-motorized networks. The influence of educational facilities is muted during weekday peak hours due to the winter vacation but shows a time-specific positive effect during weekends and weekday evenings. To address this, staggered scheduling is recommended for institutions in heavily congested areas to mitigate peak-time traffic surges. Healthcare facilities consistently demonstrate stable negative coefficients across most time periods, implying a traffic-alleviating effect. Future efforts should continue to focus on improving the road infrastructure surrounding medical facilities to prevent congestion rebound.

5.5. Parking and Transportation Facilities

Parking facilities exhibit a dual pattern of “central promotion and peripheral suppression” during peak commuting hours. Accordingly, total parking supply should be strictly controlled in core urban areas, accompanied by dynamic, differentiated pricing and congestion charges to discourage non-essential driving. At peripheral nodes, P + R capacity should be expanded to encourage mode shifts among commuters and tourists. Previous studies have indicated that both overly dense and overly sparse distributions of bus stops and routes can lead to “static” and “dynamic” bottlenecks, exacerbating congestion [40,41]. In this study, the congestion-mitigating effect of transit stop density is particularly pronounced in western peripheral areas. Priority should be given to moderately increasing station density and establishing temporary express routes during the snow season along major commuting and tourist corridors, thereby reinforcing public transport’s baseline function. In contrast, bus route density shows a significant positive correlation with congestion in the southwest and central zones, where tourism and commercial activities are highly concentrated. The overlapping of routes and mixed traffic modes intensifies roadway congestion. Therefore, optimizing route structure and implementing zoned dispatching management in high-density areas is essential to improving the efficiency of public transit operations.

This study provides context-sensitive and targeted policy recommendations for managing urban traffic congestion during peak tourism periods. Seasonal congestion in Harbin is largely driven by the spatial mismatch between diverse urban functions and infrastructure capacity. Based on the identified spatiotemporal heterogeneity of built environment impacts, we recommend an integrated strategy of demand management, supply-side optimization, and intelligent traffic guidance. Specifically, core urban areas should prioritize peak demand mitigation and operational efficiency improvements, while peripheral zones should focus on enhancing infrastructure quality and multimodal connectivity. Furthermore, targeted traffic diversion measures between major scenic attractions and transport hubs—such as shuttle services, time-based restrictions, or dynamic routing—can effectively alleviate localized pressure. These recommendations aim to strengthen urban transport resilience and promote a synergistic relationship between seasonal tourism development and sustainable urban growth. The proposed framework and findings may also offer valuable guidance for other tourism-intensive cities facing similar seasonal traffic challenges.

6. Conclusions

This study employed the MGWR model to investigate the spatial and temporal heterogeneity of the impact of built environment characteristics on traffic congestion. The results demonstrate that the MGWR model outperforms both OLS and GWR, with a substantial increase in R² and a significant reduction in AICc. These improvements highlight MGWR’s superior ability to capture spatial non-stationarity and heterogeneity. Overall, the MGWR results reveal notable variations in the direction and strength of the influence of various built environment variables across different time periods and spatial contexts.

The study reveals significant temporal differences in how various built environment factors influence traffic congestion. Regarding road network characteristics, intersection density exhibits the strongest congestion-mitigation effect during W-PM, indicating that high connectivity helps disperse commuting flows. In contrast, the effect of road density on traffic congestion differs in sign across time periods, indicating potential structural bottlenecks in the road network. The land use mix positively contributes to congestion during peak commuting hours but shows reduced impact during off-peak periods. Residential and employment functions display no significant effect across time periods, likely due to the influence of cross-district commuting patterns. In terms of destination facilities, food and beverage establishments consistently aggravate congestion, with the strongest effect observed during W-OFF. Cultural and recreational facilities exhibit a mild congestion-alleviating effect during non-commuting periods. Hotels and tourist attractions show a significant positive impact on congestion, especially during H-PK and W-PM. Education and healthcare facilities exert a relatively limited influence, although vacation periods and appointment systems may help ease local traffic pressure. These differences highlight the need for time- and function-sensitive traffic governance strategies. In terms of transport infrastructure, areas with high parking facility density are more prone to congestion during off-peak and holiday periods. Dense public transit stations help relieve congestion, particularly during evening peaks. In contrast, bus route density remains positively correlated with congestion across all periods, as areas with concentrated routes often experience localized saturation due to overlapping transportation activities.

Despite these insights, this study has several limitations. First, the temporal scope is restricted to a single week during the tourism peak season and this research does not explicitly model time-varying external drivers such as dynamic population distributions, meteorological conditions, and major events. Future research could extend the observation window to include non-peak periods and fuse multi-source spatiotemporal data with the MGWR framework to build a dynamic spatiotemporal interaction model. Second, although extensive POI data are used, the analysis does not account for POI usage intensity such as visit volumes and visit frequencies, which may bias the representation of actual activity levels; subsequent work could integrate dynamic mobility or check-in data. Third, data constraints mean that our variables and third-party inputs may underrepresent the built environment and introduce measurement bias—coverage gaps, uneven probe density, etc.; future work will add indicators such as building volume and public-space distribution and triangulate with multi-source mobility, official counts, and field audits.