Mapping Asphaltic Roads’ Skid Resistance Using Imaging Spectroscopy

Abstract

The purpose of this study is to evaluate a realistic feasibility of using hyperspectral remote sensing (also termed imaging spectroscopy) airborne data for mapping asphaltic roads’ transportation safety. This is done by quantifying the road-tire friction, an attribute responsible for vehicle control and emergency stopping. We engaged in a real-life operational scenario, where the roads’ friction was modeled against the reflectance information extracted directly from the image. The asphalt pavement’s dynamic friction coefficient was measured by a standardized technique using a Dynatest 6875H (Dynatest Consulting Inc., Westland, MI, USA) Friction Measuring System, which uses the common test-wheel retardation method. The hyperspectral data was acquired by the SPECIM AisaFenix 1K (Specim, Spectral Imaging Ltd., Oulu, Finland) airborne system, covering the entire optical range (350–2500 nm), over a selected study site, with roads characterized by different aging conditions. The spectral radiance data was processed to provide apparent surface reflectance using ground calibration targets and the ACORN-6 atmospheric correction package. Our final dataset was comprised of 1370 clean asphalt pixels coupled with geo-rectified in situ friction measurement points. We developed a partial least squares regression model using PARACUDA-II spectral data mining engine, which uses an automated outlier detection procedure and dual validation routines—a full cross-validation and an iterative internal validation based on a Latin Hypercube sampling algorithm. Our results show prediction capabilities of R2 = 0.632 for full cross-validation and R2 = 0.702 for the best available model in internal validation, both with significant results (p < 0.0001). Using spectral assignment analysis, we located the spectral bands with the highest weight in the model and discussed their possible physical and chemical assignments. The derived model was applied back on the hyperspectral image to predict and map the friction values of every road pixel in the scene. Combining the standard method with imaging spectroscopy may provide the required expansion of the available data to furnish decision makers with a full picture of the roads’ status. This technique’s limitations originate mainly in compositional variations between different roads, and the requirement for the application of multiple calibrations between scenes. Possible improvements could be achieved by using more spectral regions (e.g., thermal) and additional remote sensing techniques (e.g., LIDAR) as well as new platforms (e.g., UAV).