Database of Bioacoustics Monitoring in Smallholder Agroecosystems of Makueni County, Kenya

This dataset was generated through a passive acoustic monitoring study conducted across agroecological landscapes in Makueni County, Kenya, encompassing four sub-counties: Mbooni, Kaiti, Makueni, and Kibwezi West. Autonomous acoustic recording devices were deployed at 15 strategically selected monitoring sites to establish a baseline dataset on bird diversity and soundscape dynamics across smallholder farming landscapes. The dataset enables the assessment of how agroecological conditions, farming systems, and surrounding landscape characteristics influence avian diversity, community composition, and soundscape dynamics over time.
The monitoring sites were strategically distributed across representative agroecological zones, spanning the Upper Midlands, Lower Midlands, Inner Lowlands (drylands), and transitional (ecotone) landscapes. Site selection was designed to capture the diversity of farming systems in the region, including simplified cropping systems, diversified cropping systems, and integrated agroforestry systems. The sites also represent contrasting agricultural water management regimes, with some located in exclusively rainfed farming systems and others encompassing both rainfed and irrigated production systems.
To account for the influence of the surrounding landscape on bird communities, several monitoring sites were established in close proximity to fragmented forests, pasturelands, seasonal water bodies, and human settlements. This broad environmental gradient provides a heterogeneous landscape context for assessing how agroecological conditions, farming practices, and adjacent habitats shape avian diversity and community composition.
Makueni County lies within Kenya's semi-arid region and is characterized by predominantly smallholder mixed farming systems, high rainfall variability, and increasing pressures from climate change. Passive acoustic monitoring offers a standardized, non-invasive approach for assessing bird communities, which serve as sensitive indicators of ecosystem health, habitat quality, and environmental change. As one of the first long-term bioacoustic datasets from Kenya's semi-arid agricultural landscapes, this dataset provides an important baseline for biodiversity monitoring and supports future research on agroecology, landscape restoration, conservation, climate resilience, and sustainable land management.
Methodology:Passive acoustic monitoring was conducted using Song Meter Micro autonomous recording units deployed across the 15 monitoring sites. Each recorder was mounted approximately 2 meters above the ground and protected with a weather shelter to minimize exposure to rainfall and other environmental conditions. The recorders were configured with a sampling rate of 48,000 Hz and a gain setting of 18 dB, and programmed to record one-minute audio samples after every five minutes, operating continuously for 24 hours per day from August 2024 onwards.
Data were retrieved from the recording units at approximately three-week intervals. Although all monitoring sites followed a standardized sampling protocol, recording effort varied among sites due to occasional device interference, loss of memory cards and batteries, loss of recording units, and other logistical constraints. Consequently, the dataset contains temporal gaps for some sites and sampling periods. All acoustic recordings were collected in accordance with the standardized passive acoustic monitoring protocol developed by the Soundscapes Laboratory.
Data Analysis: Audio recordings were processed using the WilderSensing machine-learning platform, which analyses recordings in three-second segments to detect and identify bird vocalizations while filtering out non-avian sounds. Species identifications were retained using a confidence threshold of 85%, providing a balance between detection sensitivity and classification accuracy. Although the automated identification model has undergone extensive validation and demonstrates high overall accuracy, occasional false-positive and false-negative detections may occur due to similarities in bird vocalizations, vocal mimicry, overlapping calls, background environmental noise, and variations in recording quality.
This dataset provides a valuable baseline for the long-term monitoring of bird communities in smallholder agricultural landscapes and supports future research on biodiversity conservation, landscape ecology, agroecological transitions, ecosystem restoration, and the ecological impacts of land-use and climate change.