A visually persuasive radargram can still be technically weak if files are incomplete, geometry is wrong, axes differ or aggressive processing controls the apparent anomaly.
1. Inventory before interpreting
List every source file by survey area, line identifier, band or antenna and acquisition mode. Record trace count, samples per trace, sample interval, recorded time window, trace spacing and line direction. Reconcile the inventory with field notes and exported metadata.
Flag missing, duplicated, truncated, reversed or unexpectedly short files. Confirm that paired bands and cross-lines exist where expected. A processing workflow cannot restore missing coverage.
2. Validate structured data
For SEG-Y or another structured format, confirm that headers and trace records decode completely, sample values are finite and the declared sample format is correct. Compare derived line length and timing with the acquisition record. Retain the original files read-only and work from identified copies.
Screenshots and proprietary exports may assist review, but should not silently replace source traces. Record exactly which dataset supports each published figure.
3. Establish a minimal baseline
Begin with the least processing required to view the data. Review time zero, remove low-frequency drift where justified, and apply conservative gain. Preserve the recorded time axis unless a defensible depth conversion is available.
The baseline shows whether the response exists without heavy manipulation. Document every operation and parameter. Acquisition filters must be separated from post-processing and display choices.
4. Create a processing-control variant
Compare the baseline with a deliberately defined control variant, such as line-median background subtraction, while keeping line extent, time window and axes identical. Persistence across reasonable variants supports confidence that a waveform change was recorded. It does not identify the physical cause.
A feature that appears only after strong subtraction, filtering, migration or contrast enhancement is processing-sensitive. Its interpretation should be downgraded or qualified.
5. Apply corrections for a reason
- Time-zero adjustment: align the reference time using a stated method.
- Dewow: remove low-frequency instrument drift without suppressing target-scale responses.
- Gain: compensate display amplitude with time; never present colour intensity as severity.
- Background removal: suppress common horizontal components, while checking for removal or creation of meaningful behaviour.
- Migration: reposition suitable diffraction energy using an appropriate velocity model; it is not a cosmetic sharpening tool.
- Topographic correction: account for surface elevation where geometry requires it.
6. Control comparisons
Like-for-like panels should use the same coordinate direction, distance and time axes, anomaly bands and—where practical—colour scale. If each panel is individually contrast-scaled, state this prominently. Otherwise ordinary scaling differences may be mistaken for physical change.
Compare complementary bands as different measurements, not automatic confirmation. Review adjacent lines and cross-lines at their true intersections. A nearby line does not corroborate a response unless geometry and target scale support the comparison.
7. Preserve depth uncertainty
Radar measures two-way travel time. Depth conversion requires velocity, which varies with material and moisture. State how velocity was estimated—from a known-depth target, hyperbola fitting, common-midpoint analysis, core correlation or an assumption. Without calibration, use time or label depth as approximate and sensitivity-dependent.
8. Build an auditable output
Each figure should identify dataset, line, band, coordinate range, processing sequence and axis units. Archive the parameter record and the version of any script or software used. Do not overwrite the raw data. Record excluded steps such as migration, topographic correction or depth conversion so readers do not infer they were performed.
Minimum processing record
- Source file and checksum or immutable identifier.
- Acquisition metadata and geometry.
- Software and version.
- Ordered operations with parameters.
- Velocity and depth assumptions.
- Display scaling and colour treatment.
- Figure-to-source traceability.
- Known exclusions and data defects.