Decadal acoustic monitoring of toothed whales in the Gulf of Mexico post-oil spill

Data files

Nov 19, 2024 version files 1.93 GB

GOM_2010-2021.zip
1.93 GB
README.md
4.60 KB

Abstract

Shortly after the Deepwater Horizon oil spill began in April 2010, a widely spaced passive acoustic monitoring array was deployed in the northeastern Gulf of Mexico to document the impacts of this unprecedentedly large and deep offshore oil spill on oceanic marine mammals. The array was subsequently maintained for over a decade. Here we document decadal density declines for seven of eight monitored species groups, including sperm whales (up to 31%), beaked whales (up to 83%), and small delphinids (up to 43%). Declines were observed both within and outside of the surface oil footprint. Though not conclusively linked to the oil spill, the broad spatial and temporal scale of these declines observed for disparate marine mammal species is consistent with Deepwater Horizon impacts. These declines have exceeded and outlasted post-spill damage assessment predictions, suggesting that the offshore ecosystem impacts of Deepwater Horizon may have been larger than previously thought.

https://doi.org/10.5061/dryad.9zw3r22n4

Dataset Overview

This dataset supports the study of long-term changes in toothed whale densities in the Gulf of Mexico following the Deepwater Horizon oil spill, spanning 2010–2020. It contains passive acoustic detection records from five monitoring sites across various depths and regions, categorized by species and location.

Objective: To provide high-resolution, time-series data on the presence and density of multiple odontocete species in the northeastern Gulf of Mexico.
Monitoring Sites: Five locations with varied distances from the Deepwater Horizon spill site, including both deep and shallow water environments.
Species Monitored: The dataset includes detection events for a range of species, such as sperm whales (Physeter macrocephalus), beaked whales (Ziphius cavirostris), and delphinid species, categorized by frequency classes (e.g., high- and low-frequency delphinids).
Time Span: Continuous monitoring data from 2010 to 2020, allowing for analysis of trends and seasonal patterns in species presence and density.
Data Contents:
- Detection Labels: Classified odontocete detections with associated likelihood scores.
- Recording Effort: Daily effort metrics for each monitoring site to contextualize the detection data.
- Output: Organized time-series estimates of species presence by site, supporting density and trend analyses.
Potential Applications: The dataset is suitable for studies in long-term marine ecosystem health, impacts of oil spills on marine mammals, and passive acoustic monitoring methodology.

Description of the data and file structure

Detection Labels

Detection labels are provided in the Labels folder in NetCDF-compatible MATLAB .mat files. File names include the following information:

(Project) (Site) _(Deployment)(disk)
or

(Project) _ (Site) _ (Deployment) _ (disk)

Project names GofMX and GOM are used interchangeably.HARP data is typically stored across 1-16 disks within each deployment, but the total number varies, and occasionally a disk fails and is skipped during the recording process, leading to missing disk numbers.

Each file in the Labels subfolder contains a vector zID in which each row represents an acoustic detection. Three columns represent:

Detection event timestamp as a MATLAB date number (number of days since January 0, 0000).
Detection label, a number, which can be associated with a species ID using the associated row of mySpID.
Detection label score. A Softmax score provides some information on the likelihood of the quality of the label.

Values near 1 indicate that the classifier scored this event as more likely to be correct. Labels with scores near 0 are less likely to be correct.

Species codes are provided in the mySpID variable and include:

Gg: Grampus griseus
Kspp: Kogia spp.
Md: Mesoplodon densirostris
Me: Mesoplodon europaeus
Pm: Physeter macrocephalus
UD: Unidentified delphinid
UD_LF: Low-frequency delphinid
Zc: Ziphius cavirostris
SS: Snapping shrimp
Sonar: Echosounders
Ship: Ship noise

Effort

Daily recording effort information is provided for each site in .mat files in the Effort folder, with site codes in the file names (_Effort.mat) Two variables include:

Effort: A 2-column matrix where column 1 is the start of an effort period, and column 2 is the end of an effort period. Rows represent successive recording effort periods at a single site. Numbers are Matlab date numbers represented as the number of fractional days elapsed since 00/00/0000.
allEffort: A 2-column matrix where each row represents a day. Column 1 gives the index of that day over the study period, and column 2 gives the recording effort within that day at the specific site. The study period is from January 1st, 2010 to September 20th, 2021. Days may have partial effort, which may affect the total number of detections recorded on that day.

Note that recording effort differs between sites.

Sharing/Access information

Code is available on GitHub https://github.com/MarineBioAcousticsRC/2010-2021_timeseries_analysis. Output from the code using this data reproduces density estimates and figures in the publication, with additional figures and tables indicating seasonality and diel patterns of occurrence by species.

Passive acoustic recordings were collected using High-frequency Acoustic Recording Packages (HARPs) at five monitoring stations between May 2010 and March 2020 (Table S1). The total recording duration across the five sites amounted to 37 instrument-years of continuous data sampled at 200 kHz. Sites include three deep sites and two shallow shelf sites that monitor primarily shallow water species.

Echolocation clicks and other impulsive signals from odontocete species were detected using an impulse detector and then classified into species groups based on spectral characteristics and inter-click intervals. Automated detection and semi-supervised clustering techniques identified signals from target species, allowing for group-based density estimation of each species across time. Processing steps included high-pass filtering, calibration of hydrophones, and applying a neural network classifier to attribute signal identities. Acoustic presence data were converted into local density estimates, accounting for detection probability, group size, vocal probability, and false detection rates to analyze long-term trends. This approach allowed for robust monitoring of marine mammal density changes in the region

The dataset was collected using a passive acoustic monitoring array, deployed in the northeastern Gulf of Mexico shortly after the 2010 Deepwater Horizon oil spill and maintained for over a decade. High-frequency Acoustic Recording Packages (HARPs) were placed at five monitoring sites, capturing echolocation clicks and other impulsive signals from odontocete species, which were then classified into species groups based on spectral characteristics and inter-click intervals. Automated detection and semi-supervised clustering techniques identified signals from target species, allowing for group-based density estimation of each species across time. Processing steps included high-pass filtering, calibration of hydrophones, and applying a neural network classifier to attribute signal identities. Please see the associated manuscript for additional details:

Frasier, K.E., Kadifa, M.A., Solsona Berga, A., Hildebrand, J.A., Wiggins, S.M., Garrison, L.P., Frouin-Mouy, H., Gracia, A., Serrano, A., Hodge, L.E.W., Wall, C.C., Le Hénaff, M. & Soldevilla, M.S. (202X) A decade of declines in toothed whale densities following the Deepwater Horizon oil spill. Nature Communications Earth and Environment.