How An Underwater Robot Maps the DDT Pesticide Dumpsites Off Los Angeles

Off Los Angeles, almost 900 m down, two old dumpsites sit in the San Pedro Basin. Earlier studies found DDT pesticide in the mud there, and DDT sticks to particles, so wherever the mud goes, the DDT can go too. In 2023, researchers at Scripps Institution of Oceanography mapped more than 350 km² of both sites with an AUV carrying synthetic aperture sonar.

A model learned to sort the sonar pictures into six kinds of seafloor and got 9 out of 10 right on examples it hadn't trained on. The map shows where the mud lies flat, and where burrowing animals have roughened it. The authors read that rough mud as more likely to be stirred up, and any DDT with it.

In this video

• The dumpsites: waste dumping off Southern California from the 1930s to the 1970s, 15 known offshore dumpsites, and the two in the San Pedro Basin that received DDT-laced acid sludge
• The survey: a HUGIN AUV flying about 30 m above the seafloor in overlapping lines, 10 dives of about a day each, more than three times the area of a 2021 sonar survey. A Navy ROV, steered through a cable, recorded 300 hours of video
• Synthetic aperture sonar: about 180 m to each side, many pings combined as if from one much longer sonar, details a few centimeters across, more than 20 TB of raw data
• Down close: one run at about 5 m above the bottom, from about 200 m deep on the Santa Catalina slope into the basin, with 220 photos and a laser that measured the mud's height. One 10 m strip had about 270 burrow holes and heaps
• Six kinds of seafloor, picked from the ROV video: flat mud, textured mud, rocks, hard-shelled animals, plow scars and mounds. A classifier learned them from 10 m sonar tiles, 50 examples of each, and got 9 out of 10 right. Flat mud was the hardest
• The map: flat mud across the basin floor, textured mud along the slopes, rocks to the south, plow scars near both dumpsites, checked against the camera's photos
• Roughness from physics: the sonar's echo pattern compared with the laser's direct measurement. They lined up well, though not perfectly
• The mystery mounds: hundreds along the southeastern edge of the second dumpsite. The authors suggest they could be waste dumped before ships reached the site, but their source and content are unknown
• What it doesn't show: the paper doesn't measure DDT or watch the mud move. The link between rough mud and moving DDT is the authors' reading of the map. Next, they want to try the method on rocky seafloor, coral and dense vegetation

For engineers

A Kongsberg HUGIN AUV carried a HiSAS 1030 synthetic aperture sonar (100 kHz center frequency, ~180 m range, ~4 cm along-track and ~2 cm across-track resolution) at 30 m altitude, with ~45 m overlap between tracks and INS navigation with USBL updates. Training labels came from the U.S. Navy's CURV-21 ROV video. Close-range data: a still camera and laser line scanner at 5 m altitude.

Each 10 × 10 m SAS tile became 412 features by 2D wavelet scattering, classified by a linear-kernel SVM trained on 50 examples per class: 90% overall accuracy in 10-fold cross-validation, with F1 from 85.1% (flat mud) to 92.8% (biota) (Table 2). For roughness, power spectra of the corrected SAS echo level were compared with the laser's mean square slope. Textured mud showed a spectral peak at 2.72 rad/m, flat mud none, and a power-law fit between the two gave an RMSE of 0.37 and a goodness of fit of 0.76 (Fig. 16).

The paper

"Dumpsite Seabed Characterization of the San Pedro Basin, CA, From Autonomous Vehicle Observations"
Ryan A. McCarthy, Dylan Kriegman, Sophia T. Merrifield, Heidi Batchelor, Eric J. Terrill
Scripps Institution of Oceanography, University of California San Diego, La Jolla, California
Journal of Field Robotics, July 2026, doi:10.1002/rob.70310. Open access, CC BY-NC-ND 4.0.

Read the paper

Supported by the NOAA Community Grant Program (award NA25NMFX469G0096) and the U.S. Office of Naval Research (grant N000014-22-C-2006). The U.S. Navy's Supervisor of Salvage deployed the survey vehicles.

This is an independent explainer. I'm not affiliated with or endorsed by the authors or Scripps, and the research is all theirs. The animations are my own drawings, not footage from the survey; the maps and sonar pictures are schematic. Any mistakes are mine.

I'm Bora Celik from Piccard. I explain new research in ocean science and robotics.