USING AN ARTIFICIAL NEURAL NETWORK TO PREDICT PARAMETERS FOR FROST DEPOSITION ON IOWA BRIDGEWAYS

This paper investigates a new method for forecasting frost formation on Iowa bridgeways. A frost model developed by Knollhoff et al. (2001) predicts frost deposition based on moisture flux principles. The frost model requires 4 inputs: air temperature; dew-point temperature; wind speed; and surface temperature. An artificial neural network is used to predict these four inputs at 20-minute intervals for a 24-hour period. The output from the neural network models can then be used as input into the frost deposition model to predict frost formation on Iowa bridgeways. The proper development of an artificial neural network requires the dataset to be subdivided into at least a training set and a validation set. A test set can also be used to further test the model. Results showed that the predictions correlated well with the road weather information system observations, and generally perform better than output from nested grid model-model output statistics alone.