Long-term multi source analysis for asphalt binder PG selection using deep learning high air temperature modelling

Abstract

As asphalt binder is a temperature-dependent viscoelastic material, the asphalt pavement performance is affected by the temperature during its service life. Traditionally, selecting the asphalt binder performance grade (PG) is based on long-term meteorological data analysis, but this study used MODIS and ASTER remotely-sensed (RS), reanalysis ERA5-Land, and in-situ meteorological datasets to select the PG. The multiple linear regression (MLR), genetic algorithm (GA), and deep-learning (DL) techniques were used to create models to estimate the maximum air and road pavement surface temperatures, considering the Superpave specifications. Model parameters involved the land surface temperature, vegetation index, elevation, ERA5-Land air temperature, soil moisture, wind speed, thermal radiation, evapotranspiration, latitude, and climate type. Comparative analyses revealed that the DL model yielded the most reliable PG results and the proposed RS-based method was more accurate than conventional approaches and could determine the PG accurately with 1 km resolution independent of distance to meteorological stations.