2. Literature Review

This chapter presents the review of the literature regarding traffic responses to weather events and the application of weather responsive traffic advisory and control strategies. The review focuses on two main areas of the literature: (1) impact of weather events on traffic system performance (i.e. supply side impacts), and (2) traveler responses to weather events and related traffic advisory information (i.e. demand side impacts). These two areas are generally unconnected in the literature. The first area developed primarily in the traffic flow theory/traffic engineering community, whereas the second has been the purview of travel behavior researchers and demand analysts. This study is among the first to combine and juxtapose these two hitherto separate domains, recognizing that behavioral responses and system performance interact in determining the manner in which weather events and weather-related information interact in determining traffic flows and associated travel times through a network.

As has been recognized since the inception of this study, the literature is considerably richer in regard to the supply-side relative to the demand-side. This is in part due to the greater relative ease of measurement and observation in the traffic arena, compared to the demand area, which often requires direct participation of the respondent in a tracking and/or interview process. This section first presents the traffic performance-related material, followed by contributions to the behavior area.

This chapter presents the review of the literature regarding traffic responses to weather events and the application of weather responsive traffic advisory and control strategies. The review focuses on two main areas of the literature: (1) impact of weather events on traffic system performance (i.e. supply side impacts), and (2) traveler responses to weather events and related traffic advisory information (i.e. demand side impacts). These two areas are generally unconnected in the literature. The first area developed primarily in the traffic flow theory/traffic engineering community, whereas the second has been the purview of travel behavior researchers and demand analysts. This study is among the first to combine and juxtapose these two hitherto separate domains, recognizing that behavioral responses and system performance interact in determining the manner in which weather events and weather-related information interact in determining traffic flows and associated travel times through a network.

As has been recognized since the inception of this study, the literature is considerably richer in regard to the supply-side relative to the demand-side. This is in part due to the greater relative ease of measurement and observation in the traffic arena, compared to the demand area, which often requires direct participation of the respondent in a tracking and/or interview process. This section first presents the traffic performance-related material, followed by contributions to the behavior area.

2.1 Traffic Performance under Weather Events

Since the early 1950's (Tanner, 1952), it has been recognized that weather conditions affect driver behavior and the manner in which a transportation system needs to be operated. By modifying speeds, headways as well as other parameters, drivers' reactions impact the overall system performance. This section presents a detailed literature review on the classification of inclement weather conditions and their translation into measurable objective parameters. The impact of such conditions on speed-flow-density relationships is first introduced. Such impact is associated with a change in capacity, delay, volume and speed, and reflects drivers' behavior on a given road section. Once the change in these parameters is better understood, the control aspect of the study is analyzed; research studies linking weather effects to signal timing, unsignalized intersections and variable message signs (VMS) are reviewed.

2.1.1 Weather Conditions

The impact of "weather conditions" on transportation systems is a general term that may pose some confusion. Researchers have used different classification schemes for weather conditions, because these conditions differ considerably in type and in magnitude (Rakha et al., 2007). Some weather conditions are extreme in nature (tornados, floods, typhoons, hurricanes etc.) and thus may trigger a different response by the drivers. Such extreme conditions are outside the immediate focus of the present study. Other inclement weather conditions (light and heavy rain, light and heavy snow etc.) offer a less compressed time frame to the decision makers, and allow drivers to retain an acceptable amount of control on their vehicles; this control may be less than under "normal everyday" situation due to physical factors such as visibility, physical discomfort (cold or hot temperatures) and reduced pavement friction with the tires when there is precipitation or icy conditions prevail.

As mentioned earlier, most existing studies do not describe all "weather conditions" in the form of measurable objective parameters, making it difficult to explain or quantify the effect of such conditions on the transportation systems and their users. Martin et al. (2000) suggested that before analyzing the impact of such conditions, four dimensions need to be considered:

1. Severity of the condition
2. Duration
3. Geographic area of influence
4. Traffic flow or the demand served by the network

According to the literature, most inclement conditions can be classified into one of three types: "rain", "snow" and "others" (wind, fog etc.). These in their turn differ in intensity (light versus heavy). In reviewing previous research efforts, Rakha et al. (2007) reported the influence of these conditions on speed and volume as summarized in Table 2-1 through Table 2-4.

Table 2-1. Rain Effects on Speed

Researcher	Ibrahim and Hall	Kyte et al.	Smith et al.
Location	Toronto, Ontario	Idaho	Hampton Roads, Virginia
Year	1994	2001	2004
Speed Reduction in Light Rain	1.9-12.9 km/hr (1.2-8 mph)	9.5 km/hr (5.9 mph)	3-5%
Speed Reduction in Heavy Rain	4.8-16.1 km/h (3-10 mph)	9.5 km/hr (5.9 mph)	3-5%
Source: Rakha et al., 2007

Table 2-2. Snow Effects on Volume

	Freeway	Arterial
Researcher	Hanbali and Kuemmel	Knapp	Maki
Location	Illinois, Minnesota, New York, Wisconsin	Iowa	Minneapolis, Minnesota
Year	1992	1995-1998	1999
Volume Reduction in Light Snow	7-31%	-	-
Volume Reduction in Heavy Snow	11-47%	16-47%	15-30%
Source: Rakha et al., 2007

Table 2-3. Snow Effects on Speed

	Freeway		Arterial
Researcher	Ibrahim and Hall	Kyte et al.	Maki	Perrin
Location	Toronto, Ontario	Idaho	Minneapolis, Minnesota	Salt Lake City, Utah
Year	1994	2001	1999	2001
Speed Reduction in Light Snow	0.97 km/hr (0.6mph)	16.4 km/hr (10.19 mph)	-	13%
Speed Reduction in Heavy Snow	37-41.8 km/hr (23-26mph)	16.4 km/hr (10.19 mph)	40%	25-30%
Source: Rakha et al., 2007

Table 2-4. Summary of Weather Impact on Macroscopic Traffic Parameters

Factors\Reduction	Volume	Maximum Observed Flow	Capacity	Speed
Rain	-	0-20%	4-47%	-
Snow	7-47%	5-10%	30%	13-40%
Wind	-	-	-	10%
Source: Rakha et al., 2007

2.1.2 Human Factors

As mentioned earlier, parameters directly linking weather conditions (visibility factor, pavement friction factor) to the driving task (perception and execution) are rarely used. Previous studies have focused mainly on two parameters, visibility and traction.

Visibility

While perceiving a stimulus to which drivers need to react, visibility plays an important role in understanding driver behavior during inclement weather conditions. A limited amount of research focused explicitly on visibility as a factor impacting traffic flow. Mostly, low visibility has been implied by the presence of heavy rain or snow conditions that reduces the sight distance of the drivers. Brilon and Ponzlet (1996) studied visibility in the context of daylight versus darkness. Based on data collected in Germany, a 13% to 47% reduction in capacity was observed in darkness compared to daylight conditions.

On the other hand, Kyte et al. (2001) explicitly defined a critical visibility distance of 0.3 km (0.18 mile), below which the speed was reduced by 0.77 km/hr (0.48 mph) for every 0.01 km (0.0062 mile) reduction in visibility. In contrast to such results, Snowden et al. (1998) found that, based on a laboratory simulation, drivers tended to underestimate their speed under more foggy environmental conditions. Accordingly, drivers subconsciously increase their speed as they get used to the surrounding environment.

Traction

During the execution of a given response by the drivers (accelerating, decelerating, steering), traction reflects the friction that exists between the tires and the pavement. Explicit friction measurement (friction coefficient) has not been associated with different weather conditions that are classified in section 2.1.1. To study the impact of snow and ice on the highway system, the Federal Highway Administration (FHWA) offered a weather classification scheme with seven categories in ascending severity ID's. These categories are closely related to the pavement conditions as well as the resulting speed reduction (Table 2-5).

It should be noted that, while capturing the weather impact on the perception (visibility) and the execution (traction) aspect of driver behavior, few researchers focused on the judgment aspect (reaction time, weight on different alternatives). Some studies (Zeitlin, 1995) suggested that weather affected drivers' ability to make quick decisions.

Table 2-5 Speed Reduction based on Pavement Conditions

Condition	Severity ID	Percent Speed Reduction
Dry	1	0%
Wet	2	0%
Wet and Snowing	3	13%
Wet and Slushy	4	22%
Slushy in wheel Paths	5	30%
Snowy and Sticking	6	35%
Snowing and Packed	7	42%
Source: FHWA, 1977

A more elaborate multi-dimensional study was introduced by Rakha et al. (2007). For the three main macroscopic parameters, namely, maximum flow rate q_c and corresponding speed u_c, and free-flow speed u_f, a weather adjustment factor (WAF) is predicted for a given precipitation type (i.e. rain or snow), intensity level, and visibility level. The prediction model is given in the following form:

where

F = WAF

i = precipitation intensity (cm/h)

v = visibility (km)

iv = interaction term

c₁, c₂, c₃₁, c₄, c₅, c₆ = model coefficients

Using data collected in the Twin Cities, Minnesota, the WAF's were plotted for both rain and snow conditions. The results are shown in Figure 2-1 and Figure 2-2.

In Figure 2-1, the vertical lines show that only rain intensity, not visibility, influences free-flow speed and speed at capacity. No significant effect was recorded for the capacity measure. Notice that under "rain conditions", only atmospheric weather parameters (no traction related parameters) affect the three macroscopic traffic parameters. This is consistent with the zero reduction in speed for wet conditions reported in Table 2-5 (Severity ID 2). On the other hand, more significant impact can be seen in "snow conditions". Capacity is reported to be independent of snow intensity (horizontal lines, Figure 2-2). As for the non-linear plots (Figure 2-2), they indicate that other parameters (mostly related to pavement conditions) may impact the free-flow speed and the speed at capacity and are not captured in the model.

Figure 2-1. Variation in WAFs as a Function of Visibility and Rain Intensity Levels
Source: Rakha et al., 2007

Figure 2-2. Variation in WAFs as a Function of Visibility and Snow Intensity Levels
Source: Rakha et al., 2007

2.1.3 Traffic Flow Characteristics

After reviewing the different classification schemes of inclement weather conditions, the impact of such conditions on traffic flow relationships is discussed in this section.

Speed-Flow-Density Relationships

The speed-flow-density relationships used in current applications do not explicitly take into consideration the effect of weather and the corresponding departure from "dry and clear conditions". A singe calibrated flow-density curve is normally used for a given location during the entire year, irrespective of the amount of rain or snow falling, the level of visibility/darkness, the pavement conditions and the temperatures. Salonen and Puttonen (1982) studied the relationship between adverse weather and safety. They found that darkness results in a reduction of operating speed by 5 km/hr. In terms of capacity, Jones and Goolsby (1969, 1970) indicated a 14% reduction during rain; no information was provided on the severity of the rain. This severity had an important impact on such reduction as reported by Keltsch and Cleveland (1971). An average of 8% reduction was reported.

Ibrahim and Hall (1994) used a dummy variable multiple regression analysis technique to test the significance in the differences in traffic conditions between different weather conditions. The data used was collected on the Queen Elizabeth Way in Missisauga, Ontario. The three measures available were speed, volume and occupancy. Detailed weather records were available from the Pearson International Airport. The weather conditions were classified under: clear, light rain, heavy rain, light snow and snow storms. The weather data used were those for the months of October, November and December 1990, and for January and February 1991. The focus was on the off-peak weekday duration (10 AM - 4 PM). Even though two functional forms were tested for the flow-occupancy relationship (linear versus quadratic), the linear model was chosen for testing the weather effects. For the speed-flow function, based on the regression analysis, the light rain caused a drop in the free-flow speed of a maximum of 2km/hr and a change of slope between -1.67 to -4.67 m/veh. At a maximum flow of 40 veh/min (2400 veh/hr), an average drop of 13 km/hr is observed compared to clear conditions. For the light snow conditions, the free-flow speed drops by 3 km/hr and the change of slope is between -1.58 and -1.92 m/veh. At the 2400 veh/hr level, the above gives an 8 km/hr drop in speed. It should be noted that, in light precipitation, even though the changes are statistically significant, the scattering of the data points makes the above conclusions difficult to apply.

For the heavy precipitation scenarios, the changes in the speed-flow-occupancy functions are more noticeable. During heavy rain, the free-flow speeds drop by 5 to 10 km/hr and the slope changes by an amount ranging between -1.67 to -4.67 m/veh. Heavy snow causes a drop of free-flow speed of 38 to 50 km/hr and the change is slope varies between -1.67and -5.08 m/veh. Near capacity (2400 veh/hr), the speeds can be reduced by more than 60 km/hr. In terms of flow-occupancy relationship, heavy rain caused a reduction in the maximum flow by 10 to 20% and heavy rains caused a reduction of 30 to 48%.

Consistently with the above studies, Rakha et al. (2007) reported no change in the functional form (linear, quadratic etc.) relating flows, speeds and densities. The authors in this study used the Van-Aerde's model (1995) calibrated using data from Baltimore, Maryland, Seattle, Washington and the Twin-Cities, Minnesota. As seen in Figure 2-3 only three main macroscopic parameters change in value (free-flow speed, flow at capacity and speed at capacity).

Parameter	Normal	Inclement	F
Free-Speed (km/h):	106	95	0.9
Capacity (veh/h):	1888	1550	0.82
Speed-at-Capacity (km/h):	90	75	0.83
Jam Density (veh/km):	100	90	0.9
Wave Speed (wj) (km/h):	-25	-24.3	0.97

Figure 2-3. Impact of Precipitation on Flow-Density-Speed Relationships
Source: Rakha et al., 2007

The constant form is observed at different intensity levels for different weather conditions (snow versus rain - See Figure 2-4).

As Ibrahim and Hall's study focused on freeway sections, it is reported that the impact of weather conditions on traffic flow relationships and parameters is different depending on the road types. Chin et al. (2004) used loop detector data from different regions of the United States; these data were linked to different weather parameters. Weather conditions were classified into 5 categories: light rain, heavy rain, light snow, heavy snow, fog and ice. The adverse weather conditions impact was translated into loss of capacity and speed and is reported in Table 2-6.

Table 2-6. Speed and Capacity Reduction based on Road Type

Weather Condition	Highway Type
	Urban Freeway		Rural Freeway		Urban Arterial		Rural Arterial
	Capacity	Speed	Capacity	Speed	Capacity	Speed	Capacity	Speed
Light Rain	4%	10%	4%	10%	6%	10%	6%	10%
Heavy Rain	8%	16%	10%	25%	6%	10%	6%	10%
Light Snow	7.50%	15%	7.50%	15%	11%	13%	11%	13%
Heavy Snow	27.50%	38%	27.50%	38%	18%	25%	18%	25%
Fog	6%	13%	6%	13%	6%	13%	6%	13%
Ice	27.50%	38%	27.50%	38%	18%	25%	18%	25%
Source: Chin et al., 2004

The more detailed impact of weather conditions on capacity, delay, volume and speed is reviewed in the following sections.

Figure 2-4. Sample Traffic Stream Model Variation
Source: Rakha et al., 2007

Capacity and Saturation Flow Rates

Adverse weather conditions can significantly reduce the operating speed and thus the capacity in a given road segment (HCM 2000). It is suggested that speeds are not influenced by the presence of wet pavement until visibility is affected (Lamm et al., 1990). Accordingly, light rain does not have noticeable impact on traffic flow compared to heavy rain (10% to 15% reduction in capacity).

Similar to rain, heavy snow is reported to have a potentially large impact on the operating speed (Ibrahim and Hall, 1994). In the corresponding study mentioned earlier, a 30% drop in capacity is attributed to heavy snow compared to a 10% reduction in the case of light snow. The main reason behind such drop is the search for a greater lateral clearance and longer headways since the lane markings are obscured by snow accumulation.

With regard to fog, the HCM noted that a modest amount of research has been performed to quantify the corresponding reduction in capacity. Other research focused on the extent of influence of different environmental conditions on capacity. These environmental conditions were categorized into daylight versus darkness, dry versus wet, and weekend versus weekday conditions; the data were collected on 15 Autobahn sites in Germany (Brilon and Ponzlet, 1995). Table 2-7 illustrates the main findings.

Table 2-7. Reduction in Capacity from Daylight and Dry Conditions

Number of Lanes	Weekday or Weekend	Dark and Dry	Daylight and Wet	Dark and Wet
Six	Weekday	13%	12%	38%
Six	Weekend	21%	27%	-
Four	Weekday	19%	18%	47%
Four	Weekend	25%	29%	-
Source: Brilon and Ponzlet, 1995

The above results recognize the effect of the reduction in light caused by the dark clouds during winter periods.

Smith et al. (2004) at the University of Virginia studied the impact of different rainfall intensity on freeway capacity and operating speeds. The corresponding traffic (volume, time mean speed and occupancy) and weather (rainfall intensity) data were collected for a one year period between August 1999 and July 2000 on two freeway links in Hampton Roads, Virginia. While the traffic data were collected every 2 minutes using the Smart Travel Laboratory, average speed and flow rates were compiled at 15-minutes intervals. As for the weather data, they were collected by the weather station at Norfolk International Airport (three miles from the study freeway segments) at an hourly rate assuming that the intensity is constant for every 15 minutes in the course of an hour. Based on the guidelines provided by the Swedish Meteorological and Hydrological Institute and the Philippine Atmospheric Services Administration, the rainfall was classified into light rain (0.01 to 0.25 inch per hour) and heavy rain (greater than 0.25 inch per hour). Plotting the speed-flow curves, the maximum throughput observed was estimated to be the capacity. The mean of the highest 5% flow rates was used to determine the change in capacity. It was concluded that light rain decreased capacity by 4 to 10% while heavy rain decreased capacity by 25 to 30%.

Another study by Prevedouros and Chang (2004) used video surveillance data monitoring freeway and arterial roadways in Honolulu between 1996 and 2000. On average, a freeway capacity reduction of 8.3% was observed.

The results of above mentioned studies, regarding the rain effects on capacity, are summarized in Table 2-8.

Table 2-8. Summary of Rain Effects on Capacity

Researcher	Ibrahim and Hall	Brilon and Ponzlet	Smith	Prevedouros and Chang
Location	Toronto, Ontario	Germany	Hampton Roads, Virginia	Honolulu, Hawaii
Year	1994	1995	2004	2004
Capacity Reduction in Light Rain	-	12-47%	4-10%	8.30%
Capacity Reduction in Heavy Rain	14-15%	12-47%	25-30%	8.30%
Source: Rakha et al., 2007

Delay

Few researchers have been able to quantify the weather impact on the delay experienced by the drivers due to the limitation of data, the inaccuracies involved in travel time estimation and number of explanatory variables involved. Stern et al., (2003) used the metropolitan Washington D.C. network (33 road segments) to collect travel time data for each weekday between December 1999 and May 2001. These data were taken in 5 minute increments between 6:30 am and 6:30 pm. The weather data were collected via Automated Surface Observation System (ASOS) stations at three International Airports in the Washington D.C. area. The travel time was regressed against weather variables for each site using a two-step linear regression process. The final variables kept in the analysis were precipitation type and intensity, wind, visibility distance and pavement conditions.

The study found an average 14% increase in travel time when weather phenomena occur. The pavement condition was the most frequent explanatory variable followed by precipitation.

Traffic Volume and Demand

Although traffic volumes reflect the demand side of the problem, it is reviewed in this section as a traffic flow parameter. Adverse weather can reduce demand when drivers cancel or postpone their activities, thus their trips. However, an increase in demand is observed if a good portion of travelers by bicycles or on foot switches to the private vehicle use (short trips). Adverse weather can also shift the peak-hour demand if the drivers choose to leave earlier or later due to unsafe driving conditions.

In 1992, the reduction in traffic volumes during snowstorms in rural areas of Illinois, Minnesota, New York, and Wisconsin (Hanbali and Kuemmel, 1992) was quantified (shown in Table 2-9). The corresponding researchers used automatic vehicle detectors data collected during the first three months of 1991. These data include annual average daily traffic and 24-hour counts. Other data collected include "highway characteristics, level of service (in terms of snow and ice removal), and road treatment. The climate data included storm data (start and end time and date), temperature range, snow depth and type of snow". Comparing hourly traffic volumes during every snowstorm to the "normal" hourly traffic volume, a volume reduction increase with total snowfall was found. However, this reduction is less important during peak-hours and during weekdays. This may be attributed to the non-discretionary type of trips (home to work and work to home trips).

Table 2-9. Volume Reduction due to Snowstorm

Snowfall	Weekdays	Weekends
< 25 mm	7-17%	19-31%
25-75 mm	11-25%	30-41%
75-150 mm	18-34%	39-47%
Source: Hanbali and Kuemmel, 1992

The winter weather impact on traffic volume and safety was studied by Knapp et al. (2000). Traffic and weather data were collected hourly along interstate highways in Iowa during the years 1995, 1996, 1997 and 1998. The goal is to focus on significant winter storms: precipitation, air temperature below freezing, wet pavement surface and a pavement temperature below freezing for at least 4 hours with an estimated snowfall exceeding 5.1 mm/hr or 0.2 inch/hr.

Covering 64 winter storm events (618 hours), the analysis showed a traffic volume reduction ranging from 16% to 47%. The average reduction was 22.3% where the 95% confidence interval is between 22.3% and 35.8%. Based on a regression analysis, the percent volume reduction had a significant relationship with total snowfall and the square of the maximum wind speed.

Speed

The weather conditions increment impact on speed was one of the first aspects of this research to be studied. In 1977, a Federal Highway Administration (FHWA) sponsored study confirmed a decrease in speeds during inclement weather.

In the Highway Capacity Manual, the reported weather impact on speeds is based on Ibrahim and Hall's (1994) study. Conducting a regression analysis on the clear weather data, a quadratic model was found to best fit the flow-occupancy relationship; a simple linear model suited the speed-flow relationship. Moreover, comparing different relationships under different weather conditions, the differences in slope and intercept of the speed-flow function during the rainy (snowy) conditions were more significant that the differences between clear and rainy weather; in light rain, a 1.9 km/hr (1.2 mph) and 6.4 to 12.9 km.hr (4 to 8 mph) reduction in operating speeds is expected during free-flow conditions and at 2400 vehicles/hr flow respectively. In heavy rain, a 4.8 to 6.4 km/hr (3 to 4 mph) reduction in speed can be expected for the free-flow conditions and a 12.9 to 16 km/hr (8 to 10 mph) reduction for the congested conditions. Finally, light snow resulted in a 0.96 km/hr (0.6 mph) drop in free-flow speeds, while heavy snow resulted in a 37.0 to 41.8 km/hr (23 to 26 mph) free-flow speed reduction.

Smith et al. (2004) concluded that although operating speed reductions were not as dramatic as was the case with capacity reductions, statistically significant reduction (3% - 5 %) in operating speed were observed under rainfall conditions compared to no rain at all.

Another related study was conducted by Kyte et al. (2001) on a rural interstate in Idaho. All data were collected from the same four-lane, level grade freeway between 1996 and 2000. High-truck volumes and low flow rates (mostly less than 500 passenger cars per hour per lane – pcphpl) were observed. Collected traffic data include time, speed and vehicle length while weather data contains visibility distance, wind speed and direction, air temperature, relative humidity, roadway surface condition, and type and amount of precipitation. The data were recorded in five-minute intervals. The main results obtained in the study are summarized in Table 2-10.

Padget et al. (2001), investigated whether drivers of SUVs, pickup trucks, and passenger cars choose different vehicle speeds during winter weather at an urban arterial street in Ames, Iowa, between November 1999 and April 2000. The results indicated that winter-weather vehicle speeds for all three vehicle types were significantly less than their normal weather speeds, and that during the day a large percentage of the speed reduction occurs after snow began to accumulate in the gutter pans of the roadway. They also found that speed variability between vehicles types increased during different winter-weather conditions and the magnitude of the speed differences between SUVs, pickup trucks and passenger cars increased with roadway snow cover, but was always less than 5.6 km/h (3.5 mph).

Table 2-10. Impact of Environment Conditions on Speed

Factor	Speed Reduction inkm/hr	Speed Reduction in Mph
Wet Pavement	9.5	5.9
Snow Covered Pavement	16.4	10.2
Wind > 24 km/hr	11.7 (high variation)	7.3 (high variation)
Visibility < 0.28 km (critical)	0.77 per 0.01 km below critical	0.48 per 33 ft below critical
Source: Kyte et al., 2001

2.1.4 Traffic Control Related Parameters

The changes in traffic related parameters and relationships mentioned above suggest a change in the control scheme applied to manage a transportation system during inclement weather conditions. This section focuses on research performed on the relationships between weather conditions and 1) signalized intersections, 2) unsignalized intersections and 3) variable message signs (including use of the latter as part of road weather management programs).

Signalized Intersections

Even though a number of research studies tried to identify the impact of inclement weather on traffic flow parameters at signalized intersections, only a limited effort considered non-standard ways to change signal timing to accommodate such impact. In 1995, Bernardin, Lochmueller and Associates measured saturation flow rates, vehicle speeds, lost time and capacity during summer, winter and severe winter weather conditions (Martin et al., 2000). Summer conditions were defined as temperatures above 14 °F and dry roads or temperatures above 32 °F and wet roads with no ice; winter conditions were defined as temperatures between -22 °F and 14 °F and dry pavement or "well-sanded hard packed snow"; extreme winter conditions were defined as temperature below -22 °F or during snowfall, blizzard, and freezing rain. It was found that summer signal timing is not suitable for winter and extreme winter signal timing. This is mainly due to the slower vehicle speeds and inaccurate measures from detectors covered by snow. Focusing on a 24-signal network in Anchorage, the SIGNAL 85 and TRANSYT-7F signal timing optimization packages were used. SIGNAL 85 determined the final phase sequences and splits based on the chosen cycle lengths and TRANSYT-7F generated offsets giving better arterial progression. The traffic flow parameters input were modified to accommodate the weather changes. Table 2-11 shows the main results. The travel time and the delay measures are provided on an average hourly basis. The results suggest that the improved timing results in a meaningful improvement in delay, accompanied by a slight increase in the percentage of stops.

Table 2-11. Improvements based on the Winter Signal Timing Modification in Anchorage

MOE	Existing Timing (Based on Summer Conditions)	Recommended Timing	Anticipated Improvement
Total Travel Time	1630 veh-hr/hr	1416 veh-hr/hr	13%
Total Delay	930 veh-hr/hr	716 veh-hr/hr	23%
Average Delay	49.8 sec/veh	38.4 sec/veh	23%
Percentage Stops	64%	68%	-6%
System Speed	17.1 mph	19.1 mph	-12%
Source: Martin et al., 2000

It should be mentioned that the study also recommended that no modifications should be made on the all-red (1-3 seconds) and amber (4-5 seconds) times during winter conditions. However, such changes, which are associated with reduced speeds, would also increase the all-red time and decrease the amber time. This topic is still a subject of disagreement between researchers.

A study by Agbolosu-Amison et al. (2004) reveals that inclement weather has a significant impact on saturation headways, particularly once slushy conditions start. The saturation flow rates were found to decrease at 15 ~ 16% under inclement weather conditions (wet & slushy, wheel path slushy and snowy & sticky). However, they concluded that start-up lost time does not appear to be significantly affected by inclement weather.

The Minnesota Department of Transportation conducted a study based on data collected on Hwy 36 between 3-8 pm on several weekdays during different weather conditions (Maki, 1999). Any storm of three inches of snow or more was defined as inclement weather. SYNCHRO III software was used to optimize signal timing during inclement weather conditions by modifying the saturation flow rates, average speeds and lost times. The output data was then compared with those corresponding to the signal timings in use and the existing normal conditions. Based on the simulated scenarios, the small improvement is illustrated in Table 2-12.

Other interesting conclusions were made based on the empirical collected data; during inclement weather conditions, a 15-20% reduction in volumes was reported during the 3-8 pm period and 15-30% reduction during the peak-hour period (5-6pm). Moreover, consistently with the previous study, the speeds decreased from 44 mph to 26 mph (~40%); the saturation flow decreased from 1800 vplph to 1600 vplph (11%); and the start-up delay increased from 2 to 3 seconds.

Table 2-12. Improvements based on the Winter Signal Timing Modification in Minnesota

Scenario	Cycle Length (sec)	Volume on TH 36 (veh/hr)	Percentile Signal Delay/Veh (sec)	Average Number of Stops/Veh	Average Speed (mph)
"Normal" Weather	160	2513	55	0.72	16
"Adverse" Weather with Existing Timing	160	1912	52	0.72	13
"Adverse Weather" with Optimized Timing	160	1912	48	0.68	13
Source: Martin et al., 2000

In 1992, Parsonson discussed signal timing in adverse weather conditions by relating it to signal timing during congested conditions. The main recommendation was to have zero offset time (setting all corridor signals to green at the same time) in snowy corridors. This "flushing" is used normally as a management scheme for some heavy congested corridors. Also in 1992, Botha and Kruse studied how the residual ice and snow impacted saturation flow rates and start-up lost times at signalized intersections. The study used data collected at Fairbanks, Alaska. The results are summarized in Table 2-13.

Table 2-13. Saturation Flow Rates Based on Botha-Kruse Study

Category	Winter	Summer	HCM	Winter/Summer Reduction	Winter/HCM Reduction
Saturation Flow Rate (vplph)	1463	1714	1800	15%	19%
Source: Botha and Kruse, 1992

As seen above, the saturation flow rates reported in this study are about 20% less than those calculated using the Highway Capacity Manual (HCM). This indicates the fact that the HCM rates are not reflective of the specific conditions prevailing on the ground.

Gilliam and Withill (1992) used SCOOT adaptive signal control system to reduce the level of congestion at signal networks and that increased due to inclement weather. Wet weather parameters were developed and served as input for the SCOOT system (decreased saturation flow rates and travel times). An important aspect of the study was the method by which a precise traffic monitoring during different weather conditions can be ensured.

One of the most comprehensive studies was performed in 2000 (Martin et al., 2000). During the winter season of 1999-2000, saturation flow rates, start-up-lost times and vehicle speeds were collected for four approaches on two signalized intersections in Salt-Lake City (Intersection 1: 400 E & 900 S; Intersection 2: 1300 E & 500 S). The data was collected during morning and evening peak-hours. The weather data was collected and categorized based on the FHWA 1977 study (Table 2-4). For the saturation flow rate and speed measures, Table 2-14 and Table 2-15 (illustrated by Figure 2-5 and Figure 2-6) show the main results.

Table 2-14. Saturation Flow Rate (vphpl)

Road Surface Condition	Severity	700 E & 900 S		1300 E & 500 S		Average Percent Reduction
		AM	PM	AM	PM
Dry	1	1881	1736	1752	1902	0
Wet	2	1680	1711	-	-	6
Wet and Snowing	3	1751	1708	1491	1691	11
Wet and Slushy	4	-	1476	1321	1647	18
Slushy in wheel Paths	5	-	1421	-	-	18
Snowy and Sticking	6	-	-	1395	-	20
Snowing and Packed	7	-	-	-	-	-
Source: Martin et al., 2000

Figure 2-5. Average Saturation Flow Reductions by Weather Condition
Source: Martin et al., 2000

Table 2-15. Speed (mph)

Road Surface Condition	Severity	700 E & 900 S		1300 E & 500 S		Average Percent Reduction
		AM	PM	AM	PM
Dry	1	39	31.4	28.4	27.4	0
Wet	2	34.3	-	-	25.2	10
Wet and Snowing	3	31.4	29.4	-	23.5	13
Wet and Slushy	4	-	22.0	-	21.8	25
Slushy in wheel Paths	5	25.5	23.4	-	-	30
Snowy and Sticking	6	-	-	-	-	-
Snowing and Packed	7	-	-	-	-	-
Source: Martin et al., 2000

Figure 2-6. Dry Speed Percentage in Inclement Weather
Source: Martin et al., 2000

As can be seen in Table 2-14, no storm was severe enough to record a severity level of 7. Moreover, no data was available for speeds beyond severity level 5 for both intersections (Table 2 15). Nonetheless, a clear reduction in both saturation flow rate and speed is recorded as the severity level increases. For the speed values, the reductions in this study are almost identical to those reported in the FHWY study in 1977 (Table 2-5 versus Table 2-15). As for saturation flow rates, since no single definition of inclement weather conditions is provided, it is difficult to compare the reductions obtained in the study with those mentioned earlier (Salt-Lake City, Utah; Fairbanks, Alaska; Anchorage, Alaska; and Minneapolis, Minnesota).

In addition to the saturation flow rates and speed reductions, Martin et al (2000) reported the following:

1. The start-up loss time increases considerably with the severity of the road conditions, mainly due to the lesser tire traction. The greatest increase is when slush accumulates on the pavement surface.
2. Most of the northern states (cold weather) do not modify their signal timings during inclement weather.
3. A special timing plan is recommended on major corridors in Salt Lake City during inclement weather based on the following:
   1. The modified plan includes new splits and offsets but the same cycle lengths unless inclement weather traffic counts are provided and require a different cycle.
   2. There should be an increase in amber time by 10% to 15% depending on the intersection size. A 10% (0.5 seconds) increase is appropriate for intersections under 50ft wide, and a 15% (1 second) increase is suitable for 100-ft wide intersections.
   3. An increase of all red time by 1 second is recommended to consider the slower clearing at permitted/protected intersections (a 0.75 seconds longer time is needed by the "sneakers").
   4. A decrease in measured (dry) saturation flow rate of 20% is needed.
   5. The average dry speeds are to be reduced by 30%.
   6. As mentioned earlier the start-up loss time should be 23% higher when devising the modified signal timing plan.

Unsignalized Intersections

Based on the literature, there appears to be no scientific literature studying the operational aspect of unsignalized intersections during inclement weather. Instead, researchers focused on the gap acceptance and the acceleration (start-up lost time) behavior during different environmental conditions (NCHRP report, 1996).

Based on observations made by Martin et al. (2000), there is an increase of 23% (from 2 to 2.46 seconds) in start-up lost time. This is based on 112 dry weather sample points and 134 snowing weather points (conditions 4-6, See Table 2 14). The decrease of an intersection's efficiency is not solely related to the increase of the start-up lost time. It is also related to the decrease in acceptable gap at unsignalized intersections as well as for permitted left-turn movements at signalized intersections. Martin et al. suggested that the critical gap for severity levels 4 through 6 increased by 25 to 30% on average. This kind of increase is closely related to an intersection width (number of lanes on each approach) and the reduced speeds while accelerating at a slower rate.

Variable Message Signs (VMS)

The common approach to managing highway operations under inclement weather has been through the provision of information to travelers, principally through variable message signs. Agencies with more progressive programs also provide speed advisory information, reflecting a control strategy that considers weather in addition to prevailing traffic conditions in setting advisory speeds.

One of the better (and early) examples of this practice in the US is the ATMS program operated by the New Jersey Turnpike Authority (NJTA) to control 148 miles (237.9 kilometers) of their heavily-travelled turnpike. The NJTA system monitors road and weather conditions, and provides speed management and traveler information to motorists accordingly. The system includes 30 environmental sensor stations (ESS) deployed along the turnpike, with pavement temperature and condition data collected at 11 sites. Over 120 Variable Speed Limit (VSL) sign assemblies are positioned along the freeway at two-mile (3.2-kilometer) intervals. Sign assemblies include VSL signs and speed warning signs, which display "REDUCE SPEED AHEAD" messages (in 5-mph increments) and the reason for speed reductions (i.e., "FOG", "SNOW", or "ICE") (Goodwin, 2003).

Other weather-related information provided to motorists tends to address immediate hazardous conditions, such as reduced visibility due to fog or restrictions due to snow. These would impact directly the traffic flow characteristics on the immediately affected section of highway. Other impacts on travel would depend on the manner in which the information is disseminated to the public at large, at their origin location.

The extent of scientific research addressing these impacts is very limited. Very few systematic studies of user responses to this type of information appear to have been conducted. In this section we focus on the more immediate impact of information on the drivers' response during inclement weather conditions. The next section will consider a wider range of traveler responses. Rämä (1999) investigated the drivers' acceptance of weather controlled signs on Finland's south coast. For that purpose, different VMS and variable speed limit signs were adopted and 590 drivers were interviewed. The objective was to assess the reactions at various intervals after the implementation of the signs in question. Although this study is not based on real-time observation, but rather on driver recall (an unreliable approach for this type of application), only a small percentage of drivers said that they modified their behavior based on the posted message or speed limit.

Consistent with the above finding, Andrey et al. (2003) reported that most drivers access weather information prior to their trip and do not change their travel patterns. As for the real-time driving pattern, Boyle and Mannering (2004) used a simulator to assess the impact of "real-time weather/incident hazard information provided by VMS and in-vehicle information system". It was found that drivers reduce their speed under adverse conditions but increase it again downstream trying to make up the lost time. Also using a driving simulator, Ganesh Babu Kolisetty et al. (2006) investigates the effect of variable message signs on driver speed behavior under foggy conditions. Focusing on an 8.5 km stretch of an expressway in Japan, the authors reported that 40% of the subjects were clearly impacted by the VMS, 40% were marginally impacted and 20% were not impacted at all.

2.2 User Responses to Weather Events and Weather-Related Information

The performance of networks depends largely on users' response to traffic conditions, which environmental conditions, such as adverse weather, can impact by increasing the variability in performance. Understanding and modeling the relationship between users' behaviors and adverse weather is important for developing strategies that target user behavior. The literature on adverse weather and user behavior has focused on the adjustments users make when faced with these conditions. Although the majority of these studies are based on stated-preference type data that may not accurately reflect users' actual behaviors, some important insights regarding the travel choice adjustments and preferences of users for weather information can be gained. The existing literature on adverse weather and trip-making behavior has examined either the propensity of travelers to change trip-making decisions, or their preferences for supplied weather information. The next section discusses studies on travel decision adjustments with respect to adverse weather, followed by discussion of users' preferences and response to weather information.

2.2.1 Adverse Weather and Travel Decision Changes

In addition to driver responses, adverse weather may also impact a host of travel decisions either made pre-trip or en-route. Most of the existing literature examines departure time, mode and route choice adjustments and show that most travelers do make some kind of change in their travel decisions under adverse weather (Khattak and de Palma, 1997; de Palma and Rochat, 1999; Aaheim and Hauge, 2005). In a detailed survey of commuters in Brussels, the results reveal that even travelers with flexible work hours have a regular schedule, and hence do not all make travel choice changes, suggesting that changing departure times, modes or routes in response to bad weather may be governed by habit or inertial effects (Khattak and de Palma, 1997). Among the commuters' whose travel decisions are impacted by weather, a relatively high percentage indicate that departure time would most likely be adjusted relative to route and mode choice (Mannering, Kim, Ng and Barfield, 1995; Khattak and de Palma, 1997; de Palma and Rochat, 1999). One possible explanation for the preference towards departure time adjustment is its relative lower costs in terms of searching for alternatives. Adjusting routes would require users to search for alternatives to the current route, while switching modes require access to alternative modes. Also, although most commuters in the Brussels study had flexible work hours, the higher propensity towards changing departure times may indicate that commuters would most likely use this flexibility in selecting the most convenient starting and stopping times, relative to other alternatives. Similar insights had been obtained by Mahmassani and co-workers in conjunction with laboratory experiments as well as travel diary surveys of commuter behavior dynamics, albeit without explicit consideration of weather (Mahmassani and Stephan, 1988; Caplice and Mahmassani, 1992; Mahmassani et al., 1997).

Mode choice has also received significant attention in regards to adverse weather. In the Brussels commuter study (Khattak and de Palma, 1997), the results show that although a high number of respondents (69%) stated they had access to a secondary mode, only a small fraction (5%) actually switched modes with bad weather, suggesting the low impact of weather patterns on mode choices. Furthermore, since only a small percentage of respondents used bikes for commuting to work, the results suggest that the substitutability between car and transit is limited. One possible explanation is that transit may expose passengers to the elements. In a study of mode choice during winter versus summer months, the authors showed that a decrease in the number of bicycle trips in the winter was accompanied by a large increase in car use for commuting purposes (Bergstrom and Magnusson, 2003). However, these studies are based on stated preference data and may not represent actual behaviors. A revealed preference study on the impact of weather on the travel habits in Bergen, Norway suggests that the impact of weather on the substitution between public and private transport is relatively small (Aaheim and Hauge, 2005). The same study also shows that travel distances decrease under precipitation, except for commute trips where there is little discretion regarding the destination.

Although the previous studies mentioned examine several different travel choices in light of adverse weather, the literature on activity scheduling changes in response to weather is virtually nonexistent. One possible explanation for this is the difficulty in obtaining good quality weather data over space and time for timeframes longer than a day. However, one study on the impact of the perception of weather information on beach trip decisions suggests that depending on the timeframe in which activities were planned, individuals make varying efforts to distort information regarding adverse weather (Adams, 1973). The study was based on interviews of individuals at a popular beach serving the Boston metropolitan region. The results indicated that respondents with a high prior commitment to go to the beach reported a lower likelihood of rain, relative to respondents with lower prior commitments, with all individuals presented the same weather forecast. Furthermore, individuals with a lower prior commitment tend to cancel the trip when given a sixty percent chance of rain. Although the study was not based on revealed behaviors, it suggests that individuals respond to weather forecasts with the commitment of the activity in mind.

Overall, the literature shows that the impact of adverse weather on trip decisions has been examined only to a limited extent, and that almost all of these studies relied on stated preference data. This suggests the difficulty in obtaining revealed travel behavior data under varying weather conditions. Furthermore, these studies seem to have focused more on travel decisions, such as departure time and mode choice, whereas the literature on activity scheduling adjustments to weather has been virtually nonexistent.

2.2.2 Adverse Weather Information and User Response

The literature on weather information provision and user response has primarily focused on two issues: (1) the impact of information provision on travel decisions; and (2) the timing of provision. The 2000 Brussels study shows that drivers using secondary sources did not change their travel choices in numbers that were statistically different relative to drivers using their own observations (Khattak and de Palma, 1997). Similarly, de Palma and Rochat (1999) show that, of the respondents who state weather is an important factor in their decisions, only 55% used secondary sources to keep up with weather forecasts, possibly suggesting a low credit given to weather forecasts. A study by Hansen et al (2001) on weather information preferences showed that information about weather-related road conditions is important to all groups including commuters, recreational travelers, and truckers. Also drivers preferred information about road surface conditions and alternative routes above travel time and speeds. Furthermore, the conditions that all groups preferred information on were those that impact vehicle performance and travel speeds, such as accumulation of snow, ice, high winds, and road closures. Although several studies on the impact of information on trip-making behavior have been conducted, few provide any indication of response to weather information specifically. In a study of en-route switching (to switch or not to switch), the choice model estimation results indicated that delays caused by bad weather decrease the propensity to switch (Polydoropoulou et al., 1996). Several studies have examined driver responses to information (Liu and Mahmassani, 1998; Peeta et al., 2000; Peeta and Yu, 2005). However, weather was not explicitly considered in these studies.

A second issue investigated in the literature is the timing of weather information. Hansen et al. (2001) found that truckers preferred information as early as possible, relative to other consumer groups. In a study on in-vehicle information provision, Mannering et al., (1995) found that, in general, the results for weather information mimicked those of other route and road conditions. For example, users driving frequently on major highways with high average commute times have decreased preference for "ahead-distance" information. Users with more flexible work hours and who change residence frequently, with high number of accidents, have higher preference for ahead-distance information. However, some differences were observed. Also, the higher the number of car passengers, the more preference for ahead-distance information. This suggests that carpool vehicles have a higher preference for weather information, possibly due to the increased variability from picking up multiple riders. Furthermore, drivers with a greater number of alternative routes have higher preference for this information. Intuitively, if individuals have little knowledge of alternative routes in a network, information on bad weather may have little impact since they do not know any alternatives.

Previous | Next

ITS Home | US DOT Home | Privacy | Feedback

United States Department of Transportation - ITS