Laserfiche WebLink
Jennifer Wiltgen <br />Minnesota Department of Transportation <br />June 25, 2021 <br />Page 5 <br />Link Root Mean Squared Error (RMSE) <br />The average error of the model can be estimated in two ways. The first, the average arithmetic error, <br />compares the average modeled volume with the average count volume. However, analysts frequently <br />use a more restrictive measure called Root Mean Squared Error (RMSE) that does not allow a too - <br />high volume to be averaged and offset against a too -low volume. Furthermore, error on higher volume <br />facilities is weighted more heavily. It is often helpful to compare the error to the magnitude of an <br />observed traffic count. To do so, the RMSE is normalized by dividing it by the mean count. This new <br />variable is referred to as Normalized RMSE or Percent RMSE. For this project, an RMSE of 19 <br />percent with an average count of 63,000 vehicles per day was obtained. As shown in Figure 2, this <br />value is reasonable given the standard tolerates higher percentage errors with low numeric differences <br />(a result of typical count deviations on low -volume roads and the low impact of those deviations on <br />forecast lane requirements) . Model performance for low volume roads is affected by the resources <br />available to refine the geographic size of model zones, the roadway geometrics and traffic control <br />parameters in the model. <br />Figure 2: Maximum Desirable Error for Link Volumes <br />Percent Deviation <br />15 25 35 45 55 65 75 85 95 <br />Base Year Count(1000's) <br />Source: Model Validation and Reasonableness Checking Manual (FHWA, 1997) <br />Link Deviation from Count <br />60 <br />50 <br />40 <br />30 <br />20 <br />10 <br />0 <br />5 <br />Differences between modeled volumes and ground counts are expected in a model. The significance <br />of any difference depends on whether the difference affects roadway requirements, such as the <br />number of lanes, and considers variance in traffic counts. For this analysis, the model attained a 19 <br />percent RMSE, which necessitates applying an industry -standard post -model adjustment process. As <br />shown in Figure 3, high volume roadways had the best fit; some roadways, particularly low volume <br />roadways, exceeded the desirable deviation. <br />