Effect of Winter Weather and Road
Surface Conditions on Macroscopic
Traffic Parameters
Tae J. Kwon, Liping Fu, and Chaozhe Jiang
This paper presents an empirical study focusing on identifying the main
factors that affect the capacity and free-flow speed (FFS) of urban free-
ways under inclement winter weather conditions. The weather and road
surface condition factors examined include air temperature, wind speed,
hourly snow intensity, visibility, snow on ground, and road surface con­
dition describing the road slipperiness caused mainly by snow events.
Data on traffic operations and the associated weather and road conditions
observed at two freeway locations over the 2010–2012 winter seasons were
used in an extensive statistical analysis. Linear regression models were cal­
ibrated for both capacity and FFS reductions as related to various weather
and road condition variables. It was found that visibility and road surface
conditions had a statistically significant effect on both capacity and FFS.
Snow intensity was found to be significant only when the visibility factor
was excluded; this finding suggests a refutation of these two factors on
capacity and FFS. The modeling results were compared with those recom-
mended by the Highway Capacity Manual 2010, showing that, in many
cases, the manual could underestimate or overestimate the effects of win-
ter weather conditions and that the proposed models ­provided a more
reasonable estimate at a higher level of granularity.
Accurate estimation of the effect of winter weather on traffic flow
parameters such as capacity and free-flow speed (FFS) is of high
importance, particularly in countries that exhibit severe winter
weather. Capacity and FFS are critical parameters for facility
design, traffic management, and operations control. Quantifica-
tion of the effect of winter weather and road surface conditions on
traffic operations is also essential for developing efficient winter
road maintenance programs. Despite these critical importances,
past research has not thoroughly investigated the issue and has
not been able to quantify the winter weather impact on traffic
operations. What are the most significant predictors affecting the
variation of capacity and FFS? How easily can they be adapted
and utilized by highway personnel for effective traffic control
and operation? How valid are the effect estimates given in the
Highway Capacity Manual 2010 (HCM 2010)? These questions
represent the main concern of the research.
Department of Civil and Environmental Engineering, University of Waterloo, Water-
loo, ON N2L 3G1. Alternate affiliation for L. Fu: School of Transportation and Logis-
tics, Southwest Jiaotong University, Chengdu, China. Corresponding author: T. J.
Kwon, tkwon@uwaterloo.ca.
Transportation Research Record: Journal of the Transportation Research Board,
No. 2329, Transportation Research Board of the National Academies, Washington,
D.C., 2013, pp. 54–62.
DOI: 10.3141/2329-07
54
The objectives of this study are to investigate and identify the
key factors that contribute to a variation of capacity and FFS during
inclement winter weather conditions and to develop models that can
be used to estimate the capacity and FFS. Furthermore, the findings
of the proposed study will be compared with the values documented
in the HCM 2010.
The paper is organized as follows. The first section presents a
literature review of the topic. The second section describes the data
sources and their processing. Model calibration and validation are
described in the third section, followed by a discussion of the results
in the fourth section. The last section highlights the conclusions and
provides recommendations for future work.
Literature Review
Prior research has attempted to determine the effect of inclement
weather conditions during winter by investigating the relationship
between various weather factors and traffic stream variables. Lamm
et al. studied the impacts of rain on freeway capacity in conjunction
with geometry of curve using 322 curved roadway sections of rural
highways (1). Their study revealed a statistically significant relation-
ship between operating speed and degree of curvature. The study
also claimed that wet pavement does not affect vehicle operating
speed if visibility is not affected appreciably by heavy rain. How-
ever, other studies showed that even light rain can trigger changes in
capacity and FFS (2, 3). Brilon and Ponzlet reported that wet road-
way conditions caused drops in speed and capacity of 9.5 km/h and
350 vehicles per hour (vph) (4). However, their data were collected
on several German autobahns (e.g., no speed limit); as a result the
conclusions may not be applicable to other urban highways.
Kyte et al. undertook a study on the effect of inclement weather
on highway capacity and compared their results with those sug-
gested in the Highway Capacity Manual 2000 (5). Their study was
able to show that light rain or light snow caused a higher reduc-
tion in speed by 50%, but heavy snow caused a lower reduction
in speed by 20% when compared to the values suggested in the
HCM 2000 (5). They also investigated the impacts caused by other
weather variables, such as wind speed and visibility, and reported
that wind speeds higher than 48 km/h caused a reduction in speed
by 9.0 km/h, but visibility between 0.2 and 37.0 km caused a very
marginal decrease in speeds. Liang et al. conducted a similar study
using a shorter stretch of the same test sites and concluded that aver-
age operating speed was reduced by 18.13 km/h for snow events
(6). In their regression analysis, they incorporated an additional
predictor called road surface condition (RSC) as a binary variable
(1 if snow covered, 0 otherwise) and found a reduction in average
Kwon, Fu, and Jiang
speed of 3.5 km/h when the road surface was covered with snow.
However, this binary representation of RSC does not capture the
full variation of RSC that could commonly be observed in the real-
world environment. Furthermore, both of these two studies did not
consider precipitation intensity and were based on rural highways
where traffic flows rarely reach capacity.
Ibrahim and Hall also undertook an analysis using a dummy vari-
able multiple regression analysis scheme to determine whether there
were significant differences in traffic operations between rainy and
snowy weather conditions as opposed to clear weather conditions
(3). Their study confirmed that light rain and snow showed 3% to
5% reduction in speeds, whereas heavy rain and snow caused 14%
to 15% and 30% to 40% reductions in speed, respectively. However,
their findings were based on data obtained over six clear, two rainy,
and two snowy days and are therefore limited in generality.
Agarwal et al. conducted a study similar to that of Ibrahim and
Hall, but using 4 years of data from three highway locations (7). A
categorical analysis was performed to link the percent reductions in
capacity and FFS to five different weather variables: rain, snow, tem-
perature, wind, and visibility. They found that rain, snow, tempera-
ture, wind, and visibility cause 2.01% to 14.13%, 4.29% to 22.43%,
1.07% to 8.45%, 1.07% to 1.47%, and 9.67% to 10.49% reduction
in capacity, respectively, and 1.62% to 6.05%, 4.17% to 13.46%,
1.21% to 2.10%, 0.54% to 1.12%, and 6.62% to 11.78% reduction
in average operating speed. Some of these results have been incor-
porated into the HCM 2010 (8). It should, however, be noted that in
this study each of the weather variables was treated independently
without taking into account the potential joint effects that might have
contributed to their results. For instance, reductions in capacity or
FFS can be jointly affected by both snow and visibility because, dur-
ing snow events, visibility is likely to be affected by an equal amount
of intensity of snowfall. As a result, the estimates on the magnitude
of the effect of the weather-related variables may be biased.
A similar study was conducted by FHWA (9) in which the same
intensity categories as in Agarwal et al. (7) were considered. In the
FHWA study, the impact of adverse weather conditions (precipitation
and visibility) was quantified by analyzing traffic stream behavior as
well as key traffic stream parameters, including FFS, capacity, and jam
density. It was found that reductions in FFS and capacity for light snow
were in the range of 5% to 16% and 12% to 20%, respectively, and 5%
to 19% in both FFS and capacity for heavy snow. It was also found that
a reduction in visibility from 4.8 km to 0.0 km resulted in reductions
of 10% in FFS and capacity. Despite a thorough investigation, the use
of categorical variables (e.g., light and heavy), as in ­Agarwal et al. (7)
and FHWA (9), could lead to over- or ­underestimation of the effect of
a snowstorm because of its low granularity.
Kwon and Fu conducted an empirical investigation on the effects of
light rain and light snow and compared the findings with those from
the HCM 2010 (10). They found that light rain caused 2.5% to 4.7%
and 5% to 7% reductions in FFS and capacity, respectively, while
light snow caused a drop of 21% to 24% and 17% to 19% in capac-
ity and FFS, respectively, suggesting that HCM 2010 underestimates
the effects. However, their study considered only a limited number of
factors and types of winter events (e.g., heavy snowstorms and factors
such as road surface conditions and visibility were not included).
In summary, there are several important limitations in the past
efforts on quantifying the effect of winter weather–related variables
on the capacity and FFS of a highway. First, past investigations
have not been able to cover the wide range of winter weather condi-
tions that may be experienced in many wintery regions. Second, the
effects of various factors and their interactions have not been fully
55
quantified statistically. While HCM has provided a set of adjustment
factors for taking into consideration the winter weather effects, few
studies have been conducted to show the validity of these results.
Data Description and Integration
Study Sites
Two detector locations on the Don Valley Parkway (DVP) of City
of Toronto were selected as the study sites for this research. The
DVP is one of the major highways in the City of Toronto with a
maximum speed limit of 90 km/h and a total length of 15.0 km and
is considered one of the busiest municipal routes in Canada (11).
Some sections of the DVP carry a daily average of 100,000 vehicles,
which is far beyond its intended daily capacity of 60,000 vehicles
(12). These locations meet the requirements that traffic congestions
should be observed in both morning and evening peak hours and
that other required data needed to carry out this study are also avail-
able. The former requirement ensures that both capacity and FFS
can be measured by analyzing speed–flow relationships.
Figure 1 shows the two sites, DN0115DND and DN0125DND,
located between the intersections of DVP and Ellington Avenue
and DVP and Lawrence Avenue, and they are to be used for model
calibration and validation, respectively. The reason for choosing two
nearby sites is to legitimize the assumption that macroscopic traffic
behaviors observed in these two sites are similar so that one site could
be used for model calibration and the other for model validation.
Data Sources and Processing
Traffic Data
For our intended analysis on capacity and FFS, it is essential that the
relevant traffic data be available. The traffic data used in the study
have been collected from vehicle detection stations (VDS) managed
by the City of Toronto through its traffic management system called
RESCU, which constantly monitors the traffic movement from a
total of 168 VDS sites installed on major highways in Toronto.
Lane-based 20-s raw traffic data consisting of speed, volume, and
occupancy were available for each study site on a daily basis from
2002 to 2012; in this analysis, data were selected over the past three
winters to ensure consistency in external conditions. This constraint
on time intervals was also determined on the basis of the availabil-
ity of other data sources that must also be available for the same
time period (a point discussed in the later sections). To conduct a
preliminary analysis, a few chronic bottleneck areas along the DVP
were identified and scrutinized by drawing speed–flow relationships
to determine whether the flow reaches the capacity. Prior studies
indicate that use of 5- to 15-min intervals for data is appropriate
for traffic stream analysis (13, 14). Because a 5-min interval is able
to capture the detailed movement of the flow while sufficiently
smoothing the short-lived peaks in flow rates, it is therefore used
as the analysis period.
Weather Data
Weather data were obtained from Environment Canada, which pro-
vides online access to the hourly and daily archived weather data at
various weather stations across Canada. Typical observations made
56
FIGURE 1   Location of two VDS sites. (S ource : Google Earth.)
at each station include air temperature, dew point temperature, rela-
tive humidity, precipitation type, visibility, and wind speed, all on
an hourly basis with the exception of the precipitation intensity and
snow on ground, which are available in daily totals. The climate sta-
tion at Toronto Buttonville Airport was selected because it is located
near the study sites (i.e., within 5 km) and has all the data for the
entire study period. The weather data were extracted in two steps. In
Step 1, monthly data were observed to identify the dates on which it
snowed; in Step 2, daily data were subsequently reviewed to deter-
mine the times at which snow events started and ended. The dates
when snowfalls occurred and the number of hours of snowfall were
documented. Other weather parameters such as visibility, tempera-
ture, and wind speed were also recorded and averaged based on the
number of hours of continuous snowfall. Other weather parameters
such as relative humidity and pressure were not included in this
process because they were considered to have insignificant effects
on the variation of capacity and FFS (2).
RSC Data
RSC reflect both the nature (e.g., snow, slush snow, and ice) and the
amount (e.g., bare, fully covered, and partially covered) of road sur-
face contaminants, which are commonly measured and categorized
visually. Two data sources were available for determining the RSC
of the study sites. The first data source is patrol reports from the
City of Toronto that contain information about patrol date and time,
weather summary, snow accumulation on road at the time of patrol-
ling, and winter road maintenance activities (e.g., plowing or salt-
ing). The data are reported by maintenance contractors, who patrol
the designated routes three to four times depending on the severity
of snow events. Note that the snow accumulation reported is differ-
ent from total precipitation because it should also reflect the results
of traffic and road maintenance operations. In this research, this data
Transportation Research Record 2329
item was used as an input to infer the road surface condition at the
study sites because it could be reasonably assumed that accumula-
tive snow amount is somewhat correlated to road slipperiness (i.e.,
more snow on the ground leads to a more slippery road). This patrol
report was available for the entire study period whenever a snow
event caused patrollers to document their observations.
The conditions reported by patrollers represented the conditions
of the main highways in the Toronto area and are therefore limited
in representing the actual conditions of the study sites. To compen-
sate for such deficiency, video data from RESCU cameras have been
employed, which record and archive the road condition images every
5 min. The City of Toronto provided the archived image data cover-
ing the study period (i.e., equivalent to approximately 17.4 million
images). The data from two RESCU cameras located adjacent to
the two study sites were identified and used in the analysis of the
proposed study.
These two sources of data were subsequently used to estimate the
average amount of snow accumulated on the road surfaces at these
sites. This amount is assumed to proportionally reflect the slipperi-
ness of the road surface conditions; as a result, it is scaled to a sca-
lar variable called road surface condition index (RSI) with values
ranging from 0 (icy) to 1 (dry). RSI was determined in an iterative
fashion involving two steps: the first step is to determine the snow
accumulation on the road by analyzing the information available on
patrol reports and Environment Canada weather archives (e.g., snow
on ground), then the second step follows to check video images
to verify if the amount of snow found in Step 1 matches the actual
road condition. This is a labor-intensive process achieved by thor-
oughly reviewing the road condition images. Once the values for
all weather events are determined, they are normalized on the basis
of the ­maximum value to fit each value in the range of 0 to 1. RSI
could be viewed as a surrogate measure of the friction level of a road
surface. For each snow event that occurred within the study period,
when the observed snow accumulation information was available,
Kwon, Fu, and Jiang 57

the average RSI representing the overall condition of the road
surface for that event was calculated. In the process of computing
this value, image data were simultaneously reviewed to confirm that
the value truly represents the road surface condition at that time. This
process was repeated for as long as the data from snow advisory and
the data from the two RESCU cameras were available.
Modeling the Inclement
Weather Effects
Model Formulation
Once all the data sets were prepared for the 3-year period, they
were rearranged according to the dates of snow events. There was a
total of 25 days of 5-min interval data, of which there were 268 h of
snow events and 48 h of normal conditions. Data for normal condi-
tions were used to determine the capacity and FFS under normal
conditions at these sites, which are used as benchmarking values in
determining the effects of adverse weather conditions.
The traffic flow and speed data over each event were extracted and
the speed–flow relationship was plotted and visually examined for
identifying the capacity and FFS. If the capacity was not reached over
the period of an event, the corresponding data were excluded. For each
applicable event, the capacity and FFS were computed with the high-
est 5% values of the observed traffic flow and speed as recommended
in the prior research (15).
Table 1 gives the summary statistics of the variables, which were
included in the model for a subsequent multiple regression analy-
sis. It is worthwhile to mention that visibility data for capacities
and FFS were found to have a logarithmic trend; as such, visibility
data were transformed into a logarithmic scale. This makes intuitive
sense because visibility should have a diminishing effect on driving
behavior once it reaches a certain clearance level. Figure 2 shows
the speed–flow relationship over some example events, which were
arranged in a chronological order accompanied with the averaged
weather and road conditions, the computed capacity, and FFS.
Model Calibration and Validation
Multiple regression analysis was carried out to determine the
statistically significant factors that affect the variation of capacity
and FFS during inclement weather conditions. An explanatory data
analysis conducted in the earlier stage of this study indicated that
all predictors except visibility had a linear relationship with the
observed capacity and FFS. Visibility was logarithmically trans-
formed and the resulting term was found to have an improved linear
relationship with the capacity and FFS.
The overall quality of the model was assessed with the R2 value,
and the statistical significance of each predictor was tested on the
basis of the 95% confidence interval. In addition, p-value was used
to determine if a predictor is statistically significant at a level of
significance of 5%. Following these steps, two variables, RSI and
ln(visibility), were found to be statistically significant variables affect-
ing the variation of both capacity and FFS with a high explanatory
power of R2 being 91% and 84%, respectively.
Surprisingly, the snow intensity variable was not found to be
statistically significant. A correlation analysis revealed that snow
and ln(visibility) were highly correlated with a correlation value
of −70.2%. Such a strong negative correlation makes logical sense
because when there is a heavy snowstorm event, visibility is mostly
likely to be affected, thus reducing both capacity and FFS. Consid-
ering the high collinearity between the two variables, an alternative
model was calibrated including all variables except ln(visibility).
As expected, RSI and snow intensity were found to be significant
variables with R2 values of 76% for capacity model and 69% for
FFS model. For a comparison purpose, both models are used in the
further analysis. The results of multiple regression analysis for four
models (i.e., two for capacity and other two for FFS) are summarized
in Table 2.
The four calibrated models were then validated using the data
obtained from a nearby VDS site (DN0125DND) as discussed pre-
viously. The capacity and FFS were extracted from the validation
site by analyzing the speed–flow relationships in the same fashion
as that of the original VDS site (DN0115DND). Figure 3a shows
the comparisons between the observed capacities and the predicted
capacities, whereas Figure 3b shows observed FFS and the predicted
FFS. The root mean square error (RMSE) values were computed to
check the prediction performance of the four models and those were
76 vehicles per hour per lane (vphpl) and 112 vphpl (equivalent to
average deviations of 4.4% and 6.5%) for the two capacity models,
respectively, and 2.20 km/h and 2.92 km/h (equivalent to average
deviations of 2.4% and 3.2%) for the two FFS models. The results
also confirmed that the first models incorporating ln(visibility) and
RSI better explained the variability within the data sets.
Results and Discussion

Model Interpretation
TABLE 1   Descriptive Summary of Statistics (Based on 316
Hourly Observations) From the results of the statistical analysis described in the previ-
ous section, the following observations can be made on each of the
Variable Minimum Maximum Mean SD significant variables:

Snow (mm/h) 0 16.57 2.78 3.34 • Effect of visibility. As anticipated, visibility was found to
RSI 0.07 1 0.81 0.23 have a significant effect on capacity and FFS models with a posi-
Wind (km/h) 5.4 38.78 21.25 10.69 tive model coefficient sign, suggesting that as visibility increases,
Visibility (km) 1.48 24.1 9.03 5.79 capacity and FFS also increase. This makes intuitive sense and is
ln(visibility) 0.39 3.18 2.01 0.66 supportive of literature findings (6, 7). As indicated previously, this
Temperature (°C) −12.21 2.47 −5.75 3.77 variable was included in a logarithmic scale such that both capac-
Capacity (vphpl) 1,074 2,016 1,668.48 227.02 ity and FFS are reduced by a greater extent toward low visibility.
FFS (km/h) 82.42 102.32 91.57 4.93 For example, visibility greater than 1.45 km was found to have a
diminishing effect of less than 5% reductions in capacity. Similarly,
Note: SD = standard deviation; vphpl = vehicle per hour per lane. visibility greater than 1.0 km had less than 5% reductions in FFS.
58 Transportation Research Record 2329

Jan-11-2010 Jan-28-2010

Feb-03-2010 Dec-13-2010

Jan-12-2011 Jan-20-2011

Feb-10-2011 Jan-30-2012

FIGURE 2   Speed–flow relationships observed during various weather conditions: capacity and FFS comparisons.
Kwon, Fu, and Jiang 59

TABLE 2   Summary Results of Capacity and FFS Models

95% Confidence Interval

Predictor Coefficient SE t Sig. Lower Bound Upper Bound

1st Capacity Model: Calibrated Using All Variables (R2 = 91%)

(Constant) 814.27 62.25 13.08 7.46 E–12 685.17 943.36
RSI 463.41 71.71 6.46 1.68 E–06 314.69 612.14
ln(visibility) 226.51 24.69 9.17 5.67 E–09 175.3 277.72
2nd Capacity Model: Calibrated Using All Variables Except ln(visibility) (R2 = 76%)
(Constant) 1,222.89 103.71 11.79 5.57 E–11 1,007.8 1,437.98
Snow (mm/h) −31.97 7.37 −4.34 2.66 E–04 −47.26 −16.68
RSI 619.06 108.52 5.7 9.75 E–06 394 844.12
1st FFS Model: Calibrated Using All Variables (R2 = 84%)
(Constant) 75.33 1.77 42.6 7.10 E–22 71.65 79
RSI 5.15 2.09 2.47 2.23 E–02 0.81 9.49
ln(visibility) 5.84 0.73 8.02 7.86 E–08 4.32 7.35
2nd FFS Model: Calibrated Using All Variables Except ln(visibility) (R2 = 69%)
(Constant) 85.81 2.57 33.4 1.09 E–19 80.47 91.15
Snow (mm/h) −0.86 0.18 −4.7 1.21 E–04 −1.24 −0.48
RSI 9.54 2.7 3.53 1.98 E–03 3.92 15.16

Note: SE = standard error; sig. = significance.

• Effect of RSC. Road surface conditions as represented by RSI
were found to have a statistically significant effect on both capacity
and FFS during snow events. The positive model coefficient also
makes intuitive sense as supported by previous studies (16–18), which
suggest that increased RSI (i.e., better road conditions) are correlated
with increased capacity and FFS (i.e., better traffic conditions). For
example, under the given snow intensity of 5 mm/h, at RSI = 0.2
(snow covered), capacity and FFS are reduced by 44.24% and 17.01%,
respectively, whereas at RSI = 0.8 (bare wet), capacity and FFS are
reduced relatively less by 24.08% and 11.01%, respectively. Note that
this result is particularly interesting because the RSI could be affected
by winter road maintenance, which means the effect of winter road
maintenance on improving traffic operations could be quantified.
• Effect of snow intensity. Hourly snow intensity was also found
to be statistically significant in both the capacity and FFS models
in which visibility was not included. The negative model coeffi-
cient makes intuitive sense because it suggests that increase in snow
intensity is associated with decrease in capacity and FFS. It had the
anticipated impact on both capacity and FFS by demonstrating a
linear relationship with their reductions. For instance, two differ-
ent hourly snow intensity rates of 2.0 mm/h and 15.0 mm/h would
cause percent reductions of 3.5% and 26% in capacity and 1.8% and
13.5% in FFS, respectively.
It should be noted that this analysis did not find average tem-
perature during snow events to be a significant factor affecting the
capacity and FFS, which makes intuitive sense in general. Further­
more, wind speed was also not found significant in all models.
According to the literature, variations caused by wind speed are rel-
atively small but can be noticeably large when wind speed reaches
a certain level of intensity (e.g., greater than 32 km/h) (6, 18). It is
possible that the effects caused by wind speed were not captured by
the models because the data sets used to calibrate the models did not
contain many high-range wind speed data (i.e., average wind speed
used in the analysis was 21 km/h).
Comparisons of Model Results with HCM 2010
Figure 4 compares the effects of weather factors on capacity and
FFS shown in the model results with those recommended by the
HCM 2010. Note that the results on FFS were not directly obtained
from HCM 2010, which does not contain the information on the
effect of weather variables on FFS. Instead, they were obtained
from Agarwal et al., which was the source of information for the
effect on capacity that was included in the HCM 2010 (7). There-
fore, the results of this study were implemented for comparisons
against the two FFS models developed in this study. To reinforce a
fair comparison with the HCM 2010, all model estimates presented
in Figure 4 (i.e., all solid lines) were calculated by setting RSI equal
to 1.0 (i.e., dry condition) so that the effects caused purely by either
visibility or snow intensity could be captured.
As can be seen clearly from Figure 4, the proposed model has
the advantage of providing estimates on the effect of varying winter
weather conditions at a higher level of granularity. In other words,
there is no need to categorize the effects by range as was done in
the HCM 2010 because it is more prone to overestimation or under­
estimation of the inclement weather effects when step functions are
used instead.
Taking an example from Figure 4a, according to HCM 2010 (i.e.,
dotted line), the capacity percent reduction at snow intensities between
12.70 and 12.95 mm/h undergoes the greatest jump in discontinuity
from 11.04% to 22.43%. Such a huge variation does not well replicate
the actual phenomena that are observed in reality because capacity
would not possibly be reduced suddenly at one point when hourly
snow intensity is increased only by 2%. Conversely, the model esti-
mates (i.e., solid lines) well project the reality by showing a continuous
relationship between capacity reduction and snow intensity.
Figure 4, b and d, shows capacity and FFS percent reductions that
are expressed as a function of visibility. It is important to point out
that the visibility range in the original data was between 1.48 and
24.10 km (refer to Table 1), whereas the data range used in the HCM
60
(a)
(b)
FIGURE 3   Comparisons between actual and predicted results.
2010 was between 0 and 1.6 km. One can therefore argue that such
a comparison is not valid in a rigorous sense because the HCM 2010
values are beyond the data range for which the model was calibrated.
However, the model captured well the original data trend at the low
range (less than 2 km); as a result, it is reasonable to believe that the
model could be extrapolated to predict the effect of visibility at a low
value range. As discussed briefly in the preceding section, capacity
and FFS reductions associated with visibility should become larger
as it moves toward zero visibility and such trend is captured by the
model. Conversely, the HCM 2010 suggests that speed be reduced by
10.5% even at zero visibility, which does not make a practical sense.
The overestimation problem of the weather effects, when com-
pared with the estimates from the model, can also be appreciated in
Transportation Research Record 2329
Figure 4c, where speed reduction appears to increase at a fast rate in
the light snow intensity range (e.g., 0 to 2.80 mm/h). Furthermore, it
is hard to be convinced that speed reductions do not vary from 9.4%
for a long stretch of snow intensity ranges (e.g., 2.80 to 12.70 mm/h),
whereas the model is able to produce more meaningful estimates.
Conclusions
In this research, the effects of inclement winter weather on highway
traffic operations were examined by conducting a thorough analysis
of the relationship between macroscopic traffic parameters (capac-
ity and FFS) and various weather and road condition factors. Data
Kwon, Fu, and Jiang
(a)
(c)
FIGURE 4   Comparisons between model results and results found in the HCM 2010: (a) capacity affected by snow, (b) capacity affected by
visibility, (c) speed affected by snow, and (d) speed affected by visibility.
from four different sources, including traffic data from VDS sites,
weather data from Environment Canada, and road surface condition
data from patrol reports and CCTV cameras, have been collected
and processed. A multiple linear regression analysis was subse-
quently performed to calibrate the models for estimating capacity
and FFS as a function of several weather variables, such as snow
intensity, visibility, air temperature, RSI, and wind speed. The key
findings are summarized as follows:
• Visibility was found to be statistically a significant factor affect-
ing the changes of capacity and FFS with high explanatory powers.
Snow intensity, which was excluded in the analysis because it had a
high correlation with visibility, was used to calibrate an alternative
model, from which snow intensity was found to be significant for
both capacity and FFS.
• Four calibrated models were validated by both statistical and
graphical analyses. Comparisons between model predicted and actual
observations showed low RMSE of 76 and 112 vphpl for capacity
models and 2.20 and 2.92 km/h for FFS models. Model estimates and
actual observations were found to be highly comparable.
• The model estimated capacity and FFS reductions were com-
pared with those from HCM 2010. The results indicated that HCM
2010 estimates were susceptible to underestimation or overestimation
of the reductions for both capacity and FFS. Conversely, the models
were able to provide effect estimates at a great level of granularity.
• RSC was found to be a significant factor in all models. Capac-
ity and FFS reductions can be estimated on the basis of different
61
(b)
(d)
types of RSC, making it possible to quantify the effect of winter
road maintenance on improving traffic conditions.
The research can be further extended in several directions. First,
the modeling process should be incorporated with larger data sets
and wider study areas to be able to capture various levels of severity
of inclement weather events. This is particularly important for testing
the applicability of the developed models to see if the model estimates
are also representative of other roads. Second, the potential non­linear
effects among tested variables should be thoroughly examined by
analyzing their moderating effects. Third, additional variables, such
as road geometry describing the unique features of a road (e.g., num-
ber of lanes, length of curve, sight distance, etc.), can be included
because they can potentially affect a variation of capacity and FFS.
Nevertheless, the approach employed in this study should be applied
for accurately estimating the capacity and FFS reductions during
inclement winter weather conditions.
Acknowledgments
The authors acknowledge traffic engineers Simon Foo and Stephen
Huang of the City of Toronto for providing the VDS and video
data that were essential in this study. The authors also thank Lisa
­Maasland for data transfer and Tom Kalogiannis for providing patrol
reports. The research was partially funded by National Sciences and
Engineering Research Council of Canada.
62
References
  1. Lamm, R., E. M. Choueiri, and T. Mailaender. Comparison of Operat-
ing Speeds on Dry and Wet Pavements of Two-Lane Rural Highways. In
Transportation Research Record 1280, TRB, National Research C­ ouncil,
Washington, D.C., 1990, pp. 199–207.
  2. Maze, T. H., M. Agarwal, and G. Burchett. Whether Weather Mat-
ters to Traffic Demand, Traffic Safety, and Traffic Operations and
Flow. In Transportation Research Record: Journal of the Transporta-
tion Research Board, No. 1948, Transportation Research Board of the
National Academies, Washington, D.C., 2006, pp. 170–176.
  3. 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.
  4. Brilon, W., and M. Ponzlet. Variability of Speed–Flow Relationships
on German Autobahns. In Transportation Research Record 1555, TRB,
National Research Council, Washington, D.C., 1996, pp. 91–98.
  5. Kyte, M., Z. Khatib, P. Shannon, and F. Kitchener. Effect of Weather
on Free-Flow Speed. In Transportation Research Record: Journal of
the Transportation Research Board, No. 1776, TRB, National Research
Council, Washington, D.C., 2001, pp. 60–68.
  6. Liang, W. L., M. Kyte, F. Kitchener, and P. Shannon. Effect of Envi-
ronmental Factors on Driver Speed: A Case Study. In Transportation
Research Record 1635, TRB, National Research Council, Washington,
D.C., 1998, pp. 155–161.
  7. Agarwal, M., T. Maze, and R. Souleyrette. Impact of Weather on Urban
Freeway Traffic Flow Characteristics and Facility Capacity. Center for
Transportation Research and Education, Iowa State University, Ames,
2005.
 8. Highway Capacity Manual 2010. Transportation Research Board of the
National Academies, Washington, D.C., 2010.
 9. Empirical Studies on Traffic Flow in Inclement Weather. Report FHWA-
HOP-07-073. FHWA, U.S. Department of Transportation, 2006.
Transportation Research Record 2329
10. Kwon, T. J., and L. Fu. Effect of Inclement Weather Conditions on
­Macroscopic Traffic Behavior. Presented at 9th International Transpor-
tation Specialty Conference, Canadian Society for Civil Engineering,
Edmonton, Alberta, Canada, 2010.
11. Rand McNally Canada. Toronto and Area Map Book (Map). Cartography
by Perly’s, 2010, p. 3, Section D1.
12. City of Toronto. Average Weekday, 24-Hour Traffic Volume. http://
toronto.ca/transportation/publications/brochures/24hourvolumemap.
pdf. Accessed May 5, 2010.
13. Hunt, J. G., and S. Y. Yousiff. Traffic Capacity at Motorway Road Net-
works: Effects of Layout, Incidents, and Driver Behavior. In Network
Design to Reduce Conflict: Proceedings of the 2nd International Sym-
posium on Highway Capacity, Australian Road Research Board Ltd.,
Vermont South, Australia, Vol. 1, 1994, pp. 295–314.
14. Smith, B. L., and J. M. Ulmer. Freeway Traffic Flow Measurement:
Investigation into Impact of Measurement Time Interval. Journal of
Transportation Engineering, Vol. 129, No. 3, 2003, pp. 223–229.
15. Smith, B. L., K. G. Byrne, R. B. Copperman, S. M. Hennessy, and
N. J. Goodall. Investigation into Impact of Rainfall on Freeway Traffic
Flow. Presented at 83rd Annual Meeting of the Transportation Research
Board, Washington, D.C., 2004.
16. Donaher, G., T. Usman, L. Fu, L. F. Miranda-Moreno, and M. S.
­Perchanok. Quantifying the Mobility Effects of Winter Snow Storms
and the Benefits of Winter Road Maintenance. Proc., International Con-
ference on Winter Maintenance and Surface Transportation Weather,
Iowa City, Iowa, 2012, pp. 173–186.
17. Norrman, J., M. Eriksson, and S. Lindqvist. Relationships Between
Road Slipperiness, Traffic Accident Risk, and Winter Road Mainte-
nance Activity. Climate Research, Vol. 15, 2012, pp. 185–193.
18. Kumar, M., and S. Wang. Impacts of Weather on Rural Highway Opera-
tions. Final Report. U.S. Department of Transportation, 2006.
The Surface Transportation Weather Committee peer-reviewed this paper.