top of page
Search

Verifying the accuracy performance of the DJI L3 LiDAR scanner with the Emlid Reach RS4 Pro

  • Writer: Charlie Wijnberg
    Charlie Wijnberg
  • 1 day ago
  • 3 min read

Updated: 7 hours ago

DJI M400 drone with L3 LiDAR scanner
DJI M400 drone with L3 LiDAR scanner

Overview

3DroneMapping recently acquired the DJI M400 drone and Zenmuse L3 LiDAR scanner. The DJI Zenmuse L3 represents a significant advancement over the L2, offering:

  • Detection range exceeding 950 m (compared to ~450 m on the L2)

  • Up to 16 return pulses for superior vegetation penetration (compared to 5)

  • Higher pulse rate for increased point cloud density

  • Improved absolute accuracy specifications

While the technical capabilities of the L3 are well documented, 3DroneMapping elected to conduct an independent accuracy assessment under controlled field conditions. The objectives were to determine real-world end accuracy using the Emlid Reach RS4 Pro as a non-DJI base station, and to evaluate whether PPK post-processing yields superior results over RTK for LiDAR trajectory computation.

Equipment & Ground Control

All surveying was conducted at 3DroneMapping's calibration site in South Africa, covering an area of approximately 2 km.

The Emlid Reach RS4 Pro was selected as the GNSS base station. The RS4 Pro tracks full-band constellations (L1/L2/L5/L6), achieving 7 mm + 1 ppm horizontal accuracy with fixed RTK solutions in approximately 5 seconds. Its high-gain, low-noise antenna provides strong multipath resistance, making it well suited to open survey environments where signal quality directly affects trajectory accuracy.

A total of 8 ground control points and 25 independent check points were established across the site and surveyed using the Emlid Reach RS4 base and RS4 Pro rover.

Flight Parameters & Data Capture

NTRIP corrections from the RS4 base were passed via its internal WiFi hotspot to the DJI RC Plus 2 Enterprise Enhanced controller and forwarded to the M400 for real-time RTK positioning during flight. The RS4 was simultaneously set to log raw GNSS observations for PPK post-processing.

Flight parameters were as follows:

  • Altitude: 120 m AGL

  • Scan rate: 350 kHz

  • Pattern: Linear

  • Sidelap: 30%

Processing Methodology

All data was processed using the latest version of DJI Terra. Two separate projects were computed:

  • RTK only: Trajectory calculated using real-time RTK positioning observations

  • PPK: Trajectory recalculated using raw GNSS logs from the RS4 base station

No ground control points were applied during initial processing, and no manual smoothing or adjustments were introduced. This ensured a clean comparison of the two positioning methods prior to GCP correction. Point clouds were subsequently loaded into LiDAR360 and assessed against the surveyed control and check measurements.

Accuracy Results

Each dataset was adjusted to the 8 control points. The 25 check points were then tested independently against each dataset as an unbiased measure of final accuracy.

PPK vs RTK trajectory comparison

The PPK dataset produced a vertical RMSE of 1.0 cm against 25 independent check points, compared to 5.0 cm for the RTK-only dataset — a fivefold improvement in vertical accuracy. The maximum Z error was reduced from 9.4 cm (RTK) to 1.8 cm (PPK), with a significantly tighter distribution of residuals across the site.

This result is consistent with the known advantages of PPK processing. By computing the full GNSS trajectory after the flight using forward-backward smoothing, PPK is not subject to real-time link dropouts, signal degradation, or baseline limitations that can affect RTK positioning during flight operations. For LiDAR survey work, where trajectory accuracy is directly reflected in point cloud accuracy, PPK is the recommended positioning workflow.

Point Cloud Noise

One observation of note: significant point cloud noise was present across all scan patterns and scan rates tested. This manifested as outlier points above and below the primary surface, clearly visible in cross-section views of the data. The noise was present regardless of the positioning method used, indicating it is a characteristic of the scanner or the default Terra processing pipeline rather than a positioning artefact.

DJI Terra's built-in strip alignment and smoothing routines provide some level of noise reduction, however outlier points persisted in the final deliverables when using default processing settings.

For projects requiring high-accuracy terrain models, corridor surveys, or volumetric calculations, a dedicated noise filtering step in specialist software such as LiDAR360 or LAStools is recommended prior to producing final outputs.

Conclusions

The results of this assessment confirm that the Emlid Reach RS4 Pro is a capable and cost-effective base station for use with the DJI Zenmuse L3. Its full-band tracking and high-quality antenna provide positioning performance well suited to demanding LiDAR survey applications.

PPK post-processing produced substantially superior vertical accuracy compared to RTK-only observations, achieving a check point Z RMSE of 1.0 cm across a 2 km site. GCP adjustment remains recommended for survey-grade deliverables, and post-processing noise filtering should be considered standard practice for high-accuracy outputs from the L3.

3DroneMapping provides high-accuracy LiDAR and aerial survey services across Africa. For enquiries regarding UAV LiDAR, aerial mapping, or equipment accuracy assessments, contact us at info@3dronemap.com.


 
 
bottom of page