Getting usable scans out of a GPR rig is mostly about patience and knowing where it lies
I've been doing subsurface scanning for about fifteen years across utility locates, concrete inspection, and archaeological surveys. Most people pick up a GPR unit and immediately get frustrated because the output doesn't look like one of those clean images in the promotional materials. The raw data looks like noise. That is normal. The trick is learning to read what is actually there before you start making decisions off a scan. You need a few things before you even power the system on. A good GPR unit with interchangeable antennas, a tracking device if you want georeferenced results, a laptop running the acquisition software, and an idea of what you are looking for. The last point matters more than most beginners realize. If you approach a site without a hypothesis about what might be down there, you will spend hours collecting data you cannot interpret later. Antenna frequency is the first decision. Higher frequencies give better resolution but shallow penetration. Lower frequencies dig deeper but blur everything. A 500 MHz antenna will see rebar in a six-inch slab. A 100 MHz antenna might reach a void three meters down but will not distinguish individual pipes. I usually carry a dual-antenna setup — a 900 MHz for shallow detail work and a 250 MHz for anything below about a meter. That covers most job-site scenarios without switching gear every twenty minutes.
Setup procedure is straightforward. Power on the unit, let it initialize, run the manufacturer's calibration routine, then verify timing offsets with a known target if possible. I always test on a section of pavement where I know there are dowels or conduit embedded. Ten minutes of verification saves an afternoon of confusion later. The actual scanning process involves pulling the cart along parallel lines. Keep the wheels on the ground. Don't lift it between passes. Maintain consistent spacing between lines — usually half a meter to one meter depending on your resolution requirements. Mark your start and end points. Note surface conditions. Wet ground kills signal faster than anything else. I had a job once where the ground had been sprinkled the night before and I lost about sixty percent of my depth penetration compared to the dry patch I'd already scanned. I just marked that section and came back three days later when it had dried out.
Processing the data is where most people fail
Raw radargrams are nearly useless without processing. The standard pipeline runs through background removal, gain application, filtering, and migration. Background removal subtracts the common midpoint trace to eliminate horizontal banding caused by system noise and constant reflections from the surface layer. Gain compensates for signal attenuation with depth. You want hyperbolas to appear as clean arcs, not fading into the noise floor. Filtering is where judgment comes in. Bandpass filters remove frequencies your antenna isn't capable of resolving anyway. But I have seen people slap a aggressive median filter on their data and wipe out legitimate targets in the process. Start conservative. Apply a light filter, review, then escalate only if noise is genuinely drowning your reflections. Migration collapses those hyperbolas back into point sources. Without it, a single pipe shows up as a broad arc that is much wider than the actual object. Migration corrects for that. Most modern software does this automatically, but the defaults are often too aggressive. I usually set migration velocity to match the estimated dielectric constant of the material and then check a few test targets to confirm the hyperbolas are properly collapsed before running the full dataset.
Get the Full Details
A specific problem I ran into and how I handled it
Last year I was scanning a parking lot for abandoned drain lines using a 400 MHz antenna. The radargrams showed a series of strong linear reflections at about forty centimeters depth. looked like conduit or pipe running east-west across the site. I flagged those for the client and marked them on the site plan. A week later the excavation crew hit something that wasn't on my plan. A large concrete footing from a demolished structure that I had completely missed. I went back and reviewed the data. The footing was there. It was just buried under a thick layer of asphalt overlay that had different dielectric properties than the underlying soil. The contrast between the overlay and the footing was too low for my antenna to resolve clearly against the clutter from the asphalt itself. The workaround was to switch to the 200 MHz antenna and run slower pass lines with tighter spacing. The lower frequency penetrated through the asphalt layer better and the increased line density gave me enough spatial sampling to distinguish the footing from the background noise. It took longer — roughly triple the scan time for that zone — but the resulting map was accurate. I now always run a secondary lower-frequency pass when I suspect there might be layered materials with similar dielectric properties. It adds about twenty percent to the field time but prevents costly misses.
What the manuals don't tell you
Dielectric constant variation is a silent accuracy killer. Most GPR software assumes a uniform dielectric constant across your entire survey area. Real ground is not uniform. A patch of gravel next to compacted clay next to a filled trench will each have different velocities for the radar signal. If you set your velocity based on one zone and apply it everywhere, your depth estimates will be wrong in the other zones. I usually take three to five time-domain reflectometry measurements across different material types in the survey area and adjust my processing velocity for each zone separately. Depth accuracy improves from roughly plus or minus fifteen percent to plus or minus five percent that way. Another thing nobody warns you about: rebar in concrete creates such a strong return that it dominates the display. If you are scanning a slab to find voids or post-tension cables underneath, the rebar network can mask everything below it. The solution is to use a higher frequency antenna — 1 GHz or above — which gives you enough resolution to image between the rebar grid lines without penetrating deeply. You are essentially accepting that you won't see below the rebar plane and focusing on the shallow details instead.
Limitations you need to accept
Ground penetrating radar does not work in highly conductive materials. Saturated clay, saline soils, and wet fill will attenuate the signal within centimeters. I have walked away from sites where the ground was so clay-heavy that the maximum penetration was twelve centimeters regardless of antenna frequency. In those cases, you are limited to very shallow utility locating only. Electromagnetic induction methods work better in those conditions. Signal interference from power lines, rebar mesh, and even metal garbage cans left near the scan area can create artifacts that look like subsurface features. I learned to always do a quick pass without the cart — just walking the area and noting obvious surface contaminants — before committing to a full survey. Things that should have been obvious but weren't, like a row of steel fence posts running along one edge of the site, show up as bright hyperbolas that beginners can easily mistake for buried pipe. The equipment is expensive to buy and expensive to repair. A dropped cart on rocky terrain can knock out a transducer. Antenna cables fail at the connection points. I have replaced two 900 MHz antennas in four years from drop damage and one from water ingress after a sudden rain shower. Budget for maintenance.
If you need absolute depth certainty, especially for engineering or legal purposes, corroboration with a test pit or drill sample is still the only way to verify what the radar is showing. I treat every GPR survey as interpretive unless I have physical confirmation at a representative sample of locations. That habit has saved me from signing off on inaccurate maps more times than I care to count.
