BALLAST FOULING ASSESSMENT USING 2 GHZ HORN ANTENNAS – GPR AND GROUND TRUTH COMPARISON FROM 238 KM OF TRACK Roger Roberts*, Imad Al-Qadi**, Erol Tutumluer**, Andreas Kathage* *GSSI, 12 Industrial Way, Salem, NH 03079; ** University of Illinois Urbana-Champaign, Urbana, IL, 61801 roger@geophysical.com , alqadi@ad.uiuc.edu , tutumlue@uiuc.edu , kathagea@geophysical.com ABSTRACT A study utilizing ground penetrating radar (GPR) horn antennas to evaluate railroad ballast, subballast, and subgrade condition was conducted between January 2005 and September 2006. Early in the study it was noted that the data from recently developed 2 GHz horn antennas indicated high sensitivity to the void space between clean ballast aggregate. The data from clean ballast contained significant scattering from the voids. Scattering from the dielectric contrast between the fines filling the void space and the aggregate in fouled ballast was very subdued. This important feature formed the basis for development work focusing on data processing methodologies to extract this information from the data. A representative scattering amplitude envelope was constructed from the data and implemented in an automatic data processing sequence. Subsequently, data from 148 mi (238 km) of track obtained with 2 GHz horn antennas was processed and the ballast fouling condition was automatically interpreted from the GPR data. The data were collected on four different railroad tracks located primarily in the states of Massachusetts, Nebraska, Colorado, and Wyoming in the United States. Ground truth data consisting of mud spots, ballast cleaning records and limited sampling have been compared to the processed GPR data. The comparison shows good agreement between automatically-interpreted GPR data and ballast condition assessed via ground truth. INTRODUCTION A considerable amount of work has been performed using GPR to assess railroad track substructure, especially in the past few years (Al-Nuaimy et al., 2006; Kathage et. al., 2006; Keogh et al., 2006, Hyslip et al., 2005). Data are commonly obtained with antennas ranging in center frequency from 400 MHz to 1 GHz. These investigations are focused on locating reflections which are generally associated with boundaries between the either the ballast-subballast or subballast-subgrade layers. Layer-tracking algorithms are typically used to extract this information from the data (Al-Nuaimy et al., 2006; Eriksen et al., 2006). The subsequent processed data are then reviewed and edited at locations where the layer- picking algorithm becomes confused due to reflections diverging, weakening, or disappearing. Considerable information can be obtained regarding the condition of the track substructure based on the amplitude and arrival times of these reflections. For example, very high amplitude reflections may indicate the presence of high moisture (Hyslip et al., 2005). And sudden increases in arrival times can infer the presence of ballast pockets (Hyslip et al., 2005). Two studies of the electromagnetic properties of clean and fouled ballast (Clark et al., 2001 and Sussmann et al., 2002) clearly show an increased permittivity associated with fouled ballast. Keogh et al., (2006) use this information to infer ballast fouling. They calculate the ballast permittivity using a velocity-depth inversion of reflection travel times from multiple offset antenna data. Recently developed 2 GHz horn antennas have been tested and were found to provide data that are very sensitive to the scattering from void space in clean ballast (Roberts et al., 2006 1 ; Roberts et al., 2006 2 ). The sensitivity was so pronounced that a processing scheme was developed which could automatically obtain the void scattering amplitude envelope and measure the decay of the envelope versus depth. Data