Preventing Snow-Induced Traffic Isolation Through Data-Driven Control: Toward Resilient and Sustainable Highway Management

Abstract

This study develops a data-driven framework to prevent traffic isolation on snow-affected highways by analyzing vehicle detection system (VDS) data collected over the past decade in the Yeongdong region of the Republic of Korea. Specifically, we used hourly traffic volume and average travel speed between interchange to interchange (IC-IC) segments on days with cumulative snowfall exceeding 30 cm, enabling the identification of critical thresholds that trigger congestion and isolation under extreme snow conditions. By examining the correlation between hourly snowfall intensity, traffic volume, and travel speed, we identified critical thresholds that signal the onset of traffic congestion and isolation, where traffic congestion refers to temporary flow deterioration with average speeds falling below 40 km/h, and traffic isolation denotes and operational breakdown characterized by average travel speeds falling below 20 km/h and prolonged loss of roadway functionality. Results indicated that when snowfall intensity exceeded 2 cm per hour, traffic congestion generally emerged once hourly volumes surpassed 1500 vehicles, whereas traffic isolation became likely when volumes exceeded 2200 vehicles per hour. Building on these findings, this study proposes adaptive traffic control measures that can be proactively implemented during snowstorm conditions. The proposed framework further provides a basis for determining the optimal timing of intervention before isolation occurs, thereby preventing operational breakdowns and enhancing both the resilience and sustainability of winter highway operations.

1. Introduction

In recent years, the frequency and severity of extreme winter weather events—particularly heavy snowfall and snowstorms—have increased significantly across many regions as a result of climate change. These events have placed mounting pressure on highway transportation networks, especially in mountainous and snow-prone areas, where snow accumulation, freezing temperatures, and limited visibility frequently disrupt vehicular mobility and compromise public safety.

Extreme winter conditions can cause considerable degradation in traffic operations. Studies have shown that heavy snowfall and cold temperatures can reduce traffic volumes by 16–47%, with average reductions around 29%, while simultaneously increasing accident rates and prolonging travel times [1,2,3,4]. In particular, light vehicles are more affected in terms of speed reduction, while freight traffic may sometimes increase due to weather-induced rerouting. Moreover, adverse conditions such as snow, ice, and fog are associated with up to a 40% increase in crash frequency, with greater storm intensity and duration further elevating the risk [5,6,7].

Beyond immediate mobility and safety concerns, these events often lead to partial or complete road closures, isolating communities and obstructing emergency services and snow removal operations. Such isolation events amplify both human and economic losses and highlight vulnerabilities in existing traffic management systems [8,9]. The economic implications are also substantial, with blocked highways and elevated maintenance demands accounting for up to 20% of annual budgets in some jurisdictions [4,6].

The severity of these impacts is influenced by a range of factors. Major highways with higher traffic volumes and complex geometries (e.g., slopes and elevation changes) are particularly susceptible to disruption. Effective winter maintenance practices and real-time monitoring systems have proven beneficial in mitigating travel delays, reducing accident risk, and lowering economic losses [10,11,12]. Additionally, the compounding effect of concurrent hazards—such as floods or landslides triggered by extreme precipitation—further complicates road network resilience, especially when not accounted for in planning [9,13].

According to refs. [14,15], the protective action perspective—such as individual choices between in-vehicle sheltering and self-evacuation—has received limited scholarly attention, despite its critical influence on survival and traffic outcomes. While existing studies have predominantly concentrated on engineering and infrastructural responses, far fewer have examined these behavioral and decision-making dimensions of snow emergencies.

Several previous studies have examined the relationship between snowfall and traffic disruption. For example, ref. [16] reported that snowfall intensity significantly reduces traffic volume and increases the likelihood of congestion. Similarly, ref. [17] emphasized that snow-induced reductions in road capacity can lead to severe traffic isolation in highway networks. Ref. [18] highlighted the importance of traffic volume thresholds in determining when road closures become necessary during winter storms. In addition, refs. [19,20] demonstrated that simulation-based approaches can effectively evaluate traffic flow under adverse weather conditions. More recently, refs. [21,22] have discussed the integration of real-time data into dynamic traffic management systems.

However, despite these efforts, relatively few studies have explicitly identified critical operational thresholds or developed real-time, data-driven frameworks that can inform timely interventions to prevent highway traffic isolation during heavy snowfall events. Existing research has tended to emphasize either weather forecasting or post-event analyses of traffic disruption, which, while valuable, provide limited guidance for proactive operational decision-making. What remains underexplored is how threshold-based strategies—such as traffic volume limits, staged closures, or priority routing—can be dynamically integrated with real-time snowfall and traffic data to mitigate the risk of systemic isolation. Addressing this gap, the present study develops a threshold-based analytical model and simulation framework that evaluates traffic isolation risks under varying snow conditions. By linking operational thresholds with scenario-based simulations, the study not only enhances the understanding of snow-induced traffic breakdowns but also provides practical insights for dynamic traffic management and resilience planning in highway networks.

To address this gap, this study develops a threshold-based analytical model for predicting and preventing traffic isolation under heavy snowfall conditions. By integrating ten years of empirical traffic and meteorological data from the Yeongdong region of the Republic of Korea—a mountainous area frequently impacted by snowstorms—this research aims to identify the combinations of snowfall intensity and traffic volume that most reliably predict congestion and isolation. Although the empirical analysis focuses on two-lane mountainous highways, the methodological framework is broadly adaptable. By recalibrating the thresholds to reflect roadway capacity, terrain, and regional traffic dynamics, the proposed approach can be extended to diverse geographic and infrastructural contexts, thereby enhancing its generalizability and practical value for winter traffic management.

2. Materials and Methods

2.1. Study Area and Climatic Context

The Yeongdong region of northeastern Republic of Korea is a climatologically significant corridor located along the eastern slopes of the Taebaek Mountains. This area experiences some of the highest annual snowfall totals on the Korean Peninsula, often exceeding 800 mm, particularly in localities such as Daegwallyeong, Sokcho, and Gangneung. During peak winter months—January and February—daily snowfall frequently surpasses 30 cm, placing substantial operational strain on highway infrastructure.

The region’s snowfall is primarily driven by orographic lifting, whereby moisture-laden easterly winds from the East Sea (Sea of Korea) are forced upward over steep mountainous terrain, leading to rapid adiabatic cooling and concentrated precipitation on windward slopes. These atmospheric and topographic dynamics closely parallel those observed in other snow-intensive coastal and mountainous zones, including Japan’s Hokuriku region, the southern interior of British Columbia in Canada, and the fjord-adjacent areas of western Norway.

Highway networks in such regions are particularly vulnerable to winter disruptions, including road closures, vehicle immobilization, and delayed snow removal. As such, the Yeongdong region provides a representative testbed for investigating the intersection of complex topography, extreme winter weather, and transportation system vulnerability. Insights gained here may be transferable to similar high-risk corridors worldwide, informing more resilient and data-driven traffic control strategies.

The highway network in the study area includes key arterial routes such as the Yeongdong Expressway (between Daegwallyeong IC and Gangneung IC), the Donghae Expressway (between Hajo IC and Sokcho IC), and the Seoul–Yangyang Expressway (between Seoyangyang IC and Yangyang JC). These sections are all dual-lane roads, making them particularly vulnerable to capacity reductions and operational blockages when snowfall intensity rapidly increases. Despite periodic snow removal operations, the narrow carriageways and limited lane redundancy significantly limit the system’s ability to absorb sudden traffic volume surges or prolonged snow accumulation.

Figure 1 illustrates the geographical location and major expressway segments within the study area in northeastern Republic of Korea. The map highlights the key highway corridors analyzed in this research, including sections of the Yeongdong Expressway, Donghae Expressway, and Seoul–Yangyang Expressway, all of which are frequently impacted by heavy snowfall events.

Furthermore, the frequent occurrence of snow-induced isolation events in the region underscores the need for a data-informed approach to traffic management. Notable incidents include a seven-hour traffic isolation on the Seoul–Yangyang Expressway in March 2021, as identified through loop-detector data (VDS) showing zero average travel speed and sustained traffic density above 80% across multiple segments. Over 2000 vehicles were effectively immobilized due to sudden snowfall and the absence of preemptive control measures.

By integrating empirical analyses of snowfall intensity and traffic volume, this study investigates the Yeongdong region—a geographically and climatologically critical area that exhibits high operational vulnerability during extreme winter events. The proposed threshold-based framework enables the identification of preemptive traffic control strategies to mitigate isolation risks and ensure more sustainable highway operations. Moreover, the findings hold broader relevance for snow-prone highway corridors in other mountainous coastal regions globally, where similar orographic and meteorological dynamics elevate transportation disruption risks.

2.2. Data Sources and Collection Methods

This study draws on an integrated dataset comprising meteorological and traffic operation records collected over a ten-year period (2013–2022) from official government sources in the Republic of Korea. The primary objective of data acquisition was to systematically capture the dynamic interaction between severe snowfall conditions and real-time highway traffic behavior across the Yeongdong region—one of the most snow-prone corridors on the Korean Peninsula.

To enable robust analysis, two distinct but complementary data streams were utilized: (1) high-resolution meteorological observations, including snowfall intensity, temperature, wind speed, and humidity, and (2) vehicle operation data recorded by the national Vehicle Detection System (VDS), which tracks traffic volume and speed at kilometer-level resolution. These datasets were obtained from the Korea Meteorological Administration (KMA) and the Korea Expressway Corporation (KEC), respectively.

All records were time-synchronized and geospatially filtered to match predefined highway segments within the study area. This dual-source data framework enabled the identification of traffic isolation patterns under snowstorm conditions and supported the derivation of quantitative thresholds for proactive traffic control. Detailed descriptions of each data component and the associated filtering criteria are provided in the following subsections.

2.2.1. Meteorological Data

Hourly meteorological data were collected from the Korea Meteorological Administration (KMA) via the national weather portal (Weather.go.kr) for the period spanning 2013 to 2022. The dataset includes hourly measurements of snowfall intensity (cm), cumulative snow depth, air temperature (°C), relative humidity (%), and wind speed (m/s), recorded at automated weather stations (AWS) strategically positioned near key expressway segments in the Yeongdong region.

To ensure spatial relevance, only AWS sites located within a 5 km buffer from major highway corridors were selected. Temporal filtering was applied to include only those periods during which heavy snowfall conditions were present—specifically, instances where hourly snowfall exceeded 2 cm/h and total daily accumulation surpassed 30 cm under active snow warning advisories. These thresholds were established based on operational standards used by the Korea Road Traffic Authority for triggering enhanced snow control protocols.

In addition to supporting temporal pattern analysis of extreme weather events, the meteorological dataset was synchronized with traffic data to enable time-aligned modeling of snowfall impacts on traffic flow, isolation events, and recovery times. This integration was critical for identifying cause–effect relationships between meteorological anomalies and traffic network performance. The dataset also served as the foundation for deriving critical weather thresholds used in the predictive traffic control framework proposed in this study.

2.2.2. Traffic Operation Data (VDS)

Traffic operational data were obtained through the Vehicle Detection System (VDS), a nationwide sensor network deployed and managed by the Korea Expressway Corporation (KEC). VDS sensors are installed at approximately 1 km intervals along major expressways and are capable of detecting vehicle passage, type classification, lane usage, and speed using a combination of infrared and magnetic loop detection technologies.

The dataset used in this study comprised three key variables:

• Hourly traffic volume, defined as the total number of vehicles passing a designated control zone (CZ) in one direction per hour across both lanes;
• Average travel speed, computed from real-time vehicle speeds and aggregated at hourly intervals;
• Control zone codes, which are spatial identifiers assigned by KEC that facilitate the mapping of traffic patterns across specific highway segments.

To ensure spatial and temporal consistency with meteorological data, only VDS records from control zones within the selected study corridors—namely, Seoyangyang IC to Yangyang JC, and Daegwallyeong IC to Gangneung IC—were analyzed. These sections are known bottlenecks during heavy snowfall and thus provide a representative sample of traffic behavior under stress conditions.

Data preprocessing involved filtering out anomalous readings (e.g., implausible speeds, missing volume data) and aligning timestamps to the corresponding meteorological data at hourly resolution. This synchronization enabled the analysis of real-time traffic responses to snowfall intensity, allowing for the identification of congestion onset, propagation, and recovery phases. The integrated dataset serves as a critical foundation for defining isolation thresholds and validating the predictive traffic control framework proposed in this study.

2.2.3. Case Selection Criteria

• To ensure that the analysis focused on operationally significant snowstorm events, a systematic case selection procedure was applied to the full 10-year dataset (2013–2022). A subset of extreme winter weather cases was extracted using the following inclusion criteria:
• Official issuance of a heavy snow warning by the Korea Meteorological Administration (KMA), indicating the presence of hazardous snowfall conditions;
• Cumulative snowfall exceeding 30 cm within a 24 h period, as measured by automated weather stations (AWS) in the vicinity of the highway network;
• Continuous and uninterrupted VDS records for a minimum duration of six hours during the event, to ensure sufficient temporal resolution for traffic behavior analysis;
• At least one confirmed instance of traffic congestion or isolation, as reported by the Korea Expressway Corporation (KEC) through official logs or operational reports.

Applying these criteria, a total of seven major snowstorm events were identified and selected for in-depth analysis. These cases spanned a range of meteorological conditions and traffic volumes, thereby providing a diverse and representative sample for threshold modeling. The selected events include both daytime and nighttime occurrences, as well as differing intensities and durations of snowfall. This variability allowed for the evaluation of how specific combinations of environmental and operational variables contribute to traffic isolation, thereby strengthening the generalizability of the findings.

2.3. VDS-Based Traffic Monitoring System

The Vehicle Detection System (VDS) is a nationwide sensor-based traffic surveillance infrastructure installed along the Republic of Korea’s expressway network. It is designed to collect continuous, high-resolution traffic information, including vehicle count, speed, and lane occupancy. Operated by the Korea Expressway Corporation (KEC), the system plays a vital role in real-time traffic management, particularly under adverse weather conditions such as heavy snowfall. In this study, VDS data served as the primary source for analyzing traffic behavior and identifying congestion and isolation patterns during snowstorm events.

2.3.1. System Configuration and Sensor Capabilities

Each VDS unit is typically installed at approximately 1 km intervals along expressway corridors and utilizes a combination of non-intrusive detection technologies—including infrared, microwave, and magnetometer sensors—to capture vehicular movements with high temporal precision. These units are capable of collecting and transmitting real-time data that reflect key traffic operation parameters.

The primary metrics recorded by each sensor include the following:

• Traffic volume (vehicles/hour): The number of vehicles passing through a given detection point per hour in each travel direction;
• Average travel speed (km/h): Estimated by averaging the speeds of individual vehicles within a detection zone, typically spanning 150 to 200 m;
• Occupancy rate (%): Defined as the percentage of time a given lane is occupied by one or more vehicles, serving as a proxy for traffic density and congestion levels.

All sensor outputs are systematically organized into predefined Control Zones (CZs)—spatial segments of the expressway network bounded by interchanges (ICs) and junctions (JCs). Each CZ is assigned a unique identifier, which encodes the route number, segment index, travel direction (e.g., eastbound or westbound), and sequential location. This structured identification system facilitates precise spatiotemporal tracking of traffic patterns and supports detailed analysis of congestion propagation and isolation events across the study region.

2.3.2. Traffic Flow Classification Based on VDS Data

To quantitatively assess highway traffic performance during snowstorm events, this study adopted standardized classification thresholds based on average travel speed, as defined by the Korea Expressway Corporation (KEC). These operational thresholds enable consistent categorization of flow conditions and identification of critical transitions in traffic status. The classification scheme is as follows:

• Free flow: Speed ≥ 80 km/h;
• Moderate flow: 40–79 km/h;
• Congestion: <40 km/h;
• Severe congestion/Isolation: <20 km/h.

For the purposes of this study, a travel speed of 40 km/h was designated as the threshold for identifying congested conditions, while 20 km/h was used to define severe congestion or effective isolation. These criteria reflect the operational definitions used by KEC during winter weather response protocols and were applied to hourly average speed data obtained from each VDS unit.

2.3.3. Data Quality and Synchronization

To ensure the analytical validity of traffic–meteorology correlations, all Vehicle Detection System (VDS) records were timestamp-aligned with hourly meteorological observations from proximal automated weather stations (AWS). Only data entries containing complete hourly values for both traffic volume and average travel speed were retained for analysis. Records exhibiting temporal gaps, abrupt anomalies, or inconsistencies were identified through descriptive screening and subsequently excluded.

Anomalous values—such as zero speed with zero volume, unusually high occupancy during low volume periods, or abrupt speed fluctuations indicative of sensor drift—were flagged and removed. Additional filtering was performed to eliminate artifacts resulting from construction activities, sensor calibration errors, or communication failures in data logging systems.

Through this rigorous quality control process, the study constructed a clean and reliable dataset that accurately reflects real-time highway operational dynamics during snowstorm conditions. This refined dataset formed the empirical foundation for estimating traffic flow thresholds, detecting isolation events, and evaluating the resilience of snow-affected highway corridors.

2.4. Snowfall Intensity and Traffic Volume Classification

2.4.1. Classification of Snowfall Intensity and Associated Traffic Impacts

To identify the onset of traffic congestion and isolation during snowfall events, this study classified meteorological and traffic variables into severity levels based on established thresholds. Hourly snowfall intensity and traffic volume were selected as the two primary indicators, given their well-documented impact on winter highway performance [23,24,25].

Snowfall intensity was categorized according to hourly accumulation rates derived from automated meteorological station data. Following national standards and empirical studies, three intensity levels were defined: light (<2 cm/h), moderate (2–10 cm/h), and heavy (>10 cm/h) [23,24,26]. Prior research indicates that even light snowfall can reduce traffic volumes by up to 22% and speeds by 10% [18,27], while moderate snowfall often results in a 16–47% drop in volume and a 9–13% decline in speed [24,25]. In heavy snow conditions, reductions in volume can exceed 40%, with travel speeds dropping by up to 18% [23,26]. Each additional centimeter of snowfall per hour may decrease speed by approximately 0.4%, as shown by ref. [24].

In parallel, hourly traffic volume was classified using thresholds derived from Vehicle Detection System (VDS) data and documented cases of snow-induced congestion. A volume below 1400 vehicles per hour was typically associated with stable flow even under snowfall, while congestion frequently emerged when volume exceeded 1500 vehicles per hour. Severe isolation was most often observed when hourly volumes surpassed 2200 vehicles, aligning with the theoretical breakdown flow rates for dual-lane highways under inclement weather [18,23,28].

To characterize traffic status, this study adopted the Korea Expressway Corporation’s classification framework, where congestion is identified at average speeds below 40 km/h and isolation is flagged below 20 km/h [23,29]. These thresholds have proven effective in past analyses of storm impacts and were validated against incident reports and sensor data in this study.

Furthermore, it is worth noting that traffic behavior during snowfall is also affected by contextual factors such as time of day, day of week, and regional characteristics. For instance, traffic sensitivity to snow is highest during midday on weekdays and late afternoons on weekends [18]. The magnitude of operational disruption can vary by highway geometry and regional snowfall adaptation capacity [23,26].

By integrating meteorological classification with traffic behavior metrics, this study establishes a robust framework for detecting threshold conditions that lead to critical network degradation. These classifications serve as the analytical foundation for identifying preemptive traffic control points in subsequent sections.

Table 1. Typical reductions in traffic volume and travel speed by snowfall intensity, based on empirical studies.

Snowfall Intensity	Typical Volume Reduction	Typical Speed Reduction
Light (<2 cm/h)	11–22%	5–10%
Moderate (2–10 cm/h)	16–47%	9–13%
Heavy (>10 cm/h)	>40%	12–18%

presents the typical impacts of varying snowfall intensity on highway traffic volume and average travel speed, based on findings from previous empirical studies. As snowfall rates increase, both traffic volume and speed exhibit substantial reductions. Light snowfall (<2 cm/h) is generally associated with moderate operational impacts, whereas heavy snowfall (>10 cm/h) can lead to more than 40% reduction in traffic volume and speed drops exceeding 12%, thereby significantly increasing the risk of congestion and isolation. These thresholds serve as an analytical basis for evaluating traffic resilience under snowstorm conditions [23,24,25,26].

2.4.2. Traffic Volume Thresholds for Congestion and Isolation Under Snowfall

Hourly traffic volume was classified using operational thresholds derived from empirical observations of snow-induced congestion events and validated against prior studies. Under normal conditions, dual-lane expressways can accommodate volumes exceeding 2000 vehicles per hour without significant delay. However, during heavy snowfall, effective road capacity is reduced due to snow accumulation, decreased visibility, and changes in driver behavior [19,30]. In this study, a threshold of 1400 veh/h was associated with free-flow or slow-flow operation even under snowfall, while congestion frequently emerged when hourly volume exceeded 1500 veh/h. Severe isolation was most commonly observed when volumes surpassed 2200 veh/h, consistent with the breakdown capacity of dual-lane highways in adverse weather as described in the Korean Highway Capacity Manual and supported by findings in refs. [23,31].

These thresholds are corroborated by international studies. For instance, refs. [32,33] demonstrated that snowfall-induced capacity losses can reduce traffic volume by up to 2% per cm of snowfall, particularly on rural highways. Similarly, ref. [18] reported 11–22% reductions in traffic volume across northeastern Ohio during snow events, with midday and weekend periods experiencing the highest sensitivity. In Japanese studies, [30] observed volume reductions of up to 30% due to physical lane narrowing from accumulated snow. The resulting congestion was often not a function of sheer volume, but rather of reduced operational space and driver caution.

Moreover, ref. [19] found that saturation flow rates at signalized intersections decreased by 20–30% under snow and slush conditions, a finding echoed in ref. [27], who linked road surface conditions to substantial speed and capacity losses. These effects are magnified on mountainous routes or areas with minimal shoulder width, where snowplows and emergency vehicles further restrict usable lanes [23].

Given these findings, the classification framework used in this study—which applies 1400 veh/h for normal flow, 1500 veh/h for congestion onset, and 2200 veh/h for isolation—is empirically grounded and consistent with observed international trends. These thresholds enable early detection of operational stress and can inform preemptive traffic management decisions during snow events.

Table 2. Summary of regional and condition-specific impacts of snowfall on traffic volume and congestion thresholds based on prior studies.

Condition/Region	Typical Volume Reduction	Congestion Threshold Change
Heavy snow, urban roads	11–22%	Lowered by 20–30%
Lane narrowing (snow)	~30% volume drop	Severe congestion at lower volumes
Alberta highways	1–2% per cm snowfall	Varies by severity
U.S. Midwest	Significant, varies	Noted at lower volumes

presents observed reductions in traffic volume and corresponding changes in congestion thresholds under various snowfall-related conditions and regional contexts. The data highlight how snow intensity, road geometry (e.g., lane narrowing), and geographic factors influence the onset of congestion during winter weather events [19,23,30,31,32,33].

2.4.3. Speed-Based Operational Criteria and Transition Framework

In this study, average travel speed was employed as a critical indicator to classify operational traffic states under snowfall conditions, in conjunction with volume-based metrics. The classification thresholds were derived from the operational guidelines of the Korea Expressway Corporation (KEC), which utilizes real-time speed data collected by the Vehicle Detection System (VDS) to evaluate roadway performance.

Specifically, an hourly average speed below 40 km/h was used to denote congestion, while speeds falling below 20 km/h were identified as indicative of traffic isolation. These thresholds were consistently applied throughout the dataset and validated against incident response records, snow control measures, and field reports to ensure operational relevance and accuracy.

To systematically characterize transition dynamics, hourly travel speed was mapped in parallel with snowfall intensity and traffic volume. This enabled the construction of a structured classification matrix that captured the empirical progression from normal flow conditions to congestion and eventual isolation. Such a framework facilitates the identification of early warning thresholds and supports the development of predictive control strategies.

Importantly, the application of speed-based criteria contributes to enhancing the resilience and sustainability of highway operations under extreme weather. By enabling timely detection of deteriorating traffic conditions, this approach supports proactive management responses, minimizes disruption duration, and promotes continuity of critical transportation services during snow events—key components of a sustainable and adaptive road infrastructure system.

2.5. Analytical Framework for Determining Isolation Thresholds

This study employed an empirical analytical framework to determine critical thresholds for traffic congestion and isolation during heavy snowfall events. The framework was designed to integrate meteorological and traffic operation data in a manner that would allow for the identification of statistically significant relationships between snowfall intensity, traffic volume, and travel speed. The ultimate goal was to establish actionable threshold values that could support proactive traffic control decisions under snowstorm conditions.

The analytical process was structured into three main phases. The first phase involved data synchronization and preprocessing. Hourly snowfall records and VDS traffic data were aligned by timestamp across all selected case events. Time intervals with incomplete or anomalous data—such as sensor malfunctions or non-weather-related incidents—were excluded to ensure analytical integrity.

In the second phase, the synchronized dataset was used to classify traffic flow patterns during snowfall. Each hour of observation was assigned to one of three traffic flow states: normal operation, congestion, or isolation. These states were defined based on average travel speed values. Observations with average speeds of 40 km per hour or higher were considered normal. Speeds between 20 and 40 km per hour indicated congestion, while speeds below 20 km per hour represented isolation.

The third phase involved identifying the combinations of snowfall intensity and traffic volume that were most strongly associated with transitions between flow states. By plotting hourly traffic volume against corresponding snowfall intensity and overlaying travel speed classifications, it was possible to identify patterns and transition boundaries. Thresholds were not determined by arbitrary cutoffs but were inferred from consistent patterns across multiple events. For instance, repeated occurrences of congestion were observed when hourly snowfall exceeded 2 cm and traffic volume approached 1500 vehicles per hour. Similarly, sustained isolation tended to occur when traffic volume exceeded 2200 vehicles per hour, even at moderate snowfall intensities.

The analytical results were further validated by examining historical case studies involving known isolation events. These validations ensured that the proposed thresholds were not only statistically derived but also operationally grounded.

Overall, this data-driven framework enables the translation of historical traffic and weather interactions into a practical threshold model. The resulting control parameters can be integrated into real-time monitoring systems to support timely intervention and prevent traffic isolation during future snowstorms.

Figure 1 illustrates the flowchart of the study, which focuses on analyzing the Yeongdong Expressway, Donghae Expressway, and Seoul–Yangyang Expressway sections. The chart outlines the process of collecting and analyzing traffic and meteorological data for each section, with the goal of preventing traffic congestion and isolation during heavy snowfall events.

This study aims to establish sustainable traffic management and snow removal policies by utilizing data collected over a 10-year period, integrating snowfall data and traffic data from the VDS system. The flowchart visually represents the steps involved in identifying critical thresholds, formulating effective policies, and assessing traffic management sustainability under snowstorm conditions.

3. Results

3.1. Extraction and Analysis of Days with Snowfall Accumulating over 30 cm in 10 Years

This section focuses on identifying specific days within the last 10 years when the snowfall accumulation exceeded 30 cm on highway segments, which were critical for analyzing the impact of heavy snowfall on traffic flow. The data utilized in this study were sourced from historical meteorological records, which were systematically reviewed to pinpoint instances of significant snowfall events. These datasets provided insights into the timing, intensity, and geographical distribution of such snowstorms.

In this analysis, snowfall events exceeding 30 cm were selected based on the intensity of snow accumulation and their impact on road conditions. The occurrence of these significant snowfall events was cross-referenced with VDS (Vehicle Detection System) data to evaluate their relationship with traffic volume, travel speed, and congestion patterns. By extracting these dates, we aimed to identify key periods when traffic disruptions, including isolation, were most likely to occur.

The analysis enabled the identification of high-risk periods during winter seasons, and the data extracted formed the foundation for developing threshold-based proactive traffic management strategies, ensuring better preparedness for future snowstorm events.

Table 3. The typical reductions in traffic volume and travel speed based on different snowfall intensities. It categorizes snowfall as light, moderate, and heavy, and provides corresponding volume and speed reductions. These values are used to understand the operational impacts of snow on traffic flow.

Date	Route	Snowfall Section	Snowfall (cm)	Congestion Status
27 December 2016	Donghae	Geundeok IC–Hajodae IC	32	None
20 January 2017	Donghae	Hajodae IC–Sokcho IC	37	None
20 January 2017	Donghae	Geundeok IC–Hajodae IC	55	Congestion
1 March 2021	Seoul–Yangyang	Naecheon IC–Yangyang JC	50	Isolation
1 March 2021	Donghae	Hajodae IC–Sokcho IC	50	Congestion
24 December 2021	Seoul–Yangyang	Naecheon IC–Yangyang JC	49	None
24 December 2021	Donghae	Hajodae IC–Sokcho IC	49	None
24 December 2021	Donghae	Geundeok IC–Hajodae IC	42	None
15 January 2023	Donghae	Hajodae IC–Sokcho IC	45	None

presents a summary of the snowfall intensity, snowfall amounts, and congestion status for various dates and routes along the Donghae and Seoul–Yangyang Expressways. The data include significant snowfall events, with snowfall exceeding 30 cm, and the corresponding impact on traffic flow. These events were categorized based on the congestion status observed, including instances of congestion and isolation, as well as those without any significant disruptions. This table serves as a foundation for understanding how varying levels of snowfall intensity influence traffic operations and congestion patterns.

3.2. Traffic Volume Analysis Using VDS Between Yangyang IC and Seo–Yangyang IC

In this section, the Vehicle Detection System (VDS) data were used to analyze the traffic volume during periods of heavy snowfall, particularly focusing on snowstorm conditions where road capacity and traffic flow were most affected. The VDS data, which provide real-time traffic information such as vehicle count, speed, and lane usage, were analyzed to observe patterns of traffic congestion and isolation in response to various snowfall intensities.

Figure 2 presents a daily traffic volume distribution analysis, focusing on the traffic flow between Yangyang IC and Seo–Yangyang IC during the period from 1 to 31 March 2021. The chart visualizes the weekly patterns of congestion and highlights the critical traffic volume points over the eight-hour window. This analysis provides insights into the variations in traffic flow over different days of the week, as well as the impact of seasonal factors on congestion levels. Notably, the figure emphasizes significant peaks in traffic during specific hours, which coincide with higher congestion periods. The data reflect the trends of both typical and exceptional traffic volumes, allowing for a comprehensive understanding of daily operational challenges on this section of the highway.

Table 4. Traffic volume and travel speed data for the section from Seo–Yangyang IC to Yangyang IC during the 8 h congestion period on 1 March 2021. This table shows hourly data on accumulated snowfall, snowfall intensity, traffic volume, and travel speed, illustrating the impact of snowfall on traffic congestion and speed reduction.

1 March 2021 (8 h Congestion) Section: Seo–Yangyang IC → Yangyang IC
Time	Accumulated Snowfall (cm)	Snowfall (cm)	Traffic Volume (Vehicles)	Travel Speed (km/h)
0	0	0	345	109.55
1	0	0	259	106.99
2	0	0	221	106.69
3	0	0	212	99.54
4	0	0	227	99.43
5	0.2	0.2	258	98.28
6	0.2	0	426	96.23
7	2.5	2.3	854	95.86
8	5.2	2.7	1661	86.82
9	8.4	3.2	2252	61.7
10	12.1	3.7	2265	50.84
11	17.9	5.8	1891	35.94
12	22.4	4.5	1276	13.24
13	27.7	5.3	1026	8.12
14	34.1	6.4	1040	10.01
15	40.9	6.8	981	9.58
16	45.1	4.2	785	5.58
17	50.2	5.1	655	5.97
18	54.5	4.3	494	8.22
19	59.7	5.2	219	33.55
20	65.1	5.4	199	23.27
21	70.0	4.9	139	25.5
22	74.3	4.3	393	18.64
23	77.3	3.0	182	44.34

presents the traffic volume and travel speed data for the section from Seo–Yangyang IC to Yangyang IC during the 8 h congestion period on 1 March 2021. The data show hourly accumulated snowfall, snowfall intensity, traffic volume, and travel speed, highlighting the relationship between snow accumulation and its impact on traffic flow. This table demonstrates the pattern of traffic congestion as snowfall intensity increases and the corresponding reduction in travel speed.

Table 5. Traffic volume and travel speed data for the section from Yangyang IC to Seo–Yangyang IC (Opposite Direction) during the period with no congestion on 1 March 2021. This table provides corresponding hourly data for the opposite direction, showing how snowfall intensity affected traffic flow and travel speed without causing significant congestion.

1 March 2021 (No Congestion) Section: Yangyang IC → Seo–Yangyang IC (Opposite Direction).
Time	Accumulated Snowfall (cm)	Snowfall (cm)	Traffic Volume (Vehicles)	Travel Speed (km/h)
0	0	0	345	109.55
1	0	0	259	106.99
2	0	0	221	106.69
3	0	0	212	99.54
4	0	0	227	99.43
5	0.2	0.2	258	98.28
6	0.2	0	426	96.23
7	2.5	2.3	854	95.86
8	5.2	2.7	1661	86.82
9	8.4	3.2	2252	61.7
10	12.1	3.7	2265	50.84
11	17.9	5.8	1891	35.94
12	22.4	4.5	1276	13.24
13	27.7	5.3	1026	8.12
14	34.1	6.4	1040	10.01
15	40.9	6.8	981	9.58
16	45.1	4.2	785	5.58
17	50.2	5.1	655	5.97
18	54.5	4.3	494	8.22
19	59.7	5.2	219	33.55
20	65.1	5.4	199	23.27
21	70.0	4.9	139	25.5
22	74.3	4.3	393	18.64
23	77.3	3.0	182	44.34

presents the traffic volume and travel speed data for the section from Yangyang IC to Seo–Yangyang IC (Opposite Direction) during the period with no congestion on 1 March 2021. Similar to Table 3, this table includes hourly data on accumulated snowfall, snowfall intensity, traffic volume, and travel speed, but for the opposite direction. The data reveal how the snowfall intensity influenced traffic flow under no congestion conditions.

The analysis of traffic volume and travel speed using VDS data for the Yangyang IC to Seo–Yangyang IC section provides valuable insights into traffic flow dynamics under varying snowfall conditions. The observed relationship between snowfall intensity, traffic volume, and travel speed highlights the critical thresholds that lead to congestion and isolation. For comparison, the relationship between VDS data and vehicle speeds for additional highway segments is provided in Appendix A, offering a broader perspective on traffic behavior across different routes under similar meteorological conditions.

3.3. Classification of Traffic Flow Patterns Under Snowfall Events

Based on the analysis of synchronized traffic and meteorological datasets, three distinct patterns of traffic flow behavior were identified during heavy snowfall events. These patterns describe how highways respond operationally to varying combinations of snowfall intensity and traffic volume, and they serve as the foundation for threshold-based control strategies. The three patterns are (1) recovery without congestion, (2) temporary congestion followed by recovery, and (3) sustained congestion leading to isolation.

The first pattern, recovery without congestion, occurred when snowfall intensity was moderate and traffic volume remained below critical levels. In these cases, travel speed initially declined due to reduced visibility and slippery road conditions, but subsequently recovered as snowfall subsided and snow removal operations progressed. For example, when hourly snowfall was between 2 and 8 cm and traffic volume remained below 1400 vehicles per hour, average travel speeds typically remained above 40 km per hour or returned to that level within a few hours. This indicated that the highway system was able to maintain functional capacity without requiring intervention.

To further explore the relationship between snowfall intensity and highway traffic conditions, Figure 3 presents hourly variations in traffic volume and speed across the Yangyang IC–Seo–Yangyang IC segment on 1 March 2021—a day characterized by heavy snowfall. The left figure illustrates the temporal distribution of vehicle volume at each VDS (Vehicle Detection System) point, showing a notable peak near midday that exceeds the congestion threshold of 1400 vehicles/hour. The right figure displays the corresponding decline in traffic speed as snowfall intensifies throughout the afternoon. The plotted segments represent individual links between interchanges (ICs), enabling a localized interpretation of traffic response under worsening weather conditions. This analysis highlights the compounding effects of meteorological severity and traffic load on operational disruption.

The second pattern involved temporary congestion followed by recovery. In these cases, traffic congestion emerged due to either increased snowfall intensity or a sudden surge in traffic volume. However, the system eventually recovered, either as weather conditions improved or as control measures such as snowplowing were effectively implemented. This pattern was commonly observed when hourly snowfall was moderate and traffic volume ranged from 1400 to 1600 vehicles per hour. During these periods, travel speeds frequently dropped below 40 km per hour, indicating congestion, but eventually returned to normal levels as the event progressed.

Figure 4 illustrates the relationship between traffic volume, snowfall accumulation, and average travel speed on 1 March 2021, for the Yangyang IC to Buk–Yangyang IC segment. The left figure shows a sharp increase in traffic volume around 10 a.m., exceeding the congestion threshold of 1500 vehicles per hour, followed by a gradual decline in the afternoon. The right figure displays hourly and cumulative snowfall along with travel speed trends, where average speeds dropped significantly during midday snowfall, frequently falling below 40 km/h and occasionally nearing 20 km/h, indicating severe congestion and potential isolation. This mountainous segment exhibited a temporary disruption pattern with partial recovery as snowfall weakened and road maintenance responses likely took effect. These dynamics align with the second disruption type, characterized by recoverable congestion triggered by the interaction of moderate snowfall and high traffic volume.

The third pattern, sustained congestion leading to isolation, was the most critical and operationally disruptive. This pattern emerged under conditions of sustained snowfall and high traffic volume, which together overwhelmed the roadway’s service capacity. Once congestion formed, travel speeds dropped sharply, often falling below 20 km per hour, and failed to recover for extended periods, even after snowfall subsided. This pattern was typically associated with traffic volumes exceeding 2200 vehicles per hour, regardless of whether snowfall was light or heavy. In these cases, snow removal vehicles also faced mobility limitations due to blocked lanes, further delaying recovery and increasing the risk of prolonged isolation.

To exemplify this third and most severe disruption pattern, Figure 5 presents hourly traffic volume and speed data for the Yangyang IC–Seo-Yangyang IC segment on 1 March 2021. The left panel reveals an abrupt spike in vehicle count, exceeding 2200 vehicles/hour between 9 a.m. and 11 a.m., surpassing the defined isolation threshold. The right panel shows a simultaneous and persistent decline in travel speed, with values remaining below 20 km/h throughout the afternoon, despite tapering snowfall. This combination of high volume and sustained low speed highlights the system’s inability to recover promptly, indicating functional isolation of the road segment.

These three flow patterns were consistently observed across multiple snowstorm events and highway segments. Their identification provides a conceptual basis for understanding the nonlinear dynamics of highway traffic under snow conditions and highlights the importance of both meteorological and operational thresholds. The next sections build on these findings by examining the quantitative relationships among snowfall, traffic volume, and travel speed in more detail.

3.4. Relationship Between Snowfall Intensity, Traffic Volume, and Travel Speed

To better understand the onset of traffic congestion and isolation during snowfall events, this section examines the interrelationship between three key variables: hourly snowfall intensity (cm/h), hourly traffic volume (vehicles/h), and average travel speed (km/h). By integrating data from multiple snowstorm events and aligning them temporally, a consistent pattern of interaction between environmental and operational factors was identified.

The analysis revealed that as snowfall intensity increased, average travel speed decreased, particularly when combined with elevated traffic volumes. In cases where hourly snowfall remained below 2 cm and traffic volume was under 1400 vehicles per hour, traffic flow generally remained stable. Average speeds stayed above 50 km per hour, and no significant signs of congestion were observed. This suggests that under mild snow conditions, road users are able to adapt their driving behavior without causing systemic breakdowns.

However, as snowfall intensity reached the 2 to 8 cm per hour range—a level categorized as moderate to heavy snowfall—traffic speed began to show more pronounced reductions, especially when traffic volume exceeded 1500 vehicles per hour. Under these conditions, the average travel speed often dropped below 40 km per hour, indicating a transition into a congested state. In particular, VDS data from the Donghae Expressway on 1 March 2021 showed that a traffic volume of approximately 1525 vehicles per hour, combined with 6 to 8 cm of hourly snowfall, resulted in five hours of continuous congestion, with speeds fluctuating between 20 and 30 km per hour.

In the most critical scenarios, where traffic volume exceeded 2200 vehicles per hour—regardless of whether snowfall was light or moderate—travel speeds dropped below 20 km per hour and remained at that level for several consecutive hours. This trend was observed during the same March 2021 event on the Seoul–Yangyang Expressway, where the recorded traffic volume reached 2265 vehicles per hour during peak snowfall. Despite a decline in snowfall intensity later in the event, traffic speed did not recover, and the section experienced approximately seven hours of operational isolation. This indicates that excessive traffic load alone, when combined with reduced visibility and snow accumulation, is sufficient to paralyze highway segments.

These findings support the hypothesis that snowfall intensity and traffic volume interact in a nonlinear manner to affect travel speed. Moderate snowfall may have a limited impact when traffic volumes are low, but the same level of snowfall can become critical when traffic volumes approach the operational threshold. Conversely, even moderate traffic loads can trigger severe congestion if they coincide with heavy snowfall. The results underscore the importance of dual-variable monitoring to enable timely traffic control interventions during snowstorms.

3.5. Impact of Slope and Snowfall on Traffic Disruptions and Vehicle Mobility Restrictions

During snowstorms, heavy snowfall on highway sections with steep slopes often prevents trucks and certain large vehicles from passing through these areas. This issue arises primarily from the physical conditions of the road and adverse weather conditions, where the accumulation of snow and ice makes the slopes slippery and hinders vehicle mobility. Specifically, trucks are unable to maintain a minimum speed, and if snow removal operations are delayed, mobility becomes impossible on these sections.

Table 6. Blocked sections and slope percentages along various routes during the snowstorm event from 1–2 March 2021. The data highlight the impact of road slope on traffic disruptions, particularly in sections with steeper gradients, which contributed to vehicle mobility challenges during heavy snowfall.

Date	Route	Blocked Section	Slope (%)
1–2 March 2021	Donghae Line	120–122 K	2.9
		118–120 K	3
	Seoul–Yangyang Line	141–151 K	3
	Yeongdong Line	172–182 K	4
		164–165.5 K	5

presents the blocked sections along various routes during the snowstorm event from 1–2 March 2021, along with the corresponding slope percentages. The data highlight the specific locations where traffic disruptions occurred, with particular attention to the slope of the road, which can significantly impact the mobility of vehicles, especially during heavy snowfall. These sections represent critical areas where congestion and isolation were observed due to both snow accumulation and challenging topographic conditions.

3.6. Case Analysis of Traffic Isolation Incidents

To validate the thresholds identified in the previous analysis, this section presents a detailed examination of specific highway segments that experienced traffic isolation during severe snowfall events. These case studies illustrate how combinations of environmental and traffic conditions resulted in prolonged operational breakdowns and demonstrate the practical implications of exceeding critical thresholds.

One of the most severe cases occurred on 1 March 2021, along the Seoul–Yangyang Expressway, particularly between the Seoyangyang Interchange (IC) and Yangyang Junction (JC). During this event, snowfall began early in the morning and persisted throughout the day. According to meteorological data, hourly snowfall ranged between 4 and 8 cm from 08:00 to 14:00, with a peak accumulation of approximately 50 cm over a 12 h period. Simultaneously, traffic volume increased steadily and reached a maximum of 2265 vehicles per hour by midday.

VDS data from this segment indicated a rapid decline in travel speed as both snowfall and traffic volume intensified. At 09:00, the average speed was recorded at 48 km per hour. However, by 11:00, speeds had dropped below 30 km per hour, and by 12:00, they fell below 20 km per hour. This marked the beginning of a sustained isolation period, during which the highway segment remained effectively paralyzed. Despite a gradual decrease in snowfall intensity after 14:00, traffic flow did not recover until after 19:00, resulting in an isolation duration of nearly seven hours.

Another notable case occurred on the Donghae Expressway on the same day, particularly in the section between Hajo IC and Sokcho IC. Here, hourly snowfall rates were similar, peaking at 6 to 8 cm, but traffic volume remained relatively lower, reaching a maximum of approximately 1525 vehicles per hour. The travel speed declined to around 25 km per hour during the peak period, but began to recover within three to four hours as snowfall weakened and snow removal operations progressed. This event represents a case of temporary congestion rather than full isolation, reinforcing the importance of traffic volume as a determining factor in operational outcomes.

These cases highlight the compounding effects of high traffic volume and sustained snowfall on highway functionality. In both instances, heavy snowfall created adverse driving conditions, but only the segment with excessive traffic volume experienced extended isolation. Snowplows and emergency vehicles were also delayed due to blocked lanes, exacerbating the situation and delaying recovery.

3.7. Determination of Critical Thresholds for Proactive Traffic Control

Building upon the classification of traffic flow patterns and detailed case analyses, this section synthesizes the observed data to propose operational thresholds that can guide proactive traffic control measures during heavy snowfall events. These thresholds are intended to help traffic management authorities anticipate the transition from normal operations to congestion and ultimately to isolation, enabling timely interventions that reduce the risk of extended service disruption.

The findings indicate that two distinct thresholds in traffic volume are critical for assessing operational stability under snowfall conditions. The first threshold, associated with the onset of congestion, was identified at approximately 1500 vehicles per hour under snowfall rates exceeding 2 cm per hour. At or above this volume, travel speeds commonly dropped below 40 km per hour, signaling the beginning of sustained congestion. The second, more severe threshold occurred near 2200 to 2265 vehicles per hour. When traffic volumes reached this level, particularly under moderate to heavy snowfall, average travel speeds fell below 20 km per hour and remained suppressed for extended periods, indicating a condition of operational isolation.

Table 7. Summary of traffic volume, snowfall intensity, and congestion status under snowfall conditions. Observed data from multiple snowfall events, including traffic volume (v/h), hourly snowfall intensity (cm/h), and corresponding congestion status across major highway sections. The table identifies key transition points in traffic flow under varying weather conditions.

Date	Route	Research Section	Traffic Volume (v/h)	Snowfall (cm/h)	Congestion Status
24 December 2021	Donghae	Hajo–dae IC → Sokcho IC	≤500 vehicles	2–7.4 cm	None
24 December 2021	Donghae	Geundeok IC → Hajo–dae IC	≤1000 vehicles	2–7.4 cm	None
24 December 2021	Seoul–Yangyang	Seo–Yangyang IC → Yangyang JC	≤1100 vehicles	2–7.4 cm	None
27 December 2016	Donghae	Geundeok IC → Hajo–dae IC	≤1200 vehicles	2–3.2 cm	None
1 March 2021	Donghae	Hajo–dae IC → Sokcho IC	≤1400 vehicles	2–8 cm	None
20 January 2017	Donghae	Geundeok IC → Sokcho IC	≤800 vehicles	2–11.2 cm	Congestion
1 March 2021	Donghae	North Yangyang IC → Yangyang IC	≤1525 vehicles	2–8 cm	Congestion
1 March 2021	Seoul–Yangyang	Seo–Yangyang IC → Yangyang JC	≤2265 vehicles	2–8 cm	Isolation

summarizes representative cases across multiple events, categorizing each observation by traffic volume, snowfall intensity, and resulting traffic status. The data reveal a consistent transition pattern, with traffic systems showing a narrow buffer between manageable congestion and complete paralysis. Importantly, congestion was occasionally observed even at relatively low snowfall intensities (2–4 cm/h), when traffic volume alone approached or exceeded the upper threshold. This highlights the compounding nature of volume pressure on roadway resilience.

3.8. Simulation-Based Validation of Traffic Isolation Prevention Strategies

In this section, we conducted a simulation-based validation of the proposed traffic isolation prevention strategies using real-world traffic and snowfall data. The simulation was designed to evaluate the effectiveness of traffic management measures by analyzing traffic speed under varying snowfall intensities. The objective was to assess the ability of traffic control interventions to prevent traffic speeds from dropping below critical thresholds, such as 40 km/h, when traffic volumes exceed a specific limit.

Multiple regression models and machine learning algorithms were trained on the observed traffic data to simulate the relationship between traffic volume and speed. In particular, the logistic regression model was employed to predict the probability of traffic speed falling below 40 km/h when the traffic volume reached 2200 vehicles per hour.

The logistic regression model, trained with the data excluding outliers, successfully identified a clear threshold at 2200 vehicles per hour, above which the probability of traffic speed falling below 40 km/h significantly increased. This result validated the critical point for initiating preventive traffic management actions.

Figure 6 illustrates the regression curves, showing the relationship between traffic volume and speed for several models, including Linear, Polynomial (3rd degree), GAM/Spline, and LSBoost. These models were used to predict the expected traffic speed at varying volumes, with a focus on the 2200 vehicle threshold.

A total of 908 data points were used to generate regression curves modeling the relationship between traffic speed and volume under snowfall conditions (2 cm/h). Outliers were defined as follows:

• 0–500 range: speed ≤ 80 km/h;
• 500–1000 range: speed ≤ 60 km/h;
• 1000–1500 range: speed ≤ 40 km/h;
• 1500–2000 range: speed ≥ 80 km/h;
• 2000–2500 range: speed ≥ 60 km/h.

Outliers were excluded based on the reasoning that low speeds with low vehicle counts indicate congestion, while high speeds, despite high traffic volumes, reflect an unusual scenario. These definitions ensured that the models captured typical traffic behavior.

Figure 7 presents the probability of traffic speed falling below 40 km/h as a function of traffic volume, using a logistic regression model. The analysis shows that when traffic volume reaches 2200 vehicles per hour, there is a 26% probability that traffic speed will decrease below 40 km/h. Considering snowfall conditions with a rate exceeding 2 cm per hour, this threshold serves as the proposed point for initiating preventive traffic control measures. However, the specific vehicle volume for preemptive control may vary dynamically depending on local traffic conditions (e.g., uphill sections, entry ramps, etc.).

Table 8. Performance metrics of regression models: This table presents the traffic volume thresholds at which predicted speeds drop to 60 km/h (T_at_60) and 40 km/h (T_at_40), as well as the predicted speed at a traffic volume of 2200 vehicles per hour (Speed_at_2200) for four regression models: Linear, Polynomial, GAM, and Boost. The R² values indicate the explanatory power of each model. Although none of the models predict a speed drop below 40 km/h at 2200 vehicles, the results highlight the need for proactive traffic control strategies.

Model	T_at_60	T_at_40	Speed_at_2200	R2
Linear	1951.6	-	54.645	0.35834
Polynomial	1858.6	2447.4	48.96	0.36766
GAM	1809.2	1988.2	47.857	0.6658
Boost	1611.6	-	49.896	0.74124

presents the performance metrics for four regression models (Linear, Polynomial, GAM, and Boost) based on observed data after outlier exclusion. T_at_60 and T_at_40 represent the traffic volumes at which the predicted speed drops to 60 km/h and 40 km/h, respectively. Speed_at_2200 indicates the predicted speed at a traffic volume of 2200 vehicles per hour. While none of the models predict a speed reduction below 40 km/h at this traffic volume, the results suggest that proactive traffic control is required.

As shown in Figure 5 of Section 3.3, traffic volumes of approximately 2200 vehicles per hour typically occur around 10:00 AM. However, congestion becomes noticeable several hours later, between 12:00 PM and 5:00 PM. This delayed onset of congestion highlights the importance of early intervention in traffic management to prevent severe congestion. Therefore, initiating traffic control at the 2200-vehicle threshold before congestion becomes evident can significantly improve traffic flow and reduce the likelihood of traffic isolation under adverse weather conditions. Early intervention is critical to maintaining smooth traffic operations and minimizing the impacts of severe weather events on traffic dynamics.

It should be noted that the simulation assumes a snowfall intensity of 2 cm/h, which represents an extreme condition rather than a common scenario. This design choice was intended to validate the robustness of the proposed framework under severe stress. In typical conditions with lower snowfall intensities (e.g., 0.5–1.0 cm/h), threshold values may vary, requiring further scenario-based validation. Nevertheless, the results highlight the framework’s capacity to anticipate traffic isolation risks under heavy snowfall, thereby reinforcing its applicability to resilience-oriented winter road operation.

4. Discussion

4.1. Interpretation of Key Findings on Traffic Isolation Dynamics

The results of this study reveal important insights into the dynamic interplay between snowfall intensity, traffic volume, and operational breakdown on highway networks. One of the most critical findings is the nonlinear relationship between traffic volume and system stability during snowfall events. Unlike under normal conditions, where traffic flow deteriorates gradually with increasing demand, the presence of snow introduces sharp thresholds beyond which highway functionality collapses rapidly.

The analysis demonstrated that traffic congestion becomes likely when hourly traffic volume exceeds approximately 1500 vehicles during snowfall rates above 2 cm per hour. In this transitional zone, travel speed typically drops below 40 km per hour and remains suppressed for several hours. However, even more significant is the identification of an upper threshold near 2200 to 2265 vehicles per hour, beyond which traffic isolation consistently occurs. Under these conditions, average travel speeds fall below 20 km per hour and remain at that level despite eventual reductions in snowfall intensity. This indicates that once the system crosses a certain threshold, recovery becomes increasingly difficult, even with improved weather or active snow removal operations.

The observed dynamics reflect the compounded effects of physical road conditions and behavioral responses. As snow accumulates, drivers reduce speed due to decreased visibility, slippery surfaces, and increased braking distance. Simultaneously, high traffic density increases vehicle interaction and reduces maneuverability, exacerbating the risk of flow breakdown. Once congestion sets in, snowplows and emergency vehicles face access delays, further limiting the system’s ability to recover. These feedback loops illustrate how small changes in input conditions can result in disproportionate and prolonged disruptions.

Additionally, the results confirm that traffic isolation is not solely a function of snowfall severity. In some cases, highway segments experienced significant operational breakdowns even under moderate snowfall when traffic volume was excessive. Conversely, during heavy snowfall, isolation was avoidable when traffic volume remained well below the threshold. This reinforces the conclusion that both factors—snowfall intensity and traffic volume—must be considered simultaneously when assessing highway vulnerability and guiding operational responses. As illustrated in Appendix A Figure A1, Figure A2, Figure A3, Figure A4 and Figure A5, traffic isolation could be mitigated when traffic volume was sufficiently below the threshold, even in the presence of heavy snowfall.

Furthermore, slope conditions also play a significant role in exacerbating isolation during snowstorms. In steep sections, especially when coupled with heavy snowfall, the risk of traffic breakdowns increases significantly. Trucks and large vehicles, which require greater traction and slower speeds, often struggle to ascend steep grades, leading to further congestion and delayed snow removal operations. This highlights the importance of considering slope, in addition to snowfall intensity and traffic volume, when developing strategies to prevent operational breakdowns and ensure timely recovery during snowstorms.

Figure 8 shows the daily traffic volume on the Seoul–Yangyang Expressway for June 2025, highlighting the variations in traffic flow across different sections and hours of the day. The traffic volume exhibits a periodic pattern, with fluctuations depending on the time of day, season, and other factors. Each line represents a distinct zone, and the graph reveals the changes in traffic patterns throughout the month, with noticeable peaks on weekends.

Figure 9 illustrates the trend of highway traffic volume from 2013 to 2022, emphasizing the general increase in traffic over the years. This upward trend indicates that, as time progresses, traffic volume on highways continues to grow, reflecting the increasing demand for road infrastructure in the Republic of Korea. The periodicity observed in both the short-term (daily fluctuations) and long-term (yearly trends) is crucial for planning traffic management strategies, especially during adverse weather conditions like snowstorms. This growth highlights the need for ongoing infrastructure improvements and traffic management to accommodate rising demand.

In order to proactively manage traffic during severe snowfall events, it is crucial to integrate expected traffic volumes, known high-risk areas (such as sections with steep gradients), and snowstorm information. By combining these factors, traffic management procedures can be put in place to mitigate the risk of congestion and isolation.

The traffic management procedure follows a logical sequence designed to ensure effective response to adverse weather conditions:

• Weather Forecasting: The first step involves analyzing weather predictions, focusing on snowfall intensity, accumulation, and the timing of snowstorms. This helps in understanding the severity and timing of the impending snowstorm, which directly impacts road conditions.
• Traffic Volume Prediction: Based on historical traffic data and predictive models, the expected traffic volume is forecasted for the affected sections of the highway. The prediction also takes into account factors such as time of day, day of the week, and seasonal traffic patterns. This forecast enables the identification of potential congestion points and allows for early intervention.
• Public Awareness and Control Promotion: Once the forecast indicates that adverse conditions are likely to disrupt traffic flow, authorities issue preemptive warnings to the public. This includes advising travelers of potential disruptions, recommending alternative routes, or encouraging off-peak travel. By making drivers aware of the forecasted conditions, the aim is to reduce the volume of vehicles on the road, especially during peak congestion periods.
• Traffic Control Implementation: Finally, when the snowstorm begins and traffic volumes reach critical levels, proactive traffic control measures are implemented. This includes ramp metering, lane closures, and temporary roadblocks to regulate traffic flow. Additionally, snow removal operations are carried out to maintain road functionality and minimize delays.

The process described above ensures that all aspects of the snowstorm impact are addressed in a timely and systematic manner. By continuously monitoring the evolving conditions and adjusting strategies accordingly, it is possible to prevent or at least reduce traffic congestion and the risk of isolation.

Figure 10 illustrates the sequence of steps for implementing proactive traffic management during snowstorm events. The left diagram shows the layout of lane reduction and emergency response lanes at toll gates, with interchanges providing access control. The yellow rectangular boxes indicate traffic blockades, while the red rectangular boxes highlight congestion zones. The right diagram outlines the response sequence, from access restriction at junctions to primary and secondary interchange closures, as well as guidance for IC exits.

From a system resilience perspective, these findings suggest that winter highway networks possess a narrow adaptive capacity when exposed to concurrent environmental and operational stressors. Once traffic volume exceeds this capacity, the system transitions into a locked state that is difficult to reverse without external intervention. Recognizing and acting upon these critical points is essential for preventing isolation, minimizing economic and social disruption, and preserving the continuity of essential transport services during extreme weather.

4.2. Implications for Resilient and Sustainable Winter Road Operations

The findings of this study offer meaningful implications for enhancing the resilience and sustainability of winter road operations in snow-prone regions. In particular, the identification of empirically validated thresholds for traffic congestion and isolation under varying snowfall conditions provides a concrete basis for proactive traffic management, which is essential for maintaining system continuity during adverse weather events.

From a resilience perspective, the ability of a highway system to withstand and recover from disruption depends heavily on its operational responsiveness to environmental stressors. The results demonstrate that once traffic volume exceeds a critical threshold during snowfall, the system enters a state of instability that is difficult to reverse. This suggests that resilience is not only a matter of structural robustness but also of dynamic adaptability—specifically, the capacity to manage traffic flow in real time to prevent systemic collapse. By integrating traffic volume thresholds into operational protocols, highway authorities can initiate timely interventions that mitigate congestion escalation and preserve network functionality.

Sustainability, in the context of transportation infrastructure, involves ensuring reliable service delivery while minimizing environmental, economic, and social burdens. Traffic isolation during snowstorms imposes severe costs, including fuel wastage from idling vehicles, increased emissions due to stop-and-go conditions, delayed emergency services, and heightened risks for stranded motorists. Additionally, as the number of electric vehicles (EVs) on the road increases, the issue of battery depletion during extended periods of idling becomes a growing concern. EVs are particularly vulnerable to prolonged stop-and-go conditions, as the power consumption for heating and air conditioning accelerates battery exhaustion, further exacerbating traffic isolation and delaying recovery efforts.

Beyond individual inconvenience, widespread EV battery depletion could create systemic challenges, such as large-scale roadside assistance demand, towing operations, and secondary congestion from immobilized vehicles. These risks underscore the importance of integrating EV-specific resilience measures into winter road operations. Potential strategies include deploying mobile fast-charging support units, enabling dynamic rerouting to nearby charging facilities, and adopting adaptive traffic control that conserves vehicle energy. The proposed threshold-based framework supports this transition by facilitating data-informed decisions that minimize exposure to high-risk situations, ensuring that both conventional and electric vehicles can be effectively managed during extreme weather events.

Moreover, the framework aligns with climate adaptation strategies increasingly emphasized in national and international infrastructure policies. As climate variability intensifies the frequency and unpredictability of extreme weather events, transport systems must evolve to incorporate real-time monitoring, predictive analytics, and flexible control mechanisms. The use of VDS data and snowfall forecasting to trigger preventive actions represents a scalable approach that can be adopted not only in the Republic of Korea but also in other mountainous and coastal regions facing similar challenges.

Finally, the emphasis on proactive traffic volume control promotes a shift away from reactive snow clearance strategies toward anticipatory management. Rather than relying solely on physical removal of snow, which may be delayed due to congestion or equipment limitations, authorities can manage inflow and redistribute demand to protect critical segments of the network. This demand-side approach complements traditional snow removal efforts and contributes to the long-term sustainability of highway operations under a changing climate.

4.3. Policy and Operational Recommendations for Traffic Management Authorities

The empirical insights derived from this study support the development of targeted policy measures and operational protocols aimed at reducing the risk of traffic isolation during heavy snowfall events. By translating data-driven thresholds into actionable guidance, traffic management authorities can significantly improve their preparedness and responsiveness under adverse weather conditions.

One of the most immediate recommendations is the integration of traffic volume thresholds into existing snow emergency response frameworks. Specifically, when a heavy snow warning is issued and hourly snowfall exceeds 2 cm, real-time VDS monitoring should be activated to track traffic volume trends along vulnerable highway segments. If traffic volume approaches or exceeds the 1500 vehicles per hour threshold, early-stage interventions such as ramp metering, real-time alerts, and entry restrictions should be considered. These measures can slow the rate of inflow and stabilize traffic before congestion sets in.

If traffic volume continues to rise and reaches the upper threshold of 2200 vehicles per hour, more assertive control measures should be deployed. These may include temporary closures of specific highway sections, mandatory detours, or pre-designated snow route enforcement. In such situations, priority should be given to ensuring accessibility for snow removal equipment and emergency vehicles, which play a critical role in facilitating recovery once snowfall subsides.

In parallel, communication strategies must be enhanced to ensure that drivers receive timely and accurate information. Dynamic message signs (DMS), mobile alerts, and navigation system integration can be used to disseminate real-time updates regarding snowfall intensity, congestion risks, and control measures in effect. Clear and proactive messaging can help reduce unnecessary travel, encourage alternative route selection, and improve public compliance with emergency protocols. According to ref. [34], dynamic messaging plays a crucial role in ensuring real-time communication during critical traffic events. Additionally, ref. [35] emphasize that real-time alerts help to guide driver behavior and prevent congestion. Furthermore, ref. [36] highlights the effectiveness of navigation system integration in improving driver awareness, especially in autonomous or semi-autonomous vehicles during adverse weather conditions.

Policy development should also incorporate these threshold-based strategies into long-term resilience planning. Standard operating procedures (SOPs) for winter operations should be revised to reflect dynamic control based on real-time traffic and weather conditions. In addition, simulation-based training programs can be implemented for traffic control personnel to enhance situational awareness and decision-making under varying snowfall and traffic scenarios. Beyond training applications, future research can extend the simulation framework to incorporate specific traffic management actions—such as lane restrictions, staged closures, or priority routing—to assess their effectiveness in preventing traffic isolation during heavy snowfall.

Finally, the findings underscore the value of investing in intelligent transportation infrastructure. Expanding the VDS network, improving sensor accuracy, and integrating meteorological forecasting systems with traffic control centers can significantly strengthen the capacity of highway authorities to manage snow-induced disruptions. As climate change increases the unpredictability of extreme weather events, such adaptive capabilities will become essential to maintaining safe, reliable, and sustainable highway operations. Although derived from two-lane mountainous highways in the Yeongdong region, the threshold-based framework can be recalibrated for higher-capacity roads and different geographic contexts, enhancing its broader policy applicability.

5. Conclusions

This study developed a data-driven framework for identifying thresholds that trigger traffic isolation during heavy snowfall on highways. Results indicate that traffic isolation consistently emerged when hourly volumes exceeded approximately 2200 vehicles under moderate to heavy snowfall, while congestion was more likely once volumes surpassed 1500 vehicles per hour with snowfall rates above 2 cm/h. Crossing these thresholds led to rapid speed reductions and prolonged recovery times, regardless of subsequent weather improvements. These findings, validated through the 2021 Seoul–Yangyang Expressway case, provide practical guidance for determining timely intervention points and inform proactive traffic management strategies aimed at enhancing the resilience of winter highway operations.

The study provides concrete recommendations for proactive traffic control based on real-time VDS monitoring and snowfall alerts. These include implementing preemptive control measures—such as entry restrictions, ramp metering, or targeted closures—before critical thresholds are exceeded. Doing so can prevent service breakdowns, improve snowplow access, and support the resilience of highway operations.

In a broader context, the proposed threshold-based framework contributes to the sustainable management of winter transportation infrastructure by reducing fuel waste, emissions, and the economic cost of traffic paralysis. The approach is adaptable and can inform winter traffic policies in other snow-prone regions globally. For successful implementation beyond the Republic of Korea, however, several prerequisites should be considered. These include the availability of continuous traffic monitoring infrastructure such as a Vehicle Detection System (VDS) or an equivalent sensor network capable of capturing hourly traffic volume and speed, the presence of reliable meteorological observation systems that provide high-resolution snowfall intensity and accumulation data, and contextual adaptation to local highway design characteristics, including lane capacity, slope gradients, and snow removal practices. By integrating these elements, the framework can be transferred to diverse snow-prone environments.

Future research may extend this work by incorporating machine learning models to predict isolation risk in real time, integrating socioeconomic impacts, or evaluating multi–modal transportation responses. Additionally, expanding the analysis to include lane–specific data or accounting for road geometry and driver behavior could further improve the precision and applicability of the framework.