Air Temperature Prediction Models for Pavements Based on the Gene Expression Programming Approach

Abstract

As per the performance grading scheme, the selection of asphalt binder for a particular location requires information on seven-day maximum and one-day minimum pavement temperatures. Pavement surface temperatures are usually related to the surrounding air temperature. This study presents a methodology for developing air temperature predictive models using high resolution long-term weather data of India. Gene expression programming (GEP), an evolutionary computing algorithm, was used to evaluate the expressions governing the air temperature as a function of latitude, longitude, elevation, relative humidity, wind speed, solar radiation, and rainfall intensity. A new methodology to evaluate the optimum tree depth for achieving reasonably high accuracy but at reasonably smaller tree depth was also proposed. Statistical analysis involving comparing the goodness of fit and distribution of the prediction error was conducted to understand the prediction capability of the proposed models. The statistical analysis proved the reasonably high predictive power of the gene expressions corresponding to the optimum tree depth. The proposed seven-day maximum and one-day minimum air temperature predictive models have a very simple structure that can be used by field engineers for hand calculation with little effort.