# Coral Reef Soundscapes from a Global Restoration Programme (MARRS): HydroMoth Recordings and Sonotype Detections

## Facts

- Provider: Ben Williams and Kate Jones, University College London
- License: CC-BY-4.0 (https://api.figshare.com/v2/articles/29958062)
- Size: 1,481 files
- Format: Parquet (in sampled files)
- Stored at: Hugging Face, https://huggingface.co/datasets/bens-bots/marrs-global-coral-reef-soundscapes
- Access: public, readable without an account
- DOI: https://doi.org/10.5522/04/29958062
- Category: Environmental Science
- Source: https://rdr.ucl.ac.uk/articles/dataset/_b_Coral_Reef_Soundscapes_from_a_Global_Restoration_Programme_b_/29958062
- Page: https://scigantic.com/archive/9dfee2c7-a49a-5b24-8ef5-a4b403d6bedc
- Keywords: Environmental Science, bioacoustics, Parquet, Hugging Face

## About

Passive acoustic recordings from 45 tropical coral reef sites in Australia, Indonesia, Kenya, Mexico and the Maldives, collected between October 2021 and June 2023 to test whether reef restoration with the MARRS method (standardized rubble stabilization and coral out-planting) brings back ecosystem function. Sites are classed as degraded, healthy, early-stage restored (under 3 months) or mid-stage restored (32 to 53 months). A HydroMoth recorder at each site sampled on a 1-in-4-minute duty cycle (1-in-2 in Indonesia) for about a month, at 16 kHz, giving over a year of audio. This Hugging Face version, parsed into Parquet, stores 1-minute raw clips as Parquet per site (raw_ _ , e.g. raw_kenya_H1), 5-second clips of fish and invertebrate sonotype detections per country produced with a human-in-the-loop agile modelling pipeline (detections_ ), plus an embeddings/ folder; about 975 GB. Load with Hugging Face datasets or pyarrow. The original dataset is Williams and Jones (2025), University College London Research Data Repository, doi:10.5522/04/29958062, licensed CC BY 4.0. Cite it and Williams et al. (2025), bioRxiv doi:10.1101/2025.09.24.678197.

## Load the data

```python
from datasets import load_dataset
ds = load_dataset("bens-bots/marrs-global-coral-reef-soundscapes")
print(ds)
```

## Data structure

File types in sampled files: Parquet

Example file: `detections_australia/train-00000-of-00014.parquet`

## Open it

- In a hosted notebook with the data mounted: https://scigantic.com/archive/9dfee2c7-a49a-5b24-8ef5-a4b403d6bedc
- Full record as JSON (schema card included): https://api.scigantic.com/api/archives/9dfee2c7-a49a-5b24-8ef5-a4b403d6bedc
- Over MCP: `get_schema_card` with id `9dfee2c7-a49a-5b24-8ef5-a4b403d6bedc` at https://api.scigantic.com/mcp

## Related topics

- [Parquet datasets](https://scigantic.com/topics/parquet)
- [Environmental Science datasets](https://scigantic.com/topics/environmental-science)
- [Bioacoustics datasets](https://scigantic.com/topics/bioacoustics)
