Background & Summary

Given the speed with which the Earth is losing biodiversity through human activities and climate change1,2,3,4, it is crucial that we map and monitor the structure, function, and biodiversity of ecosystems as quickly as possible to set baselines and to study and better understand these changes. Most of these data occur at relatively fine spatial scales, but losses accumulate over broad spatial scales and form the basis for international biodiversity targets (e.g. Intergovernmental Science-Policy Platform on Biodiversity and Ecosystem Services, Convention on Biodiversity). Ground observations are costly and time-consuming, making it challenging to collect data at broad spatial scales that can be replicated across the planet’s diverse terrestrial habitats5. One solution is to use remote sensing data from airborne or spaceborne instruments, but these data cannot identify most species, apart from sufficiently large plants and animals6. Ideally, these large datasets should be paired with contemporaneous in situ datasets representing a wide spectrum of biological diversity7,8.

Many animals emit sounds9 that have evolved to be heard within a soundscape of naturally produced noises like wind, thunder and rainfall10. Anthropogenic noise pollution is also increasing and impacting biodiversity11. Passive acoustic recording devices (autonomous recording units, ARUs) offer a scalable and innovative approach to biodiversity assessment in the form of non-invasive surveys, particularly useful for monitoring elusive or nocturnal species, and detection of natural resource exploitation, such as chainsaws or gunshots e.g.12. A new and inexpensive generation of ARUs allow researchers to expand bioacoustics into an array of applications, from monospecific density assessments at specific sites13, to documenting species presence and behaviour, particularly for bird monitoring14, to ‘soundscape ecology’ comprising animal-produced sounds (biophony), noises from the landscape (e.g. rain, wind and thunder; geophony), and human-derived sounds (anthrophony - incorporating technophony), across landscapes15,16. The combination of biophony, geophony, and anthrophony is recorded in any single ARU deployment, thus analyses of these components together to monitor landscape-level changes in biodiversity presents an exciting proposition.

BioSCape is a collaboration between the US and South Africa, supported by the United States government (NASA), the South African government (National Research Foundation, NRF), the South African Environmental Observation Network (SAEON), and the United Nations Educational, Scientific and Cultural Organization (UNESCO). The project integrates remotely sensed UV/visible to short wavelength infrared (UVSWIR) and thermal imaging spectroscopy and laser scanning (LiDAR) with a suite of surface measures of biodiversity. The BioSoundSCapes project is a BioSCape sub-project seeking to relate ground-based measures of acoustic diversity (biophony, geophony and anthrophony) to the contemporaneous spectral and structural imaging obtained by NASA remote-sensing instruments. This research took place in the Greater Cape Floristic Region (GCFR), South Africa, a global biodiversity hotspot experiencing significant anthropogenic and climate-driven challenges17.

The BioSoundSCape project used a large-scale, multi-season (dry and wet) deployment of ARUs to record bioacoustic signals across the GCFR, sampling an area of roughly 460 hectares spanning an extent of 119,058 km2. The GCFR is home to the Fynbos biome, world-renowned for its botanical richness18. The region is also home to numerous fauna, many of which vocalize, e.g., birds, amphibians, and insects. The region harbours eight endemic bird species19. Regions that have maintained and generated biodiversity over evolutionary time are especially important to conserve20 and may be crucial to species’ survival into the future21. The GCFR has been crucial in this respect for amphibians, acting as both a refugium and centre of speciation22,23, with severe impacts of climate change predicted24. The GCFR has been considered to contain a globally significant amphibian composition25, with 61 species and high endemism (62% of species are endemic to GCFR). As much as 68% of the GCFR’s Amphibia are considered threatened and 7% are data-deficient or not yet evaluated26. Both bird and amphibian taxa give insight into habitat quality, but frogs may be sensitive indicators27,28. Birds can be useful indicators of biodiversity, but this is not always the case, and so supplementing bird data with that of other taxa improves accuracy of assessment29. Thus, mapping and understanding the relationship between habitat structure and human impacts, along with data on the presence of birds and amphibians will help inform conservation efforts by identifying local biodiversity hotspots, and potential stressors and threats.

Although monitoring programmes have assembled data on bird e.g.30, and to a lesser extent amphibian e.g.31, presence across the GCFR, the combined use of acoustic data and artificial intelligence algorithms allows relatively uniform surveying of areas for extended periods (ARUs were deployed for at least four days, in each season), and throughout the day and night. Combined with remote sensing, these data may reveal patterns about how species’ presence changes with habitat structure, time since fire and anthropogenic habitat transformation. Remotely-sensed biodiversity indices can be acquired without the need for human capacity on the ground and can cover vast areas. Additionally, having automated processes to deliver useful indices and indicators of ecosystem state and health makes the translation to management responses faster and less onerous than once-off research (that may require considerable manual processing and discovery of suitable algorithms and analytic procedures). The BioSoundSCapes project aims to establish opportunities for both continuous remotely-sensed data flows and automated biodiversity assessment. A prominent value of acoustic surveys is that the original data are preserved unhindered by human observer bias. As new methodologies and technologies emerge to understand and analyze sound data, these can be used with the BioSoundScape dataset to advance our understanding of the biodiversity of the region32.

The BioSoundSCape project used ARUs to capture acoustic data to estimate animal diversity (focusing on birds and frogs), complemented by remote sensing techniques for vegetation habitat characterization. Protected Area rangers and volunteers in the project placed ARUs at 521 sites in the Greater Cape Floristic Region, South Africa, during the wet (July-August) and dry (October-November) seasons of 2023 noting that rainfall in the eastern areas are not as strongly seasonal as the western areas. The ARUs recorded one minute in every ten over 4–10 days, producing 825,832 minutes of recordings in.wav format. Here, we describe the study area, data collection and validation procedures, metadata and information on accessing these data.

Methods

Study area

The Greater Cape Floristic Region (GCFR) is a distinctive and biologically diverse area in southwestern South Africa that includes the Fynbos biome, Succulent Karoo biome, and patches of AfroTemperate Forest in the east and in fire refuges33. Characterized by a Mediterranean-type climate, the region experiences wet, mild winters and hot, dry summers, with fire playing a crucial role in the evolution of life history traits. The Fynbos biome (Fig. 1), characterized by shrubs and reed-like plants (restios), and the Succulent Karoo biome (rainfall <250 mm) dominated by woody-stemmed succulent shrubs, exhibit high plant endemism and diversity. Fire-return intervals in the Fynbos are typically between 7–25 years, a regime that may have been established around 15 million years ago34. In contrast, fire in the Succulent Karoo35 and AfroTemperate Forest36 is extremely rare, and these systems have not evolved to burn. Fire frequency in the Fynbos is a key determinant of species composition and community processes37. The three biomes have markedly different plant communities and vegetation structure.

Fig. 1
figure 1

The distribution of the Fynbos, Forest, and Succulent Karoo Biomes within the GCFR. Data from the South Africa National Vegetation Map.

Full size image

Data acquisition

AudioMoths and protocols

AudioMoth devices (hereafter, ARUs) are compact and cost-effective acoustic recorders12, and were deployed for environmental sound recording. These devices can record full-spectrum audio at variable sample rates, tailored to the project’s requirements. They typically record sounds from within a 50-m radius, but this can vary with topography, geophony, and vegetation structure38. ARUs store data on microSD cards and are powered by rechargeable AA batteries, making them suitable for extended field deployment. The configuration of the ARUs, including sample rate, gain level, and recording schedules, was customized using the AudioMoth Configuration App (see Table 1). We used AudioMoths device versions 1.1.0 and 1.2.0, and ARUs were programmed to use firmware version 1.5.0 or 1.8.1. These firmware versions share a persistent sampling configuration, and ARUs were configured to synchronize time using the AudioMoth chime smartphone app39, ensuring accurate timestamping across all recordings. For data collection, the devices were set to record one minute of audio at the start of every ten-minute interval (Table 1). This sampling strategy balanced the needs of gathering comprehensive acoustic data with memory, power, and battery life constraints, allowing a representative sample of the acoustic environment throughout the study period. Depending on vegetation at the site, ARUs were mounted on, in order of selection: large tree or shrub branches, bundles of small shrub stems, or rock crevices.

Table 1 Programmed settings used for AudioMoth recorders.
Full size table

BioSoundSCape team members, volunteers, and Protected Area rangers placed two to 15 AudioMoths at clusters of sample sites in 65 winter (wet season, June-August 2023) and 39 spring (dry season, October - December 2023) deployments (Figs. 2, 3). Sound data were collected at 521 unique sample sites – 505 sites in the wet season and 489 sites in the dry season, with 472 of these sites co-located as wet-dry season pairs (Fig. 3). Sample sites had been identified following a comprehensive protocol derived from a random stratification (see below) or vegetation plot locations that had been selected by the BioSCape project40. Field team members used smartphone apps (Gaia GPS, MAPS.ME app) or handheld GPS devices for navigation (Garmin Etrex) and the ArcGIS Survey123 smartphone app for data collection41, as well as backup data collected via physical field notes. The data collected include location coordinates (to within 20 m), date, time, and attribute information about the deployment point: approximate height of AudioMoth, estimated age of vegetation (time since last fire), presence of invasive alien plants, proximity to a wetland, and photographs of the deployment site in all four cardinal directions as well as a photograph of the mounted AudioMoth (Fig. 5). Date, time, and location coordinates were also spoken verbally into the first recording of the ARU at each site. GPS locations were cross validated to ensure spatial accuracy by capturing location details on both the handheld GPS devices (Garmin Etrex) and the Survey123 smartphone app. Strict protocols around ensuring consistency in time stamps and GPS locations also guaranteed the timing and positions of ARU placements. Survey123 allows collection of geographic information on any mobile device, offering time-cost efficiency in collecting field data, and also collected photographs to help find ARUs for retrieval and to provide an image of “on the ground” vegetation structure and topography. After 4–10 days of data collection (median = 7), teams retrieved devices, downloaded the audio data, and prepared the ARUs for subsequent deployments.

Fig. 2
figure 2

Number of 1-minute recordings within wet and dry season campaigns.

Full size image
Fig. 3
figure 3

Location of the sampling points in relation to the LVIS & AVIRIS flight boxes and vegetation type.

Full size image

Data download and pre-processing

Audio recording files were captured using the standard firmware AudioMoth naming mechanism which records the date and time of the file as the file name (YYYYMMDD_HHMMSS.WAV). Field teams downloaded audio data from the ARU SD cards onto 1 TB USB hard drives. Each set of recordings from a given site was downloaded directly into a folder with the unique deployment identifier (SiteID), given as < device > _ < begin_date > where < device > is the AudioMoth unit ID, and < begin_date > is the date of deployment. The combination of < device > and < begin_date > uniquely identifies each deployment, since each device can only be at one location on a specific date. Note that a given sample site typically had a wet and dry season deployment, each of which generated a unique SiteID.

The 1 TB USB hard drives were collected by the field coordinator at the end of each field season. Audio data from the 1 TB hard drives were then transferred to two 16 TB hard drives - one for primary storage and one for backup storage. Finally, the audio data were transferred to a secure server at Sonoma State University (SSU) for long-term storage and data post-processing.

Data post-processing

Once transferred to SSU, we used an R code script to rename the files to include the unique SiteID and recording time. The file naming convention for the audio recordings is < device > _ < begin_date > _ < rec_time > .WAV, where:

< device > = 8-character device number (e.g., s2lam001)

< begin_date > = date when recordings began at site in YYMMDD

< rec_time > = start date and time for 1-minute recording in YYYY-MM-DD_hh-mm

Time was in 24-hr format based on local South Africa time (GMT + 2).

Example file name: s2lam001_230713_2023-07-15_15-40.WAV (sampling with device s2lam001, with a deployment that began on 2023-07-13. This specific recording started at 15:40 (UTC 13:40) on 2023-07-15, on the third day of deployment at the site. The R code also deleted the first and last recording from the batch of audio files to reduce the chances of field team conversations at the times of deployment and retrieval being saved. Photographs of each deployment were downloaded from the online Survey123 point layer and renamed to correspond to SiteIDs. The file naming convention for these photographs is <device>_<begin_date>_<photo>.jpg (e.g., s2lam001_230713_mounted.jpg), where <photo> = “mounted” (an image of the recorder), “upward” (view of the canopy), or the viewing direction “north”, “east”, “south”, or “west”. These photographs were zipped for distribution on the Oak Ridge National Laboratory Distributed Active Archive Centre (ORNL DAAC).

Stratification

There were 54 acoustic sampling locations chosen based on existing vegetation or bird survey plots. On average, our sample locations were 16.5-m away from the center of these target plots.

The remainder of the ARU sampling locations were randomly selected using a randomized stratified sampling protocol within BioSCape AVIRIS NG (imaging spectrometer) flight boxes (Fig. 3). We used the 2020 South African National Land Cover (SANLC) GIS layer to classify land-cover into broad classes of Forest-Woodland, Grassland, Shrubland, Shrubland-Fynbos, Shrubland-Karoo, Shrubland-Karoo-Nama, Bare, Wetland, Water, Agriculture, Urban-Residential, Roads & Rail, and Excluded.

A Shuttle Radar Topography Mission digital elevation model (DEM) was used to classify the study region into classes Low: 0–500 m, Medium: 500–1000 m, and High: >1000 m. Another GIS layer represented fire history based on “veld age”, with classes: No data or No Fire, 1-to-6 years, 6-to-12 years, 12-to-17 years, 17-to-25 years, 25+ years. The fire history layer was only available for CapeNature protected areas, and was current at the time of site selection. We used a GIS overlay to create stratification zones with combinations of land-cover, elevation, and fire history classes across the entire GCFR (Fig. 4). Within these strata, we generated random points within protected areas, AVIRIS NG flight boxes, and 20–500-m from roads and trails. During the campaign, a field coordinator chose points to form a deployment cluster, while considering different strata. All sample locations were within the AVIRIS NG flight boxes and 60% were in LVIS (lidar) flight boxes of the BioSCape remote sensing data collection (note that the two sensor set flight boxes differ in size and shape).

Fig. 4
figure 4

Distribution of sample sites by (A) land cover, (B) elevation, and (C) fire history. Of the 521 sites total, 505 were visited twice and are reflected twice in this figure; n = 994.

Full size image

Data Records

Audio recordings

In total, we collected 825,832 one-minute.wav file audio recordings during the two campaigns (Wet season = 430,974 minutes; Dry season = 394,858 minutes) (Figs. 2, 3). Each visit to a site has its own data record and batch of recordings, identified by SiteID (see metadata description below). There is a total of 994 unique SiteID values with associated recording data and metadata.

All audio recording data and associated metadata are publicly available for download at NASA’s ORNL DAAC at https://doi.org/10.3334/ORNLDAAC/237242.

Site photographs

Also available on the DAAC1 are photographs associated with each deployment (e.g., SiteID). There are five to six photographs per deployment: one image of the mounted ARU, four directional images at the site pointing in each of the four cardinal directions, and for sites with canopy cover, a photograph pointing up (Fig. 5).

Fig. 5
figure 5

Directional photographs as well as a photograph of the mounted ARU were taken at each deployment to verify landscape information for each site. The above photographs are an example of images from one location: (a) mounted AudioMoth, and facing (b) east, (c) south, (d) west, and (e) north.

Full size image

Description of site metadata

An associated metadata comma-delimited file is distributed with the acoustic recordings on the ORNL DAAC42 and also available on the BioSoundSCape GitHub repository (https://github.com/pointblue/BioSoundSCape/tree/main). This file has the following fields:

SiteID: unique deployment identifier, given as s2lamXXX_YYMMDD where XXX is the AudioMoth unit ID, and YYMMDD is the date of deployment. Note that most locations were sampled in both the wet and dry season of 2023. These sites have unique SiteIDs. Paired samples are tracked with the PairID field.

AVIRIS: Y = in a BioSCape AVIRIS NG flight box; otherwise N

LVIS: Y = in a BioSCape LVIS flight box; otherwise N

WetlandType: NBA2018_National_Wetland_map5 GIS layer overlay with classes: Channelled valley-bottom, Depression, Estuary, Floodplain, River, Seep, Wetland Flat

LandCoverClass: SA_NLC_2020_GEO_CAPE GIS layer overlay with classes: Forest-Woodland,Grassland,Shrubland,Shrubland-Fynbos,Shrubland-Karoo,Shrubland-Karoo-Nama,Bare,Wetland,Water,Agriculture,Urban-Residential,Roads & Rail, Excluded (planted forests, mines and landfills)

ElevationClass: Shuttle Radar Topography Mission DEM layer overlay with classified into Low: 0–500 m, Medium: 500–1000 m, High: > 1000 m

FireClass: CapeNature “veld age” GIS layer overlay with classes (only available in CapeNature reserves): No data or No Fire, 1-to-6 years, 6-to-12 years, 12-to-17 years, 17-to-25 years, 25+ years

FieldWetlandType: Entered by field team members at each site with Survey123. Describes a wetland type. This field was only available for the dry season campaign and was not always entered. Classes include: Not near (100 m) any water, Within 50 m of a river or stream, In or near (25 m) wetland, In or near (10 m) seep

FieldVeldAge: Entered by field team members at each site with Survey123. Describes an estimate of the age of the site, mainly for fynbos sites. This field was only available for the dry season campaign and was not always entered. Classes include: Young (burned, < 6 yrs), Intermediate (6–17 years), and Old (>17 yrs).

FieldAliensWithin20m: Entered by field team members at each site with Survey123. Describes an estimate of alien plant species infestation. This field was only available for the dry season campaign and was not always entered. Classes include: Rare to Very Scattered, Scattered to Medium, Dense to Closed, None

Latitude: Latitude in decimal degrees, datum WGS 84

Longitude: Longitude in decimal degrees, datum WGS 84

PairID: Unique ID that indicates which wet and dry season sites belong to the same pair (e.g., spatially proximate)

RecordingNum: Number of audio recordings available at the site.

Technical Validation

To ensure recording data integrity, we tested all ARUs prior to each field season, and any malfunctioning ARUs were removed from circulation. Any ARUs that malfunctioned during the field season were also removed from circulation. To ensure audio data was accurately matched to location data, field teams recorded the location data by GPS (Garmin Etrex), smartphone app (Survey 123), and paper. Field teams also spoke the site coordinates and deployment date and time into the first recording at each site. When downloading the data, the field coordinator listened to the first recording to ensure that it matched the handwritten field notes before downloading into a folder with the appropriate site name.

Post-processing of the sound data was performed by co-authors at SSU (Clark) and Point Blue Conservation Science (Salas, Snyder) in California, USA. This team maintained a relational database of information related to properties, contacts, deployments, locations, recordings, and equipment in Airtable following protocols in43. Locations from S123 and the Garmin Etrex were compared and vetted for accuracy with a GIS distance analysis, ultimately selecting instrument type for the final coordinates. Sound recordings were renamed and organized on a server at SSU. In this process, first recordings and the Airtable database were evaluated to resolve issues with site names, recording times, and geospatial locations. Further, recordings that were not complete or empty files were deleted. Original raw sound files from the field are maintained on the server.

Roughly ten percent of locations were intentionally selected based on pre-existing vegetation plots where other research has occurred and would be occurring in the same year to allow for collaborative research with the broader BioSCape Project. The remaining locations were selected using a randomized sampling scheme to eliminate individual bias in choosing where to place audio recorders (e.g., using this sampling scheme a field team member could not place an audio recorder at a site simply because they knew there were birds present).

Usage Notes

Acoustic data can be analyzed in relation to various biotic and abiotic variables and their gradients, like vegetation structure, elevation, climate variables, disturbance regimes or distance to anthropogenic disturbance or structures. Due to the stratified sampling approach used in this project, this dataset covers gradients of vegetation structure, elevation, rainfall and human disturbance, allowing assessment of the effects of these factors on biodiversity.

Processing large amounts of audio recording data can be challenging. There are several tools available to assist researchers in identifying sounds in audio data, such as Arbimon44, Raven Sound Analysis45, and BirdNET46.