# Hidden Markov model for acoustic pesticide exposure detection and hive identification in stingless bees

## TL;DR

This study explores using sound and a Hidden Markov Model to detect pesticide exposure and identify hives in stingless bees, offering a non-invasive monitoring tool.

## Contribution

The study introduces a Hidden Markov Model for detecting pesticide exposure and hive identification in stingless bees using acoustic data.

## Key findings

- The model showed higher performance in detecting pesticide exposure when analyzing data from individual hives.
- The model accurately classified individual hives, suggesting unique acoustic 'fingerprints'.

## Abstract

Pollinator populations are declining globally at an unprecedented rate, driven by factors such as pathogens, habitat loss, climate change, and the widespread application of pesticides. Although colony losses remain difficult to prevent, precision beekeeping has introduced non-invasive strategies for monitoring hive conditions. Acoustic data, combined with machine learning techniques, has proven effective in detecting stressors and specific events in honeybee colonies; however, such methodologies remain underexplored for stingless bees, a group of native pantropical pollinators. Meliponiculture, the practice of keeping stingless bees, is an expanding field that offers significant economic and conservation benefits. Stingless bees are particularly susceptible to pesticide toxicity, even at residual concentrations, underscoring the critical need to prevent hive losses and to understand the impacts of sub-lethal pesticide exposure on these species. This study addresses the challenge of detecting airborne pesticide exposure by aiming to identify stress responses in hives of the stingless bee Tetragonisca fiebrigi when exposed to chlorpyrifos, a commonly used insecticide. We employed a Hidden Markov Model (HMM) with MATLAB’s Hidden Markov Model Toolkit (MATLABHTK) to analyze acoustic data from eight hives under both exposed and unexposed conditions, assessing the potential of acoustic monitoring as an indicator of pesticide-related stress. Initial analysis across multiple hives indicated moderate model performance. However, hive-specific analyses yielded higher performance in detecting pesticide exposure. Furthermore, the model accurately classified individual hives, suggesting the presence of a distinct acoustic ’fingerprint’ for each hive. These findings advance the field of stingless bee bioacoustics and provide initial evidence that acoustic monitoring of stingless bee hives could be a useful and non-invasive tool to detect airborne pesticide contamination.