116 Views

Plant-Insect Ecosystems

Section Symposium

Honey Bees as Global Environmental Monitors of Pollutants, Pesticides, Pathogens & Emerging Threats On-Demand Presentations

Machine learning prediction models: A tool for understanding pesticides monitoring with honey bees

On-Demand

Presenting Author(s)

Silvina Niell (she/her/hers)

Professor (Prof. Adj.)
Universidad de la República
Paysandu, Paysandu, Uruguay

Co-author(s)

FJ

Florencia Jesús

Universidad de la República
Montevideo, Uruguay
RH

Ricardo Hladki

Universidad de la República
Montevideo, Uruguay
MD

Melina Dorrego

Universidad de la República
Montevideo, Uruguay
RG

Romina Genolet

Universidad de la República
Montevideo, Uruguay
RD

Rosana Díaz

Universidad de la República
Montevideo, Uruguay
SD

Sebastián Díaz-Cetti

National Institute of Agricultural Research INIA La Estanzuela
BB

Belén Branchicella

National Institute of Agricultural Research INIA La Estanzuela
GN

Gastón Notte

Universidad de la República
Montevideo, Uruguay
ES

Estela Santos

Universidad de la República
Montevideo, Uruguay
NG

Natalia Gérez

Universidad de la República
Montevideo, Uruguay
VC

Verónica Cesio

Universidad de la República
Montevideo, Uruguay
HC

Héctor Cancela

Universidad de la República
Montevideo, Uruguay
YM

Yamandú Mendoza

Mesa Apicola Artigas
HH

Horacio Heinzen

Universidad de la República
Montevideo, Uruguay

The ubiquitous presence of pesticide residues in honey bees and beehive products (wax, trapped pollen, honey, nectar, beebread, honeycombs) has been found all over the world during the past decade. In Uruguay, during two campaigns in different zones (2014-2017; 2019-2020) honey bees and their beehives have been assessed as pesticide residues biomonitors of agroecosystems. Pesticides profile found in each apiary reflected the land use within its ecosystem. But still the question if and/or how are these findings affecting honey bees and the environment is unsolved. Aiming to contribute to answer, machine learning prediction models have been explored. In particular, the models evaluated are Support Vector Machines (SVM) and recently compared to Neural Networks (NN). These are supervised classification models which make predictions based on the variables defined and need to be trained and validated. The two classes have been defined as “affected” and “non-affected” agroecosytems based on the pesticide load according to their land use. The variables used were biological observations of colony strength, pesticide number of detections in the apiary and indexes constructed with hazard quotient´s structure. The selection and combination of different variables allows making various models which can be compared. From the total number of samples, a set (90%) is used for training and the other to validate the models. Accurate predictions were obtained with some of the SVM and NN models assayed. These are promising results which indicate that machine learning models can help to interpret the pesticide residue data obtained from monitoring with honey bees.