Bridge and Roadway Frost. Occurrence and Prediction by Use of an Expert System

Eugene S. Takle Iowa State University, Ames, Iowa

Search for other papers by Eugene S. Takle in
Current site
Google Scholar
PubMed
Close
Full access

Abstract

A survey of the characteristics of frost occurrences on bridges and roadways derived from questionnaires completed by highway maintenance personnel and analysis of more than 4000 frost observations in Iowa reveal that bridge frost occurs up to 58 times annually in certain parts of the state and, roadway frost, as many as 35 times. Certain bridges or stretches of road seem frost-prone because of their location or because of adjacent features. An expert system designed to provide 20-b forecasts of roadway and bridge frost has been constructed from analysis of meteorological conditions and has been evaluated against human forecasters. The expert system, when supplied with perfect forecasts of commonly forecast meteorological variables, produced accuracy com- parable to or higher than human forecasters. Human forecasters were observed to provide relatively unbiased forecasts for bridge frost but were highly biased toward reducing false alarms for roadway frost. The expert system, by contrast, is configured so that the decision threshold can be adjusted to give unbiased forecasts or forecasts that are biased in either direction without significant degradation of accuracy. A suggestion is made for the use of such a system as a management tool for separating forecast accuracy from (possibly nonmeteorological) decision-threshold criteria.

Abstract

A survey of the characteristics of frost occurrences on bridges and roadways derived from questionnaires completed by highway maintenance personnel and analysis of more than 4000 frost observations in Iowa reveal that bridge frost occurs up to 58 times annually in certain parts of the state and, roadway frost, as many as 35 times. Certain bridges or stretches of road seem frost-prone because of their location or because of adjacent features. An expert system designed to provide 20-b forecasts of roadway and bridge frost has been constructed from analysis of meteorological conditions and has been evaluated against human forecasters. The expert system, when supplied with perfect forecasts of commonly forecast meteorological variables, produced accuracy com- parable to or higher than human forecasters. Human forecasters were observed to provide relatively unbiased forecasts for bridge frost but were highly biased toward reducing false alarms for roadway frost. The expert system, by contrast, is configured so that the decision threshold can be adjusted to give unbiased forecasts or forecasts that are biased in either direction without significant degradation of accuracy. A suggestion is made for the use of such a system as a management tool for separating forecast accuracy from (possibly nonmeteorological) decision-threshold criteria.

Save