The missing queen: a non-invasive method to identify queenless stingless bee hives

Abstract

Stingless bee hives are commonly managed in the global tropics and subtropics. However, current monitoring methods for these hives are invasive and subjective, relying on manual assessments conducted by beekeepers. This approach may harm bees and, if performed by non-specialists, can lead to the death of key reproducing individuals: the mother queens. These queens are vital for maintaining hive health, as their absence can lead to colony death. Our study is aimed at exploring the potential of three hive indicators: temperature, humidity, and sound, as predictive factors for discerning the presence or absence of a mother queen in Tetragonisca fiebrigi hives. To do this, we collected data on these variables from six hives, including three queenless hives and three queenright hives, monitored over four consecutive days. Temperature and humidity were recorded every 15 min during this period, and 15-min audio recordings were made each day. We then employed five machine learning algorithms (extreme learning machine, K-nearest neighbors, multilayer perceptron, random forest, and support vector machine) to classify the data. Our findings revealed that all tested algorithms consistently achieved accuracy rates exceeding 90%, whether using acoustic or microclimatic variables. However, the highest accuracy was achieved with the microclimatic dataset. This approach holds great potential for reducing the damages caused by manual inspections while also enabling faster and more precise identification of the health of beehives. By implementing remote monitoring systems based on these indicators, beekeepers can benefit from improved efficiency and accuracy in assessing hive conditions.