1	CLARUS Road-Weather Routing for Crash Risk Aversion
2	CLARUS Monitoring Stations
3	Linking Crashes and Weather A regression model was created Dependent Variable – A crash occurring within 50 miles of a weather station during a particular hour. Independent Variables Temperature (Air, Road and Dew Point) Precipitation Types Precipitation Intensities Visibility Wind Speed (Average and Gust) Atmospheric Pressure
4	Linking Crashes and Weather First cut: What variables are significant?
5	Linking Crashes and Weather The regression model implies linear effects, but… Temperature changes may have greater effects around freezing What is the critical visibility level? Road temperatures are critical around freezing What about correlations between some of the variables? Back to the raw data Where are the tipping points above or below which the regression modeling may be effective?
6	Tipping Points About 20% of the hours observed around the 4 stations had a crash
7	Tipping Points Precipitation Rate and Visibility
8	Tipping Points Wind Speed (average and gust)
9	Decision Tree Construction A set of regression models applied under specific conditions. Allows for evaluating continuous variables for regions of interest Evaluated subsets of data where crash risk was greater than 20% for all levels of other variables shown to be significant i.e. the effect of dew pt, visibility, wind speed when air temperature is < 0 deg C.
10	Crash Risk Algorithm
11	Equations For each path on the tree, a regression model was created as done originally. The exponential of the parameter estimate multiplied by the variable value yields the odds of a crash
12	Road Network Data OpenStreetMap (OSM) data were loaded into a database to comprise the road network Length or travel time the typical cost of a road segment
13	Crash Risk Aversion Algorithm
14	Road Segment Weather Interpolate weather data for the road network using inverse distance weighting (IDW) IDW not the most rigorous spatial interpolation method, but best choice with only 4 CLARUS stations Inverse distance weights, calculated from road segment centroid, stored in the database for each road segment
15	Crash Risk & Cost Calculation Classical shortest time problem, but with crash risk considered as part of the cost
16	pgRouting
17	Routing Web Service Apache server programmed in with the framework (and RESTful and AJAX-compliant) Client application written in Javascript using GeoExt (ExtJS); web mapping powered by OpenLayers Routing data sent in Javascript Object Notation (JSON)
18
19
20
21
22
23
24