Você está na página 1de 10

Analysis of Impact of Adverse Weather

on Freeway Free-Flow Speed in Spain


Francisco Javier Camacho, Alfredo García,
and Enrique Belda

Weather conditions affect traffic flow characteristics, including speed and representative models, these stations have to be working for a long
capacity. Several studies have stated that rain, snow, wind speed, and time (generally more than a year). During that time, they may get
visibility loss cause reductions in speed and capacity. Understanding broken or give abnormal data for several months until they are fixed.
these relationships is important to managing the traffic flow appropri- Thus, a debugging process is important in all studies in order to
ately. This paper presents new research that evaluates the free-flow speed delete incorrect data.
reduction caused by inclement weather conditions, including rain, snow, Another important issue in these studies is the representation of the
wind speed, and visibility loss. Fifteen freeway locations in northwestern weather information. To ensure that road weather and traffic condi-
Spain were selected for the study. Data were collected in 15-min intervals tions are correlated, it is necessary that both data collecting stations
by weather and traffic stations for almost 3 years, from 2006 to 2008. are close enough. If weather stations are not close enough to traffic
Individual correlations between the weather and traffic variables were stations, the weather information may not be representative.
examined to select the most important weather variables and to iden- Usually traffic data are obtained from several sensors placed on the
tify speed trends and thresholds. All climate conditions were divided road pavement, although some researchers have used other methods
into four groups: no precipitation and temperatures above 0°C, no pre- to obtain traffic data. For example, Chin et al. (1) used 127 video-
cipitation and temperatures below 0°C, rain, and snow conditions. A mul- tapes to record traffic conditions between 1996 and 2000 on a Chinese
tiple nonlinear regression analysis was performed with the final variables. freeway to study capacity.
Results showed that rain and snow both caused a reduction in speed, with Several studies divide weather conditions into different groups,
a more dramatic reduction during snow conditions. Wind speed over 8 m/s which are easier to analyze. Most of them use precipitation (either
affected traffic speed, while the effect of visibility loss presented a logarith- rain or snow), wind speed, and visibility conditions as the main vari-
mic form. It was also determined that the location caused the variables ables. It is also necessary to define a neutral climate, the one that will
to affect the speed differently, so further research should utilize a greater be compared with others to determine the effects of weather on speed
number of sites. and capacity.
Weather variables are also interrelated. Thus, rain or snow precip-
itation may also affect visibility loss, depending on the phenomenon
Adverse weather conditions affect traffic flow characteristics, espe- intensity. This is the main reason why light and heavy rain and snow
cially speed and road capacity. Understanding the relationship between are often separated in other studies.
traffic flow characteristics and weather effects is important for several Some studies divide weather phenomena into several groups, but
reasons, including providing appropriate information to drivers and in most cases these groups are based on meteorological criteria, not
road management. on traffic behavior. Thus, for rain conditions, the Highway Capac-
Several studies have been developed in order to identify how ity Manual (2) indicates that the main reason why drivers reduce
traffic flow is affected by different weather conditions, such as rain, their speed is not the wet pavement but the visibility loss due to the
snow, fog, or wet pavement. These weather conditions are variable, rain intensity. It reports a speed reduction of 2 km/h for light rain
depending on the climate examined, location, and time interval and between 5 and 7 km/h for heavy rain.
considered. Collecting snow weather data is more difficult than rain data,
Usually the method used to analyze the impact of weather con- because snow is a more infrequent phenomenon. As a result, there
ditions on traffic flow behavior is the collection of both weather are fewer studies that evaluate the impact of snow conditions on traf-
and traffic data in several locations over a very long period of time, fic behavior. In general, speed reduction in snow conditions is much
in order to get as much intensity as possible for each variable. higher than for rain. Heavy snow causes a reduction in speed between
Almost all of these studies use traffic and weather stations to col- 37 and 42 km/h, according to HCM (2).
lect data. Because of the large amount of data necessary to develop Olson et al. (3) studied the speed difference between dry and wet
pavement conditions by classifying whole days into “dry” and “wet”
Department of Transportation, Universidad Politécnica de Valencia, Camino de Vera, conditions. They reported that, at a 95% confidence level, there was
s/n. 46022, Valencia, Spain. Corresponding author: F. J. Camacho, fracator@ no difference between both conditions. Lamm et al. (4) arrived at
posgrado.upv.es. the same conclusion after studying 24 freeway segments.
Hawkins (5) studied the impact of adverse weather on speed in
Transportation Research Record: Journal of the Transportation Research Board,
No. 2169, Transportation Research Board of the National Academies, Washington,
the United Kingdom. He divided the weather conditions into nine
D.C., 2010, pp. 150–159. different groups, selecting dry pavement and total visibility as the
DOI: 10.3141/2169-16 neutral condition. He reported that the most dramatic effect over

150
Camacho, García, and Belda 151

traffic speed was caused by high wind speed (speed reduction of Cities (Minneapolis and St. Paul, Minnesota); Seattle, Washington;
almost 13 km/h) and snowy and icy conditions (speed reductions and Baltimore, Maryland. Detectors grouped data into 5-min inter-
of 30 or 40 km/h). vals. Weather variables considered were precipitation (intensities
Ibrahim and Hall (6) studied capacity, traffic volume, occupancy, divided into three categories) and visibility (divided into five cate-
and speed in 30-s intervals. Two sites were selected and data were gories). Results showed that heavy precipitation conditions induced
collected from 10 a.m. to 4 p.m. Because of the time period, only a speed decrease from 6% to 9%. Speed reduction is higher in snow
light conditions were considered to avoid using day/night as a new conditions, in accordance with previous studies. Visibility also had
variable. Weather conditions were classified as fair, light rain, heavy a larger effect in snow conditions compared with rain.
rain, light snow, heavy snow, and snowstorm. The study concluded
that light rain reduced the traffic speed very slightly (less than 2 km/h),
whereas heavy rain conditions had a more dramatic effect over speed OBJECTIVES
(5 to 10 km/h speed reduction).
Brilon and Ponzlet (7) studied the speed fluctuations of 15 German Fifteen locations from two northwestern freeways in Spain were
motorway segments. Traffic data were recorded from 1991 to 1993. studied in order to analyze the influence of adverse weather on traf-
Weather conditions were obtained from several meteorological sta- fic conditions. Weather stations and traffic stations were used to col-
tions placed between 5 and 50 km away from the corresponding lect data between January 2006 and November 2008. Traffic stations
road segment. Climate was divided into five categories: dry, wet, collected data in 1-h periods, while weather stations collected data in
wet and dry, some snow, and icy and snowy conditions. Variables 15-min periods. For comparison purposes, traffic intervals were split
such as the day of the week, placement, month, truck traffic, and light into four similar 15-min intervals.
conditions were added to the analysis. The results showed that wet To obtain more accurate models, weather conditions were divided
pavement caused a traffic speed reduction of approximately 5.9 km/h into four different climates: normal, below 0°C conditions, rainy,
on two-lane freeways. and snowy. In all of these, several variables were studied in order to
Holdener (8) analyzed the effect of rain on the speed and capacity give a nonlinear regression model, including variable interactions.
of freeways. The results showed that rain had an important impact on Fifteen traffic stations were studied, providing some informa-
speed. Wet conditions caused a speed decrease of between 0.2 and tion about variability. This variability has been examined in order
37.9 km/h. If the traffic flow was near the road capacity, the speed to estimate how the weather–traffic relationship varies depending
decrease was 13.9 km/h. on the site.
Edwards (9) analyzed the drivers’ behavior during three weather The authors also studied the possibility of using variable intensity
conditions: dry, rainy, and foggy. Traffic data were obtained from a intervals instead of continuous expressions, as other researchers have
two-lane freeway section, considering only the outside lane. Weather done. For that purpose, least significant differences (LSD) intervals
conditions were recorded by one observer, divided into five cate- have been used.
gories: sunny, cloudy, heavy rain, light rain, and foggy conditions.
Both the average and the 85th percentile speeds were investigated.
Results showed that drivers reduced their speed by about 4.8 km/h DATA COLLECTION
during heavy rain conditions.
Kyte et al. (10) measured the traffic flow speed near a crash site In this study, 15 two-lane freeway segments have been recorded
of the I-84 between 1996 and 2000 during fair weather conditions between January 2006 and November 2008 to obtain both weather
and in rain, fog, reduced visibility, and high wind conditions. and traffic data. Seven of those locations were from the A-6 freeway,
They stated that the presence of snow on the pavement reduced the covering 94.3 km, while eight stations belonged to the A-52 freeway,
passenger speed by more than 16.4 km/h. A wet pavement caused covering 57.4 km.
a 9.5 km/h reduction in passenger car speed. High wind conditions Both freeways are located in northern Spain, influenced by the
also made the drivers reduce their speed by 11.7 km/h. Poor visibility Atlantic climate. This region is the most cloudy and wettest in Spain,
reduced the speed by 21.6 km/h. with annual rainfall between 800 and 1,500 mm. Average temperatures
In 2001 Kyte et al. reported the influence of visibility, pavement are 9°C in winter and 18°C in summer.
conditions, rain, and wind speed on traffic speed. Data were col- Each one of the traffic stations was separated by an average dis-
lected during two winters between 1997 and 1999 on a road segment tance of less than 100 m from its corresponding weather station.
of the I-84 in Idaho. Results showed that light rain caused a reduc- Thus, the weather conditions registered in the weather stations were
tion in speed from 14.1 to 19.5 km/h. Heavy rain caused a 31.6 km/h assumed to be the same as those registered on the road. Traffic data
reduction in speed. were registered in 1-h intervals, while meteorological data were reg-
Smith et al. (11) published an investigation in which they studied istered in 15-min intervals. In order to analyze and correlate traffic
the impact of adverse meteorology over capacity and operating weather data, traffic data were split into 15-min aggregation periods.
speeds. Traffic data (traffic volume, median speed, and occupancy) Traffic stations registered traffic volume, average speed, and truck
were recorded in 15-min intervals between August 1999 and July percentage. Meteorological stations recorded a total of 16 weather
2000 in Virginia. Weather data were collected in 60-min intervals variables:
and had to be divided into 15-min intervals. Night data were not
used in order to avoid using light conditions as a different variable. • Snow layer depth (mm),
Rain was divided into three groups (none, light, and heavy), based • Water layer depth (mm),
on external recommendations. The results indicated that speed was • Precipitation amount for last 24 h (mm),
reduced between 3% and 5%, regardless of the rain intensity. • Wind direction (in a 360° base, north direction is 0°),
Rakha et al. (12) measured traffic speed, capacity, and impact on • Relative humidity (%),
the traffic stream behavior caused by adverse weather in the Twin • Precipitation intensity (mm/h) (rain or snow),
152 Transportation Research Record 2169

• Atmospheric pressure (bar), main climates. A regression model was developed for each of the
• Global radiation, following climates:
• Salinity,
• Air temperature (°C), • Climate 1 includes all no-rain and no-snow situations, only
• Dew temperature (°C), if air temperature is above 0°C. Weather variables considered in
• Freezing temperature (°C), this case are heavy traffic percentage, visibility, and wind speed.
• Road surface temperature (°C), It is considered as the neutral climate against which to compare
• Road subsoil temperature (°C), the rest of the climates.
• Wind speed (m/s), and • Climate 2 is a similar climate to the previous one, but con-
• Visibility index (m) from 0 (null visibility) to 2,000 (total siders only situations below 0°C.
visibility). • Climate 3 includes rain situations. Weather stations used in this
research did not distinguish between rain and snow precipitation,
New variables previously not considered by other studies have been so it considers only precipitation above 0°C. Precipitation intensity
used, such as water layer depth. It is also important to note that the is included as a variable, in addition to the variables considered in
average distance between weather stations and road measurements Climates 1 and 2.
is small compared to most previous studies. • Climate 4 includes snow situations. The difference between
this climate and Climate 3 is that in this case the air temperature is
below 0°C. It also includes all cases in which there is a snow layer
over the road pavement, regardless of the air temperature. The
DATA REDUCTION
analysis of the effect of the snow layer depth is also considered in
this climate.
Because of the large amount of data recorded, the debugging process
was very important in this research. Some sensor errors were impor-
In Table 1 data amounts used in the analysis are shown, depending
tant, affecting both weather and traffic stations. This type of error
on the climate and grouped by variable values.
appears when one sensor reports unusual data often for several
After classifying all data into the different climates, a nonlin-
months until the sensor is fixed. After examining all available
ear multiple regression analysis was performed. Due to the differ-
data, several stations were removed from the analysis, resulting
ent speed and traffic distributions depending on the location, the
in 15 locations being used for the study.
analysis was done separately for each of the locations.
The traffic flow characteristics varied greatly depending on the
Several global model equations were developed, based on the dif-
location, so the first step was to analyze the road conditions that ferent trends shown by variables in the analysis. An attempt was
would be possible to examine. The results showed that the higher made to fit interactions into the equations, but no relevant results
traffic volume recorded was 2,000 vehicles per hour (vph) in both were achieved. During the analysis phase, R2 value, mean squared
lanes, not reaching the capacity at any point and time. Due to the error (MSE), and correlations between independent values were
low traffic volumes reached, it was possible to analyze the speed considered. After several attempts, the model forms that best fit all
only in free-flow conditions, so traffic volume was not considered the locations were the following:
as an independent variable for the regression analysis. The average
speed in each site varied from 100 to almost 140 km/h. Climate 1:
Traffic volumes under 80 vph were removed from the analysis.
Only daily data were used to avoid using an additional variable. ⎛ vis ⎞
v = a + b i I t + c i log ⎜ + d i Vw (1)
Thus, data were used only from 8 a.m. to 6 p.m. ⎝ 2, 000 ⎟⎠

Climate 2:
DATA ANALYSIS
⎛ vis ⎞
v = a + b i I t + c i log ⎜ + d i Vw (2)
The data analysis was carried out in several steps, individually for ⎝ 2, 000 ⎟⎠
each one of the chosen locations. The relationship between speed
and each of the weather variables for each of the locations was exam- Climate 3:
ined first. The aim of this phase was to identify trends and thresholds
and to determine which variables were going to be used in the fol- ⎛ vis ⎞ f
v = a + b i I t + c i log ⎜ + d i Vw + I p (3)
lowing steps. Thus, variables that did not have a visible effect on ⎝ 2, 000 ⎠⎟ e
speed were not considered. Other variables, such as water layer
depth, were not considered for future analysis because they were Climate 4:
dependent on the sensor location and were highly correlated to pre-
cipitation intensity. Thus only the precipitation intensity variable ⎛ vis ⎞ f
v = a + b i I t + c i log ⎜ + d i Vw + I p + g i s (4)
⎝ 2, 000 ⎟⎠
was used for the analysis.
e
After the analysis, the following variables were considered: truck
percentage, visibility, wind speed, precipitation intensity, snow layer
where
depth, and air temperature. The influence of air temperature on speed
was not as clear as the rest of variables. It seemed to be independent, It = truck percentage (%);
but an important gap at 0°C could be detected, so it was decided to vis = visibility (m), from 0 to 2,000;
incorporate it in the analysis. Vw = wind speed (m/s);
Considering all previous values, trends, and thresholds, it was Ip = precipitation intensity (either rain or snow) (mm/h); and
decided that all weather situations could be grouped into four s = snow layer depth (mm).
Camacho, García, and Belda 153

TABLE 1 Variable Range Distribution for All Climates

Variable Value Climate 1 Climate 2 Climate 3 Climate 4 Total

Truck percentage 0–10 60,907 466 3,903 1,726 67,002


11–20 92,858 530 6,172 2,965 102,525
21–30 109,248 793 7,727 3,562 121,330
31–40 85,794 1,418 7,240 2,804 97,256
41–50 48,773 1,166 4,832 1,763 56,534
51–60 14,093 628 1,914 830 17,465
61–70 3,335 224 798 244 4,601
71–80 752 94 157 69 1,072
81–90 41 1 14 4 60
91–100 4 0 4 0 8
Traffic volume (vph) 0–200 114,009 3,299 13,903 6,128 137,339
200–400 187,617 1,299 13,069 5,837 207,822
400–600 75,688 640 4,197 1,373 81,898
600–800 23,649 60 1,016 301 25,026
800–1,000 8,708 10 357 129 9,204
1,000–1,200 3,670 7 165 92 3,934
1,200–1,400 1,618 3 46 63 1,730
≥1,400 846 2 8 44 900
Wind speed (m/s) 0–1.9 213,872 4,025 12,502 3,144 233,543
2–3.9 135,513 774 11,563 4,146 151,996
4–5.9 48,644 329 5,653 3,590 58,216
6–7.9 14,678 162 2,190 2,106 19,136
8–9.9 2,625 29 710 679 4,043
10–11.9 416 1 122 203 742
≥12 57 0 21 99 177
Visibility (m) 2,000 407,959 4,802 28,654 11,484 452,899
1,800–1,999 1,020 41 633 505 2,199
1,600–1,799 525 22 354 234 1,135
1,400–1,599 463 16 281 230 990
1,200–1,399 428 27 260 179 894
1,000–1,199 446 35 255 178 914
800–999 515 26 230 159 930
600–799 574 51 287 197 1,109
400–599 864 97 447 246 1,654
200–399 1,755 142 781 375 3,053
0–199 1,256 61 579 180 2,076
Precipitation intensity (mm/h) 0 — — 9,066 10,291 19,357
0.1–2 — — 20,748 3,338 24,086
2.1–4 — — 1,968 269 2,237
4.1–6 — — 544 55 599
6.1–8 — — 207 10 217
8.1–10 — — 105 4 109
10.1–12 — — 52 0 52
>12 — — 71 0 71
Snow layer depth (mm) 0 — — — 378 378
1–10 — — — 8,601 8,601
11–20 — — — 2,126 2,126
21–30 — — — 1,245 1,245
31–40 — — — 716 716
41–50 — — — 308 308
51–60 — — — 291 291
61–70 — — — 69 69
71–80 — — — 109 109
81–90 — — — 52 52
91–100 — — — 24 24
101–110 — — — 9 9
111–120 — — — 39 39
Total 415,805 5,320 32,761 13,967 467,853

NOTE: — = not applicable.

In Figure 1 there are graphs of all the regression coefficients. It itation coefficient shows that it has much more effect for Climate 4
can be seen that in almost all of them there is a huge variability, (snow conditions) than for Climate 3 (rain conditions). The wind speed
showing that the traffic behavior depending on adverse weather coefficient is the most variable, showing above and below 0 values,
conditions is highly variable with the location. which may indicate that speed does not depend on wind speed.
It can also be seen that, in a particular location, coefficients for dif- Some regression coefficients appear not to have a physical expla-
ferent climates usually have similar values. An analysis of the precip- nation, such as a higher traffic speed with a deeper snow layer. The
154 Transportation Research Record 2169

Constant Truck percentage


160 0
0 2 4 6 8 10 12 14 16
120 -0.1

120
Regression coefficient

Regression coefficient
-0.2
100
Climate1 -0.3 Climate1
80 Climate2
Climate2
Climate3 Climate3
60 -0.4
Climate4 Climate4

40 -0.5

20
-0.6
0
0 2 4 6 8 10 12 14 16 -0.7
Location Location

(a) (b)

Visibility Wind speed


15 3

2
10
Regression coefficient

Regression coefficient
1

5
Climate1 0 Climate1
Climate2 0 2 4 6 8 10 12 14 16 Climate2
Climate3 Climate3
Climate4 -1
0 Climate4
0 2 4 6 8 10 12 14 16
-2
-5
-3

-10 -4
Location Location
(c) (d)

Precipitation intensity Snow layer depth


35 0.2

30 0.1

0
25 0 2 4 6 8 10 12 14 16
Regression coefficient

Regression coefficient

-0.1
20
Climate1 -0.2 Climate1
15 Climate2 Climate2
Climate3 -0.3 Climate3
Climate4 Climate4
10
-0.4
5
-0.5

0 -0.6
0 2 4 6 8 10 12 14 16
-5 -0.7
Location Location
(f) (g)

FIGURE 1 Regression coefficients for all climates: (a) constant, (b) truck percentage, (c) visibility, (d) wind speed, (f ) precipitation
intensity, and (g) snow layer depth. (Note: e coefficient has not been used to avoid confusion with exponentials.)

reason for this is that these coefficients might be fitted with very less of the climate. After that, the average speed was subtracted from
few data. the individual speed and added to 120 km/h.
To determine the best regression coefficients possible, a global The regression equations used for the global analysis were of the
analysis was carried out by considering all the locations together. same form as the ones used in the individual analysis. Results
Because each one of the different traffic stations registered different showed that the regression coefficients tended to be close to the
speed values, it was previously necessary to move all the speeds to a average of the previously obtained values for individual regression.
unique reference speed (it was decided to be 120 km/h). To do so, it However, the wind speed regression coefficients did not behave cor-
was necessary to determine the average speed of every site regard- rectly. In fact, for Climate 1 the wind speed regression coefficient
Camacho, García, and Belda 155

Climate 2:
Traffic speed (km/h)

⎛ vis ⎞
v = a + b i I t + c i log ⎜ + d i Vw (6)
⎝ 2, 000 ⎟⎠

Climate 3:

⎛ vis ⎞ f
v = a + b i I t + c i log ⎜ + W i d i (Vw − 8 ) + I p (7)
⎝ 2, 000 ⎟⎠ e

Climate 4:
Wind speed (m/s)
⎛ vis ⎞ f
FIGURE 2 LSD intervals for wind speed for Climate 1. v = a + b i I t + c i log ⎜ + W i d i (Vw − 8 ) + I p + g i s (8)
⎝ 2, 000 ⎟⎠ e

was positive, indicating that as the wind speed increases, so does the where W is the dummy variable (0 when speed is below 8 m/s, 1
traffic speed. otherwise).
LSD intervals were obtained to determine if that behavior was true The regression coefficients for the developed models are in
or if the wind speed coefficient was influenced by the large amount Table 2. As mentioned previously, the model for Climate 2 is
of data of low wind speed situations. An example of that LSD analy- slightly different from the other models, but the coefficient affect-
sis is shown in Figure 2. The figure shows that before 11 m/s, the traf- ing wind speed is also called d. Table 3 shows the details of the
fic speed versus wind speed has a particular trend, while beyond that analysis of variance, including the R2 coefficients and the standard
speed, it behaves almost randomly and LSD intervals are longer, deviations for all models. The standard deviations are 5 or 6 km/h
because the amount of data at this point is much smaller. for Climates 1, 2, and 3 and 8 km/h for Climate 4.
Because of that trend, it was decided to incorporate a dummy By using the global model, it is possible to analyze both the speed
variable which is equal to 0 when wind speed is lower than 8 m/s, impact as a result of a change of global climate conditions (i.e., Cli-
and 1 when wind speed is higher. Because the traffic speed seems mate 1 to Climate 2 changes) and the impact of each of the variables
to show a decreasing trend with the increase of wind speed, that separately over the speed by examining the regression coefficients.
dummy variable was multiplied by wind speed. This was possible
for all climates except for Climate 2, which has a high wind speed
data amount much more reduced than other climates. In this case, Effect of Global Climate Conditions on Speed
the initial model was maintained.
Besides wind speed, other coefficients seemed to behave correctly, By examining the regression coefficients among all climates, it is
so other similar transformations were not carried out. The model possible to obtain general behaviors according to the climate. Thus,
equations are shown below: assuming no trucks, wind speed of 0 m/s, total visibility, small pre-
cipitation (1 mm/h), and no snow layer on the pavement, the speeds
Climate 1: provided by the models are the following:

⎛ vis ⎞
+ W i d i (Vw − 8 )
Climate 1: 129.7 km/h,
v = a + b i I t + c i log ⎜ (5)
⎝ 2, 000 ⎟⎠ Climate 2: 127.7 km/h,

TABLE 2 Multiple Regression Analysis

Climate Parameter Estimation Standard Error T-Statistic P-Value

Climate 1 a 129.72 0.0160621 8,076.15443 .000


b −0.353685 0.00055141 −641.420434 .000
c 2.54137 0.0360639 70.4685295 .000
d −0.607541 0.0551818 −11.0098076 .000
Climate 2 a 127.749 0.169144 755.267701 .000
b −0.323244 0.0040076 −80.6577503 .000
c 0.813488 0.143411 5.67242401 .000
d −0.229905 0.0376527 −6.10593663 .000
Climate 3 a 122.74 0.0967415 1,268.74196 .000
b −0.305221 0.00212833 −143.408682 .000
c 2.27213 0.0648867 35.0168833 .000
d −0.596222 0.115342 −5.16916648 .000
f 4.00669 0.102804 38.9740672 .000
Climate 4 a 116.028 0.330229 351.35618 .000
b −0.357527 0.00452176 −79.0681062 .000
c 4.60032 0.14274 32.2286675 .000
d −1.08099 0.109376 −9.88324678 .000
f 12.8298 0.313385 40.9394196 .000
g −0.133796 0.00414638 −32.2681472 .000
156 Transportation Research Record 2169

TABLE 3 Analysis of Variance Results

Climate Source Sum of Squares DF Mean Square F-Ratio P-Value R2 (%) Standard Error

Climate 1 Model 9,660,760 3 3,220,253.33 134,771.9 0 56.92 4.88


Residual 9,933,350 415,724 23.894098
Total (corr.) 19,594,110 415,727
Climate 2 Model 143,390 3 47,796.6667 2,186.9 0 58.00 4.67
Residual 116,205 5,317 21.8553696
Total (corr.) 259,595 5,320
Climate 3 Model 755,326 4 188,831.5 5,516.3 0 49.01 5.85
Residual 1,121,320 32,757 34.231462
Total (corr.) 1,876,646 32,761
Climate 4 Model 752,026 5 150,405.2 2,349.0 0 45.78 8.00
Residual 893,963 13,962 64.0282911
Total (corr.) 1,645,989 13,967

NOTE: DF = degrees of freedom, corr. = corrected.

Climate 3: 124.2 km/h, and In Climates 1 and 3, the traffic speed decreases 0.6 km/h per each
Climate 4: 120.7 km/h. meter per second of wind speed increase, so the effect is important.
This effect is higher (1 km/h) in Climate 4. In Climate 2, the effect
According to those results, in general, in no precipitation and air is lower.
temperature below 0°C conditions, a 2 km/h speed reduction can be
detected. The speed reduction due to rain conditions, with insignif-
icant rain intensity (1 mm/h), is 5.5 km/h. Snow conditions have the Effect of Precipitation Intensity on Speed
largest impact on speed, with a speed reduction of 9 km/h.
The precipitation intensity is a variable that only exists in Cli-
mates 3 (rain) and 4 (rain in freezing and snow conditions). In this
case, it is important to point out that Climate 4 shows a much higher
Effect of Truck Percentage on Speed
regression coefficient than for rain conditions (over three times).
Although the analysis of non-weather-related conditions with Figure 5 shows the evolution of speed depending on precipitation
speed is not the aim of this paper, this variable had to be added to intensity. The figure shows that the speed reduction in snow condi-
the regression models because of its important effect. According tions is much more dramatic than in rain conditions. Both of them
to the global regression, the truck percentage coefficients are very tend to asymptotic values of 122.7 km/h and 116.0 km/h, respec-
similar for all climates, showing that the speed reduction due to tively. Therefore, speed variations of 7.00 km/h and 13.7 km/h are
the heavy traffic percentage is almost not affected by the climate. obtained for heavy rain and heavy snow, compared to the refer-
ence climate with no visibility loss. It is important to point out that
Moreover, the individual regression coefficients for each site are
these results consider only the precipitation intensity effect and
very similar, so the heavy traffic amount also has a similar effect
do not consider the effect of visibility loss. The rain effect appears
in all sites.
to reach its maximum at an intensity of 5 mm/h, while snow reaches
it at 7 mm/h.
Many studies that develop speed models based on rain and snow
Effect of Visibility on Speed conditions gather their intensities into two or three groups. Often, these
groups are determined based on meteorological recommendations,
Visibility varies from 0 to 2,000 m, being 2,000 (total visibility) in
not by considering their effect on speed.
most cases. As a result, a logarithmical model was used. In this research, an LSD analysis has been done to determine if
By examining the regression coefficients, it can be noted that it is possible to make different groups for the precipitation values.
the effect of visibility reduction is the highest in Climate 4 and the Those analyses showed no significant results, so the regression
lowest in Climate 2. Climates 1 and 3 appear to behave in a sim- model was maintained instead of creating groups for different
ilar way. Considering a 300-m visibility, the speed reductions precipitation ranges, depending on their effects on speed.
obtained are 4.8, 1.5, 4.3, and 8.7 km/h for Climates 1, 2, 3, and
4, respectively.
Figure 3 represents the speed variation for all climates. Climates 3 COMPARISON WITH PREVIOUS MODELS
and 4 are also represented with 1 mm/h precipitation intensity.
After obtaining the models, a comparison with previous models was
done. Because most previous studies give ranges of speed reduc-
Effect of Wind on Speed tions under adverse weather conditions, evaluation of the MSE is
not possible.
The wind speed coefficient is similar in Climates 1 and 3 and larger Most other models also gather rain and snow intensities and pro-
(in absolute value) for Climate 4. Figure 4 shows the traffic speed vide a unique value. Some examples are the Highway Capacity Man-
variation due to wind speed. ual (2), Ibrahim and Hall (6), Smith et al. (11), and Rakha et al. (12).
Camacho, García, and Belda 157

FIGURE 3 Speed variation depending on visibility conditions.

FIGURE 4 Speed depending on wind speed.


158 Transportation Research Record 2169

FIGURE 5 Effect of precipitation intensity on speed for Climates 3 and 4.

A few studies used regression models. One of these studies used a Rakha et al. (12) determined that the maximum speed reduction
model for passenger cars (10): in snow conditions was reached with 3 mm/h intensity. This value
is lower than the one determined in this study (7 mm/h).
v = 100.2 − 16.4 i snow − 9.5 i wet + 77.3 i vis − 11.7 i windd (9) Values given for light intensities are those associated with 1 mm/h
precipitation intensity. Values associated with heavy intensities are
where
the asymptotic ones. All of them are compared to the values given by
snow = 1 if there is a snow layer on the pavement, 0 if not; Climate 1 with no visibility loss and wind speed lower than 8 m/s.
wet = 1 if the pavement is wet, 0 otherwise (In accordance to
the present model, it will be 1 for Climates 3 and 4 and 0
for Climates 1 and 2);
vis = 0.28 if visibility is equal or higher to 0.28 km, and the TABLE 4 Existing Rain and Snow Investigations: Speed Reduction
visibility value (in km) for lower values; and
wind: 1 if the wind speed exceeds 24 km/h (6.7 m/s) and Light Heavy
Light Rain Heavy Rain Snow Snow
0 if not.
HCM (2) 2 km/h 5–7 km/h 1 km/h 37–42 km/h
Rain and Snow Effects Ibrahim and 2 km/h 5–10 km/h 3 km/h 38–50 km/h
Hall (6)
Each one of the existing studies considers rain and snow precipita- Kyte et al. 9.5 km/h ND 25.9 km/h ND
tions in a different way. The analyzed methods are listed in Table 4, (10)
including the proposed models in this research. Smith et al. 2.9–4.8 km/h 2.9–4.8 km/h ND ND
Previous data referred to the reduction in speed due only to precip- (11)
itation intensity, but did not consider any other associated effect, such Rakha et al. 2% to 3.6% 6% to 9% 5% to19% 5% to19%
(12)
as visibility loss. The research developed by Kyte et al. (10) does not
consider precipitation intensities, but only the road pavement surface. Camacho 5.5 km/h 7.0 km/h 9.0 km/h 13.7 km/h
et al.
Thus, the corresponding speed reductions are associated with the (this study)
lowest possible intensity.
Camacho, García, and Belda 159

The present model makes the speed less dependent on precipi- sion, so speed is not affected when visibility is about 2,000 m, but
tation intensity than previous research. As all the previous models, the induced speed reduction was large when the visibility values
it also shows a higher speed reduction in snow than in rain condi- were low. It was also determined that the influence of any weather
tions, but the difference between light and heavy snow is not as variable on speed was higher in snow conditions than in other cases.
dramatic as in those models. These differences, depending on the
precipitation intensity, may exist because other studies do not con-
sider visibility as a separate variable, including its effect on these ACKNOWLEDGMENTS
speed reductions.
The authors thank the Center for Studies and Experimentation of
Public Works (CEDEX) of the Spanish Ministry of Public Works,
Visibility Effects which subsidized this research, included in a research project,
METEOSAFETY: Study to Reduce Accidents in Case of Weather
In this study visibility is introduced in a logarithmical way, making Problems. The authors also thank the Ministry of the Interior, General
speed reduce at higher rates as the visibility decreases. Snow con- Directorate of Traffic, Spain, for cooperation in field data gathering.
ditions cause a reduction in visibility to be more affected than in
other climates. Rakha et al. (12) also reported this behavior, giving
a maximum speed loss due to visibility of 10%. According to the
REFERENCES
model presented by Kyte et al. (10), the maximum speed reduction
is 21.6 km/h. 1. Chin, S. M., O. Frazese, D. L. Greene, H. L. Hwang, and R. C. Gibson.
Temporary Losses of Highway Capacity and Impacts on Performance.
http://cta.ornl.gov/cta/publications.shtml#2002, 2002. Accessed May
Wind Effects 2008.
2. Highway Capacity Manual. TRB, National Research Council, Wash-
Few studies have researched the effect of wind speed on traffic ington, D.C., 2000.
3. Olson, P. L., D. E. Cleveland, P. S. Fancher, L. P. Kostyniuk, and L. W.
behavior. Kyte et al. (10) included its effect in their regression Schneider. NCHRP Report 270: Parameters Affecting Stopping Sight
model, by considering it as a dummy variable that reduced the Distance. TRB, National Research Council, Washington, D.C., 1984.
speed 11.7 km/h with a wind speed higher than 6.7 m/s. In this 4. Lamm, R., E. M. Choueiri, and T. Mailaender. Comparison of Operat-
study it has also been identified as a change in speed behavior when ing Speeds on Dry and Wet Pavements of Two-Lane Rural Highways.
the wind speed reaches 8 m/s, but in this case a regression coeffi- In Transportation Research Record 1280, TRB, National Research
Council, Washington, D.C., 1990, pp. 199–207.
cient has been adjusted. Comparing all climates to the neutral cli-
5. Hawkins, R. K. Motorway Traffic Behaviour in Reduced Visibility Con-
mate with total visibility, speed reduction is reached at wind speeds ditions. Proc., Second International Conference in Vision in Vehicles,
of 27.2 m/s, 42.3 m/s, 18.1 m/s, and 10.3 m/s for Climates 1, 2, 3, Elsevier, Amsterdam, Netherlands, 1988.
and 4, respectively. 6. Ibrahim, A. T., and F. L. Hall. Effect of Adverse Weather Conditions
on Speed-Flow-Occupancy Relationships. In Transportation Research
Record 1457, TRB, National Research Council, Washington, D.C.,
1994, pp. 184–191.
CONCLUSIONS 7. Brilon, W., and M. Ponzlet. Variability of Speed-Flow Relationships
on German Autobahns. In Transportation Research Record 1555,
In this paper several models have been developed to estimate the TRB, National Research Council, Washington, D.C., 1996, pp. 91–98.
speed reduction at free-flow conditions under adverse weather 8. Holdener, D. J. The Effects of Rainfall on Freeway Speeds. ITE Journal,
conditions. Data were obtained from 15 freeway locations in the Vol. 68, No. 11, 1998, p. 16.
northwestern Spain during almost 3 years, in a 15-min interval 9. Edwards, J. B. Speed Adjustment of Motorway Commuter Traffic to
Inclement Weather. Transportation Research Part F, Vol. 2, No. 1, 1999,
format. pp. 1–14.
Several individual and global analyses have been developed. 10. Kyte, M., Z. Khatib, P. Shannon, and F. Kitchener. Effect of Weather
From these it was concluded that for this type of study it is neces- on Free-Flow Speed. In Transportation Research Record: Journal of
sary to consider a large number of sites, due to the high variability the Transportation Research Board, No. 1776, TRB, National
of the weather–traffic relationship among different locations. Research Council, Washington, D.C., 2001, pp. 60–68.
11. Smith, B. L., K. G. Byrne, R. B. Copperman, S. M. Hennessy, and N. J.
Four global climates were distinguished, considering several vari-
Goodall. Investigation into Impact of Rainfall on Freeway Traffic Flow.
ables in a multiple nonlinear regression. Variables included were Presented at 83rd Annual Meeting of the Transportation Research Board,
truck traffic intensity, visibility, wind speed, precipitation intensity, Washington, D.C., 2004.
and snow layer depth. 12. Rakha, H. A., M. Farzaneh, M. Arafeh, and E. Sterzin. Inclement
It was determined that the rain made the speed reduce from 5.5 km/h Weather Impacts on Freeway Traffic Stream Behavior. In Transporta-
tion Research Record: Journal of the Transportation Research Board,
to 7.0 km/h, depending on its intensity. Speed reduction due to snow
No. 2071, Transportation Research Board of the National Academies,
varied from 9.0 to 13.7 km/h. Washington, D.C., 2008, pp. 8–18.
Wind speed affected speed only when it goes beyond 8 m/s. The
speed reduction due to visibility loss had a logarithmical expres- The Surface Transportation Weather Committee peer-reviewed this paper.

Você também pode gostar