Drone Magnetometry Maps Coastal Magnetic Anomalies Off Cascais, Portugal

A drone-borne magnetic survey flown along the Atlantic coast of Cascais, western Portugal, has yielded one of the more candid accounts yet published of what unmanned aerial magnetometry can—and cannot—establish about the shallow subsurface. The study, published in Results in Engineering, documents an integrated workflow that combines low-altitude UAV magnetic acquisition with geostatistical modelling and automated anomaly clustering, while explicitly separating what its authors could verify from archived flight records from what remains dependent on processing files that were never preserved.

The retained evidence is substantial. Four MagDrone R3 acquisition records dated 9 June 2021 contain a total of 2,049,937 raw magnetic samples, acquired at a sampling frequency of 200 Hz, with GNSS position updates arriving at roughly 5 Hz. The georeferenced observations span approximately 38.67803–38.69592° N and 9.46047–9.41876° W, defining a coastal sector where urban development, transportation links and coastal engineering coexist with heterogeneous natural materials. The archived files also preserve triaxial accelerometer channels, latitude, longitude, GNSS altitude, satellite counts, GNSS quality fields and RTCM correction streams—a level of raw documentation that allowed the authors to re-verify the acquisition itself.

That verification mattered, because several figures from the original analysis could not be reconciled with the retained files. An earlier reported flight speed of 3.1 m s−1 could not be reproduced; trajectory-based checks of successive GNSS positions instead indicate typical in-motion ground speeds close to 9.8–10.0 m s−1. A previously reported total of 3,167,412 observations was replaced by the 2,049,937 samples actually present in the archive, and an unsupported 97.4% data-acceptance rate was dropped. The archived GNSS altitudes range from roughly 10.0 to 66.9 m, but these are GNSS-referenced heights, not sensor clearance above local ground, and the archive lacks the terrain model needed to confirm an earlier claim of terrain-draped flying at 3.8–4.6 m above ground level.

The survey used a DJI Matrice 600 Pro multirotor carrying a MagDrone R3 system with dual Bartington Mag-03MSES70 triaxial fluxgate magnetometers suspended beneath the aircraft on a non-magnetic mount. That configuration embodies a well-known trade-off in UAV magnetometry: increasing sensor–platform separation reduces electromagnetic interference from motors, batteries, power distribution and electronic speed controllers, but a suspended sensor can oscillate, introducing periodic magnetic variations of its own. The accelerometer channels in the archived records document appreciable dynamics, with standard deviations of approximately 0.075–0.095 g along X, 0.116–0.170 g along Y and 0.335–0.399 g along Z—direct evidence that motion-induced noise deserves attention alongside electromagnetic interference.

Processing of the raw data involved magnetic compensation, robust outlier detection based on the Median Absolute Deviation, temporal correction for geomagnetic fluctuations, and detrending to remove regional magnetic field variation. The resulting Total Magnetic Intensity dataset ranged from approximately 43,686 nT to 44,124 nT—a total range of about 438 nT across the surveyed area—reflecting genuine spatial heterogeneity in this mixed coastal environment rather than instrumental noise alone. The original analysis reported a residual-field standard deviation of roughly 28.4 nT after compensation, a figure the authors caution includes spatially varying magnetic signal, not merely measurement uncertainty, and should not be conflated with the separate error-budget quantity of 1.5 ± 0.5 nT whose component calculations could not all be reproduced.

The geostatistical core of the workflow used experimental variograms to characterize spatial continuity, with spherical, exponential and Gaussian models evaluated. The spherical model proved most satisfactory, yielding a nugget variance of approximately 18 nT², a partial sill of approximately 1247 nT², and a practical range of approximately 42.3 m—meaning magnetic observations remain correlated over distances of several tens of metres before becoming effectively independent. Directional variograms showed only limited anisotropy, so an omnidirectional model was used for ordinary kriging, which produced a continuous magnetic-field map of the sector. Cross-validation statistics preserved from the original analysis report a mean error of −0.3 nT, a root-mean-square error of 4.2 nT, a normalised RMSE of 1.5%, a mean standardised error of −0.07, a root-mean-square standardised error of 1.02, and an observed-versus-predicted correlation of 0.987.

On top of the kriged surface, the original analysis combined statistical thresholding with DBSCAN density-based clustering to flag spatially coherent anomaly candidates, reporting 4,712 anomalous nodes, 23 candidate groups and seven retained clusters, with a mean silhouette coefficient of 0.73. Here the archival honesty becomes pointed: the processed interpolation grid, the DBSCAN parameter file and the thresholded input table were not retained, so the exact anomaly threshold, the eps and min_samples settings, the distance metric, the scaling scheme and the bootstrap validation parameters cannot be recovered.

Subject of Research: An integrated UAV magnetometry workflow for magnetic anomaly detection and environmental assessment in coastal Cascais, Portugal

Article Title: An integrated UAV magnetometry workflow for magnetic anomaly detection and environmental assessment in Cascais (Portugal)

Article References: Muñoz, M. M., Garcia Vieira de Sá, A., Bos, M., & García, I. C. G. (2026). An integrated UAV magnetometry workflow for magnetic anomaly detection and environmental assessment in Cascais (Portugal). Results in Engineering, 32, Article 113542. https://doi.org/10.1016/j.rineng.2026.113542

Image Credits: AI Generated

DOI: 10.1016/j.rineng.2026.113542

Keywords: UAV magnetometry, Cascais, Portugal, magnetic anomaly detection, ordinary kriging, DBSCAN clustering, geostatistics, fluxgate magnetometer, coastal geophysics, reproducibility, environmental assessment, drone survey