Database of Bioacoustics Monitoring in farmscapes of Mandla, Madhya Pradesh, India

The study was conducted in Mandla district in Madhya Pradesh (MP), a central Indian state located between 22°13' and 23°11' N latitude and 79°57' and 81°12' E longitude (Survey of India, 2026). The district is characterized by a heterogenous landscape comprising of agriculture land, forests, rangelands, waterbodies and human settlements. The region is predominantly inhabited by tribal communities who actively play an active role in managing natural resources such as community owned forests and waterbodies. Local livelihoods are closely linked to forests which provide a wide range of products. Important wild trees such as Mahua, tendu and jamun that produce culturally important non-timber forest produce and are important for birds are retained and protected within the agricultural landscapes. Although the community practices traditional farming methods, conventional methods are gaining popularity and are gradually replacing traditional agroecological farming methods. NGOs such as FES (Foundation for ecological security) and PRADAN (Professional Assistance for Development Action) with the support of government agencies such as IIFSR (Indian Institute of Farming Systems Research) have been playing a key role in the revival of traditional farming methods by integrating diversified livelihood options. Within this multifunctional landscape of Mandla, we monitor the status of diversity of birds using acoustic monitoring as an indicator to understand the long-term impacts of landscape variables and farming practices.
Methodology:Acoustic data collection:
We used Song meter Micro devices at all the 15 locations and the devices were fixed at an approximate height on 2 meters above ground. The devices were further protected with a shelter to avoid direct contact with rain during the monsoon season. The devices were set at a sampling rate of 44100Hz with Maximum record length of 1min every 10 minutes and gain of 18dB for 24 hours a day and from August 2024- November 2025. Data was collected every 2-3 weeks but there are huge data gaps at certain locations due to rains, inaccessibility, loss of memory cards etc leading to varied sampling effort at different locations. The data was collected following a protocol developed by the soundscapes lab.
Bird acoustic data analysis:
The data collected was processed using machine learning tools to analyse the patterns in the audio files by Wildersensing (an innovative leader organization focusing on biodiversity monitoring and reporting, that provide specialised services on comprehensive ecological assessments and sustainability practices). The tool scans a file in 3 second chunks and tries to identify the species that has made a sound (excluding non-bird sounds). Each possible identification is given a confidence rating (was set at 85%). Each identification made is assigned a probability which is similar to how humans identify a sound. The organisation has carried out validation checks from projects across the globe and have high accuracy ratings. However, there could be errors due to one of two reasons – a false positive where the wrong species is identified and a false negative where it misses something that should have been identified but such errors are rare. Bird calls are complex and some species have very similar calls while others mimic other species or a phrase in one bird’s song can be very similar to another which could lead to erroneous identification. The tool allows us to adjust the probability of the data being correct and typically only calls which have a probability over 80% is reported. Increasing the probability to 90% will halve the number of records while increasing the quality. We set the probability at 85%.