Dataset: Zooplankton In Situ Videos

Name: In Situ Amphipod and Copepod Video Output Captured by the Hoodsport ORCA Profiling Mooring Mounted SPC-2 Zoocam in the Hood Canal, Puget Sound, Washington from August to September 2018 (Zooplankton Swimming project)
Published: 2024-05-21
Keywords: zooplankton, in-situ videos, camera, Puget Sound, hypoxia, oceans

View Data: Data not available yet

Cite This Dataset

Spatial Extent: N:47.421817 E:-123.112583 S:47.421817 W:-123.112583

Hoodsport, hood Canal, Puget Sound, WA, USA

Temporal Extent: 2018-08-01 - 2018-09-30

Project:

Causes and consequences of hypoxia and pH impacts on zooplankton: Linking movement behavior to vertical distribution. (Zooplankton Swimming)

Principal Investigator:

Julie E. Keister (University of Washington, UW)

Co-Principal Investigator:

Daniel Grunbaum (University of Washington, UW)

Scientist:

Paul Roberts (University of California-San Diego, UCSD)

Student:

Amy Wyeth (University of Washington, UW)

Contact:

Amy Wyeth (University of Washington, UW)

BCO-DMO Data Manager:

Sawyer Newman (Woods Hole Oceanographic Institution, WHOI BCO-DMO)

Version:

Version Date:

2024-05-21

Restricted:

Validated:

Current State:

Preliminary and in progress

In Situ Amphipod and Copepod Video Output Captured by the Hoodsport ORCA Profiling Mooring Mounted SPC-2 Zoocam in the Hood Canal, Puget Sound, Washington from August to September 2018 (Zooplankton Swimming project)

Abstract:

This dataset consists of videos of zooplankton swimming taken by an in-situ camera system (the SPC UW ZooCam) that was deployed on the Hoodsport ORCA profiling mooring in Hood Canal (Puget Sound), WA in summer 2018. Understanding zooplankton population dynamics is challenging, largely because traditional methods for quantifying zooplankton distributions are costly, limited in scope, and require extended analysis by trained analysts. We developed a novel methodology that combined remotely deployed camera systems, Machine Learning-based identification of zooplankton, and video-based tracking technology to quantify copepods’ and amphipods’ in situ swimming behaviors in a seasonally hypoxic and acidified fjord.

Expand/Collapse All

Description

The primary file of this dataset is a tabular file of data containing an overview of each in situ video included in this analysis, the nearest CTD cast to the recording location of that vide, and selected environemntal data from the nearest CTD cast.

Methods & Sampling

Videos were recorded only under low-flow (< 1 cm/s) conditions, using infrared lighting to avoid modifying zooplankton behavior. The video camera recorded swimming in the X (left, right) and Z (up, down) directions, observing true vertical motion and projected horizontal motion. Videos were recorded at an average frame rate of 20 frames per second. The frame rate varied slightly during videos due to data throughput limitations of the onboard, single-board computer. However, each video frame was precisely timestamped, and the exact difference in time between frames was used when converting speeds from pixels per frames to mm per second. Each frame was 650 x 876 pixels, and each pixel was 0.088 mm2, corresponding to a 57.2 x 77.1 mm field of view.

Problems/Issues

Many images are blurry or the organism was imaged at a non-oblique angle, making taxonomic classifications somewhat subjective, even for trained taxonomists. The classifications represent the best possible visual identification using a machine learning algorithm, and should not be considered definitive. If there were too many passive particles (marine snow) in a video, the particle tracking technology was not always able to stitch together reliable zooplankton swimming trajectories. In those cases videos were dropped from the analysis and not provided in this dataset.

Data Processing Description

Zooplankton and other particles appeared in videos as grayscale images against a dark background. Regions of Interest (ROIs) from each frame were identified using the Python package OpenCV (version 4.7.0) (Bradski 2000). OpenCV-fitted contours and rotated bounding boxes were used to calculate width, height, and area metrics for each ROI. The nominal “length” of each ROI was defined as the longer of its width and height measurements. Position, defined as the optical centroid of the zooplankter outline, and size metrics were saved for each ROI that were potentially zooplankton (areas larger than 30 pixels) within each frame.

Zooplankton positions were assembled into swimming trajectories using the Matlab software Tracker3D (Chan and Grünbaum 2010). Small particles in videos were used to reconstruct a background flow field with Particle Image Velocimetry (PIV), using the Python package OpenPIV (version 0.23.9) (Liberzon et al. 2021).

Each zooplankton ROI was assigned a taxonomic identification using a Machine Learning (ML) image classification algorithm. The algorithm was generated using a pre-trained convolutional neural network 50 layers deep (ResNet-50) with PyTorch, an open-source, Python-based machine learning framework.

BCO-DMO Processing

- Changed column name of video column to "vid_id" to match the "vid_id" columns of the supplemental files
- Changed date time format in nearest_ctd_cast from %m/%d/%y %H:%M to %Y-%m-%d %H:%M
- A human readable datetime column was created from the unix_datetimestamp value present in the filename, this column is named "datetime_utc"
- All video data has been compressed and uploaded to this data package in a zip folder called "fps_20.zip" the contents of this zip are represented in the primary tabular data file associated with this dataset

More Information about this dataset

Funding Sources

Funding Source	Award Number
NSF Division of Ocean Sciences	OCE-1657992

Deployments

Deployment	Synonyms	Start Date	Platform	Investigator
UW_SPC-2_Zoocam		2018-06-26	ORCA-UW-Hoodsport	Julie E. Keister (Chief Scientist)	data info

Instruments

Underwater Camera

Supplied Name: SPC-UW-Zoocam

Supplied Description:

The SPC-UW-Zoocam designed and built specifically for this project. The Zoocam was custom-built by Paul Roberts in the Jaffe Imaging Laboratory at the University of California San Diego. It is an underwater camera system with a 500-mL imaged area that captured still images while profiling using lighting in the visible wavelength range (images submitted as a separate dataset) and videos of zooplankton using IR lighting (while stationary and under low-flow conditions.

Instrument Type

Generic Name: Underwater Camera

Community Identifier: http://vocab.nerc.ac.uk/collection/L05/current/180/

Generic Description:

All types of photographic equipment that may be deployed underwater including stills, video, film and digital systems.

Parameters

Date and time formats are described in python strftime() syntax.

Supplied Name	Supplied description	Supplied Units	Standard Name
vid_id	Video identifier, also the timestamp of when the recording started in Unix epoch time	unitless	time_begin
datetime_utc	Datetime value derived from the unix timestamp represented in the image filename. Format: %Y-%m-%d %H:%M:%S	unitless	time_begin
total_frames	The total number of frames in the video, a way to measure the video length	frames	image_analysis
avg_frm_rate	The average frame rate of the video (frame rates varied slightly during videos)	frames per second	image_analysis
nearest_ctd_cast	Date and time of the nearest completed CTD cast to the recorded video Format: %Y-%m-%d %H:%M	unitless	cast
nearest_ctd_offset	Time difference between the nearest completed CTD cast and the video recording	unitless	cast
depth	The depth the camera was parked when the video was recorded	meters (m)	depth
oxygen	Oxygen concentration from the nearest CTD cast at the depth the video was recorded	milligrams per liter (mg/L)	O2
temperature	Temperature from the nearest CTD cast at the depth the video was recorded	degrees Celcius	temp

Status message

Database

Contribute Data

Dataset: Zooplankton In Situ Videos