Research ArticleUNDERSEA ROBOTS

Toward adaptive robotic sampling of phytoplankton in the coastal ocean

See allHide authors and affiliations

Science Robotics  13 Feb 2019:
Vol. 4, Issue 27, eaav3041
DOI: 10.1126/scirobotics.aav3041
  • Fig. 1 Survey area and Chla distribution.

    (A) Map of the operational area with the three AUV surveys indicated [surveys 1 and 3 (southeast of Runde) and survey 2 (north of Runde)], whereas the ship-based sampling stations are denoted by numbers (stations 1 to 7). In the lower corner, a small map of Norway is shown with Runde marked in red. (B) Chla (OC4ME) product from Ocean and Land Color Instrument, a multispectral medium-resolution instrument on board ESA’s Sentinel-3. The image from 2 May shows the Chla patchiness in the coastal zone, some months before the surveys. Runde is shown encircled in black.

    Credit: ESA CC BY-SA IGO 3.0
  • Fig. 2 Environmental conditions from buoy data.

    Time series of (A) water temperature at four different depths, (B) wind speed and wind gust, and (C) significant wave height, measured at a buoy located some 14 km inshore of the AUV survey area. The entire month of June 2017 is shown, with gray background marking the period of AUV and ship measurements.

  • Fig. 3 Data from survey 1 (MODE 1.fixed) using fixed tracking of the 3D Chla surface.

    Asterisks mark a point where the AUV behavior transitions, as the depth adjustment is followed by increased [Chla]. This is also shown in Fig. 4. (A) Chla concentration versus traveled distance accumulated over ground. (B) Temperature and salinity curves from AUV and the R/V Gunnerus from the same area. (C) AUV depth versus traveled distance. Note the adjustment of survey depth as the AUV follows the 3D surface (in 200-m increments).

  • Fig. 4 Volumetric estimate and AUV path.

    (A) Volumetric representation (3D kriging) of the [Chla] isosurfaces after the initial survey phase (MODE 1) covering the sides of the volume. The AUV path is overlaid (black line), and N, E, and Z are labels for north, east, and depth, respectively. (B) Volume after the adaptive survey of the internal volume (MODE 2), rendering more of the internal [Chla] structure. The depth adjustment (marked with asterisk) in the interior made the AUV stay in the region of high concentration. (C) Side view of the same volume as in (B) to highlight the adaptive depth adjustments performed by the AUV. The color scale of the interior isosurfaces is affected by the transparency. Plots constructed in Mayavi (65).

  • Fig. 5 Survey 3 using undulating tracking control during the survey.

    (A) [Chla] versus traveled distance accumulated over ground. (B) AUV depth versus traveled distance.

  • Fig. 6 Comparisons of in situ [Chla] derived from the AUV (ECO Puck), FRRf profiling measurements, and in vitro [Chla] extracted from water samples.

    The differences in [Chla] at the surface between the two sensors are due to the hydrography of the two stations despite proximity.

  • Fig. 7 Sample imaging information from the SilCam obtained at stations 4 and 5.

    Data presented are averaged from both stations and discretized into three depth ranges from 0.5 to 10 m (top), 10 to 30 m (middle), and 30 to 50 m (bottom). The left column shows the particle number distribution from the two magnifications of SilCam that were deployed, with the montages of particles from the red shaded region presented in the images in the middle column of the figure and the blue shaded region in the images in the right column of the figure. The dashed line in the particle number distributions represents the average fitted Junge distribution (59) (between 100 and 300 μm) from all depths.

  • Fig. 8 A 3D conceptual view of our approach.

    (Left) After each yo-yo envelope, two SCM peak depths are found and the final SCM is determined. (Middle) The final SCM depth is then assimilated into the GP model’s 3D SCM surface. (Right) AUV transects (black dashed line), peak detection (black/white crosses), MODE 1 [red lines (box)], MODE 2 [blue lines (hourglass)], and true distribution (green blobs), with the 3D surface of the SCM depth shown as the tan surface.

  • Table 1 AUV survey information and details.
    SurveyDate and timeDurationMean velocity (m/s)Area (m2)Mode
    120.06.17, 10:02 a.m.1 hour 33 min1.5700Fixed depth
    220.06.17, 12:10 a.m.1 hour 58 min1.5900Fixed depth
    322.06.17, 11:06 a.m.1 hour 27 min1.6700Undulating ±2.5 m
  • Table 2 FRRf survey data gathered from stations 1 to 7, shown in Fig. 1A.
    NameAreaTide
    Station 1—19.06.17, 10:15 a.m.Survey 2Ebb
    Station 2—19.06.17, 13:55 a.m.Survey 2Low
    Station 3—19.06.17, 16:15 a.m.Between S2 and S1Rise
    Station 4—20.06.17, 10:30 a.m.Survey 1Ebb
    Station 5—20.06.17, 13:05 a.m.Survey 1Low
    Station 6—21.06.17, 10:10 a.m.West of S1High
    Station 7—21.06.17, 14:45 a.m.West and between S2 and S1Low

Supplementary Materials

  • Supplementary Materials

    The PDF file includes:

    • Fig. S1. LAUV platform.
    • Fig. S2. Experiment overview.
    • Fig. S3. Survey 2 data.

    Download PDF

    Files in this Data Supplement:

Stay Connected to Science Robotics

Navigate This Article