Deep learning modelling of long-term satellite remotely sensed data for low-temperature asphalt binder selection

This study has focused on selecting the asphalt binder performance grade (PG) using remotely sensed (RS) data including MODIS (Aqua and Terra), ASTER and FLDAS products. Three models have been created based on multiple linear regression and deep learning and used to estimate the annual minimum air temperature on the surface of the road pavement, considering the Superpave specifications. Parameters of the models include land surface temperature, latitude, vegetation index, elevation, snow cover fraction, soil moisture, evapotranspiration, wind speed, and climate type. The PG selection has been performed according to the proposed and three identified conventional approaches and comparative analyses have clarified that the proposed method yields reliable results in estimating the asphalt binder performance temperature and is more accurate than conventional approaches in determining the PG. According to the results, the proposed RS-based approach has the potential to determine the PG accurately in 1 km resolution.