11.2 Road Temperature Predictions
Road temperature predictions were made for each of the 16 road segments (including one
bridge) and for each RWIS site in the Des Moines and Ames area (see Fig. 6.1). The
MDSS provides three sets of road temperature forecasts for each segment of road:
untreated, recommended treatment and actual treatment. The "untreated" forecast is the
expected road temperature given that the weather occurs (snow fall on the road, etc.) and
no treatments are made (the road is left "as is"). "Recommended treatment" is the
expected road temperature, given that the treatment suggested by the MDSS is applied.
"Actual treatment" is the expected road temperature, given the treatment that was entered
into the system by IADOT personnel during the event. The "untreated" forecast is the
most unrealistic, since treatment will nearly always take place. The most appropriate
field to verify is the "recommended treatment" field, since it represents a treatment
scenario for each event. Only results for that field are discussed here.
RMSE and bias plots of road temperature forecasts using the recommended treatments
were examined for the entire season. The statistics are broken down into monthly periods
(January, February and March). The stats are cumulative, using selected segments/points
that have nearby RWIS observations used for calculating road temperature forecast error.
For all of the months there were considerably higher RMSE values during the heat of the
day (9am 3pm; 0-3 hr, 21-27 hr and 45-48 hr lead times; Fig. 11.8). These spikes in
RMSE were more prominent during the months of February and March when solar influx
was stronger, resulting in a larger diurnal temperature swing. In January, daytime RMSE
values were up to 4.5C, while they were as high as 6C during March. During the night
the RMSE values were between 1.5-3C during all months. Strong positive road
temperature biases were found during the morning hours (lead times 18-24 and 42-28 hr),
followed by weaker negative (cold) biases in the afternoon (Fig. 11.9). This daytime
swing in bias may be attributable to differences in the depth over which road temperature
is forecast (skin) and the depth of pavement that surrounds the measuring device ("puck",
several centimeters). Response time differences may have caused a phase shift in the
diurnal heating/cooling cycle. Cold biases up to 1.5C were present during the night, and
were typically strongest during the hours between midnight and sunrise. Road
temperature RMSE values were very similar at each lead-time, both with and without the
use of FSL models. A comparison of RMSEs using four different forecast initialization
times showed very similar results to those found for state parameters (Fig. 11.10).
108
Previous Page
