Detection of 3D tree root systems using high resolution ground penetration radar

Abstract:

Knowledge of root systems and its distribution are important for biomass estimation as well as for the prevention of subsurface distribution network damages. Ground penetration radar (GPR) is a promising technique that enables a non-invasive imaging of tree roots. Due to the polarisation-dependent reflection coefficients and complicated three-dimensional root structure, accurate measurements with perpendicularly polarized antennas are needed. In this study, we show GPR data from two planes and one chestnut at two locations with different soil conditions. Perpendicular 10 x 10 cm grid measurements were made with a shielded 250 MHz antenna in combination with a high precision self-tracking laser theodolite that provides geo-referenced traces with a spatial resolution of ~ 2 cm. After selecting potential root hyperbolas within the perpendicular GPR profiles, the corresponding three-dimensional coordinates were extracted and visualized in planar view to reveal any linear structure that indicates a possible tree root. The coordinates of the selected linear structures were projected back to the surface by means of the laser-theodolite to indicate the locations for groundtruthing. Additionally, we interpolated the measured data into a 3D cube where time slices confirmed the locations of linear reflection events.
We validated the indicated predictions by excavation of the soil with a suction dredge. Subsequent georeferencing of the true root distribution and comparison with the selected linear events showed that the approach was able to identify the precise position of roots with a diameter between 3 and 10 cm and a depth of up to 70 cm. However, not all linear events were roots; also mouse channels were found in these depths, since they also generate GPR hyperbolas aligned in linear structures. Roots at a second location at depths of 1 to 1.20 m did not generate identifiable hyperboles, which was probably due to an increased electrical conductivity below 86 cm depth. The demonstrated approach is a promising tool for semi-linear root detection, whereas advanced 3D processing and migration is needed for more complicated root structures.