The subjective opinion was that the time-lagged ensemble of mesoscale models (WRF
and MM5), which were being run hourly out to 15 hours, had a better handle on
precipitation timing and amount. In addition, because of the "hot start" process, it was
felt that the models did a better job of identifying the prediction of very light precipitation
cases. As can be seen above, the WRF and MM5 data were given 80% of the total
weight for QPF in the RWFS.
Subjective analysis of the performance of the RWFS QPF predictions, both with and
without the inclusion of the FSL models, was conducted as part of the case studies in
Section 10.
Results and Recommendations: The ability of models to predict precipitation
start and stop time and amount varies greatly between models. If a data fusion
system that adjusts itself based on observations similar to the RWFS is utilized,
care must be taken to ensure that the weights given to individual prediction
modules are appropriate. If only low quality verification data are available, the
weights should be fixed based on experience and/or expert opinion.
11.1.4 Insolation
It is well recognized that incoming solar radiation (insolation) is a critical parameter for
predicting road temperature. During the winter of 2003, insolation values were calculated
in SNTHERM-RT from weather model cloud layer data. This approach was used
because, until recently, operational weather models did not explicitly output insolation
data. For the 2004 season, it was the opinion of the labs that it would be better to employ
the explicit calculation of insolation provided by the mesoscale models, rather than using
RWFS cloud layer predictions to estimate insolation in SNTHERM-RT; therefore, the
MDSS was configured to utilize the insolation values from the models this year. This
was the right approach as some test cases indicated that the road temperature error was
reduced by about 50%, but several data quality issues related to modeled insolation had
to be resolved during the demonstration period.
As part of the performance assessment task in 2003, CRREL analyzed SNTHERM-RT
using actual weather measurements in Iowa to determine its skill given "perfect" weather
inputs. The goal was to assess the skill of the model itself and this could only be done if
the model was provided highly accurate weather inputs. Without insolation data near the
Iowa routes, it was difficult to fully assess model skill as the insolation values had to be
estimated from cloud cover observations. As part of the MDSS project in 2004, a
pyranometer was installed at the Ames garage. The primary objective was to measure
short wave radiation so that a more thorough analysis of SNTHERM-RT could be
performed after the field demonstration.
100
Previous Page
