Tech guide — Processing
GPR Data Processing Steps: Gain, Filters, Migration
From raw traces to a readable radargram: what each processing step does, the order it runs in, and how it can mislead.
Quick AnswerGPR data processing cleans and corrects raw radar traces before anyone reads them. Time-zero sets the surface, dewow and band-pass filters remove drift and noise, background removal strips banding, and gain restores deep echoes. Migration then uses the measured wave speed to collapse arches before time is converted to depth, and the raw data is always kept. For the instruments behind the data, see more on EM Cable Locator vs GPR.
What is GPR data processing, and why does raw data need it?
GPR data processing is the chain of corrections and filters that turns raw radar traces into an image where real targets stand out and sit at the right depth. Ground penetrating radar records each echo as a trace in time. A raw radargram is dominated by a strong direct wave, slow drift, antenna banding and echoes that fade with depth. Processing adds no information; it reorganises what was recorded so that it can be read and measured.
Every step is a choice, and a poor one can hide a target or invent one. Good practice, now spelt out in the 2022 revision of PAS 128, is to keep the raw data, log every step and have trained staff do the work.
In what order are GPR processing steps applied?
Clean first, then amplify, then use the wave speed to collapse arches and convert time to depth, and only then interpret. Order matters: gain applied before drift removal amplifies the drift too. The flow below suits utility work.
GPR processing steps at a glance
| Step | What it corrects | Risk if overdone |
|---|---|---|
| Time-zero and DC shift | Sets the trace start at the ground surface; removes a constant offset | Wrong zero shifts every depth by the same amount |
| Dewow (low-cut) | Removes slow drift that pulls traces off the zero line | Too wide a cut distorts shallow echoes |
| Band-pass filter | Removes noise outside the antenna's frequency band | Too narrow a band causes ringing and smears detail |
| Background removal | Strips horizontal antenna banding and the direct wave | Also strips any real flat reflector, such as a pipe running along the line |
| Gain | Restores echoes weakened by depth and spreading | AGC flattens true amplitude; heavy gain lifts noise into "targets" |
| Velocity analysis | Finds the wave speed from arch shapes or known depths | One velocity used across mixed ground gives wrong depths |
| Migration | Collapses arches to points; puts dipping layers in place | Wrong velocity leaves "smiles" or "frowns" |
| Depth conversion and topography | Turns two-way time into depth; corrects for slopes | Ignoring slope bends flat services on the drawing |
In grid surveys the same flow runs on every line before the lines are stacked into a volume and cut into depth slices; see: 3D GPR vs 2D GPR for when that is worth it.
How do the different gain functions behave?
A gain function multiplies each sample by a factor that rises with time, and the curve shape decides whether the image stays faithful.
- Constant gain: one factor for the whole trace, which preserves relative amplitude but is rarely enough alone.
- Spreading and exponential compensation (SEC): a gain matched to the ground's attenuation that keeps the strongest reflector strongest, which helps tell metal from plastic.
- Automatic gain control (AGC): scales each short window to a similar level, showing faint deep echoes but destroying the amplitude differences that carry meaning.
A sensible routine produces an SEC version for interpretation and an AGC version as a search aid.
Which filters clean the data, and what can they remove by mistake?
Dewow and band-pass filters remove noise outside the useful band, while background removal strips banding and is the step most likely to delete something real. Dewow subtracts a running mean along each trace. A band-pass filter keeps roughly the transmitted band, about 200 to 600 MHz for a 400 MHz antenna.
Background removal subtracts the average trace over a window, so ringing vanishes and point targets stand out. A pipe running along the scan line looks like a flat band, however, and it vanishes too. Scan lines are therefore laid across the expected route, and processed images are checked against raw ones before a section is declared clear.
Noise we meet most often on Indian sites
Reinforced concrete roads produce repeating multiples between bars and surface, and wet roads after rain add ringing near the top. Mobile towers and heavy electrical plant inject spikes, while hand-pushed carts bounce on broken paving. Filtering handles most of this, yet a repeat line at another time or on a slightly shifted path often settles a doubtful target faster.
What does migration do, and why does it depend on velocity?
Migration moves the energy of each arch back to its apex, so a pipe appears as a point at its true position. Antennas see sideways as well as down, which is why a small target draws a hyperbola. Kirchhoff or frequency–wavenumber (F-K) migration undoes that spread, provided the wave speed is correct.
With the right velocity, the arch collapses to a tight spot; too low leaves it partly intact, and too high turns it into a "smile". Velocity comes first, from hyperbola fitting or a service at measured depth, and the depth conversion reuses it, so an error shows twice. Our field checks on that figure are in read GPR Depth Accuracy. On sloping sites, a topographic correction then keeps a level pipe drawn level.
What should a processed GPR deliverable include?
The interpreted drawing, the processing log, radargram images at doubtful points and the untouched raw data let anyone retrace each mark. A one-page note listing antenna, time window, trace interval, velocity and filter settings is enough. Read beside existing records, this utility data shows designers where underground utilities run before civil engineering work begins. For the annotated images we attach, the Mezux guide: GPR Radargram Library shows examples from Indian sites.
Processing also feeds each line's quality level. A line seen on raw data and two processed versions, with a matching locator trace, earns a higher grade than one that appears only after heavy gain. Such grading lowers the chance of accidental utility strikes and service disruptions. The method behind these utility maps is part of SUE survey explained simply, which helps reduce risks on construction projects.
What does a processed GPR survey cost?
Processing and interpretation are part of our per-metre GPR rate: ₹24 per metre standard and ₹32 per metre with an L-section, GST extra. Orders start at ₹50,000 in Delhi NCR and ₹1,00,000 in other states. On large programmes the whole order is reduced by 25% beyond 100 km, 40% beyond 500 km and 50% beyond 1,000 km. NCR scans begin on the booking day or the next working day. In other cities, an operator who lives nearby takes the job at once or within a working day, and mobilisation or travel, if it applies, is priced in the quote. If you want the raw files and processing log, say so at the start; share the site location and the deliverables you need.
Sources
- R. Drews, “GPR Processing”, Advanced Geophysics course notes, University of Tübingen (DC shift, dewow, gain, topographic correction). scripts.guz.uni-tuebingen.de
- ASTM International, D6432-19, “Standard Guide for Using the Surface Ground Penetrating Radar Method for Subsurface Investigation”. store.astm.org
- University of British Columbia, EOSC 350 “Geophysics for Practising Geoscientists”, GPR notes (velocity, CMP/WARR soundings, display). www.eoas.ubc.ca
- BSI, PAS 128:2022 “Underground utility detection, verification and location. Specification” (updated application of post-processing in detection surveys). knowledge.bsigroup.com
Frequently asked questions
Does processing change what GPR found?
It should not create or delete targets, but aggressive filters can. That is why raw data is kept and processed images are compared with it.
What is dewow in GPR?
A low-cut filter that removes slow drift along each trace, so the echoes sit around a zero line.
Is AGC gain bad?
It is limited, not bad: a good search aid for faint targets and a poor basis for comparing echo strength, so use SEC or linear gain for interpretation.
Why does migration need the wave speed?
The shape of each arch depends on velocity, so collapsing it needs that same value; a wrong one leaves arches or smiles.
Can I get the raw GPR files?
Yes, if agreed at the start. Raw files plus a processing log let another specialist review the work.
Where are SEC, F-K and time-zero explained?
Short definitions are listed in our glossary; see details on GPR, SUE & Land Survey Glossary terms.
Get a written quote with processing
GPR technique
Written by Saniya Parveen · Technically reviewed by Rahul Rai · Updated 8 November 2026