Using Data Driven Artificial Intelligence to Reveal Pesticide Induced Changes in Pollinator Behaviour
Lead Research Organisation:
University of Sheffield
Department Name: Computer Science
Abstract
When crops are treated with pesticides both pests and beneficial insects are affected. It
would be useful to know how different chemicals affect different species but the current ways
we use to determine the impact of such treatments have substantial limitations: They often
take a long time (e.g. to detect colony-scale outcomes) or only test a small part of an insect's
behaviour.
Our project uses a pioneering new tracking method to monitor bees during foraging trips. We
use state-of-the-art AI techniques to process the recorded 3d flight paths to detect subtle
changes in behaviour in response to pesticide exposure. We use this to explore both the
mechanisms through which pesticides affect pollinators and the agricultural implications of
the behavioural changes. The objectives are to (a) produce a sensitive method for detecting
treatment induced behavioural change in bumblebees, (b) look at the effect of treating plants
with fungicides on the foraging behaviour of bumblebees.
We will perform the experiments in a controlled environment agricultural (CEA) research
facility, with a crop of strawberries, using industry-standard techniques. We will use
commercial colonies of the common eastern bumble bee, which are commonly used in CEA
to ensure the crop is pollinated (to maximise yield). These bees will then be used to test the
impact of insecticidal treatments. The colonies will be exposed to either field-realistic
quantities of widely used neonicotinoid pesticides, fungicides or control. The colonies will be
allowed to forage within the controlled environment on strawberry plants, a portion of which
have also been treated with fungicide. This experimental design will allow us to investigate
the impact of the two pesticides on behaviour, and also the effect of the fungicidal plant
treatment on foraging (e.g. whether bees avoid/prefer the treated plants), and also allow us
to look for interactions between the treatments.
The project requires the development of three tools:
(1) the tracking tool we use works by taking photos with a flash of bees foraging while
wearing retroreflective tags. The tags appear as bright spots in the photos, allowing the bees
to be tracked. Different bees are impossible to distinguish with this technique so we will be
extending the method with coloured tags to allow multiple bees to be uniquely tracked.
(2) the 3d path of the bee might provide important information about the impact of the
pesticides, so we will combine multiple tracking cameras and then use recently developed
mathematical tools (a method for Bayesian inference) to reconstruct the 3d flight path of
each bee.
(3) To make sense of this huge dataset of 3d flight paths we will consider several
approaches to extract relevant features from the raw data, which will then be used to train a
classifier to distinguish between the different treatment groups.
The research will provide a novel, sensitive tool for detecting behavioural changes in flying
insects, and explore the impact of specific pesticides (and their interactions) on key
pollinators. We anticipate the new assay will allow manufacturers to produce more targeted
pesticides (and thus support growers) and, through its use by other researchers, provide
legislators with the evidence base required for regulatory decision making. Medium term
impacts are expected for the wider public through improved biodiversity
would be useful to know how different chemicals affect different species but the current ways
we use to determine the impact of such treatments have substantial limitations: They often
take a long time (e.g. to detect colony-scale outcomes) or only test a small part of an insect's
behaviour.
Our project uses a pioneering new tracking method to monitor bees during foraging trips. We
use state-of-the-art AI techniques to process the recorded 3d flight paths to detect subtle
changes in behaviour in response to pesticide exposure. We use this to explore both the
mechanisms through which pesticides affect pollinators and the agricultural implications of
the behavioural changes. The objectives are to (a) produce a sensitive method for detecting
treatment induced behavioural change in bumblebees, (b) look at the effect of treating plants
with fungicides on the foraging behaviour of bumblebees.
We will perform the experiments in a controlled environment agricultural (CEA) research
facility, with a crop of strawberries, using industry-standard techniques. We will use
commercial colonies of the common eastern bumble bee, which are commonly used in CEA
to ensure the crop is pollinated (to maximise yield). These bees will then be used to test the
impact of insecticidal treatments. The colonies will be exposed to either field-realistic
quantities of widely used neonicotinoid pesticides, fungicides or control. The colonies will be
allowed to forage within the controlled environment on strawberry plants, a portion of which
have also been treated with fungicide. This experimental design will allow us to investigate
the impact of the two pesticides on behaviour, and also the effect of the fungicidal plant
treatment on foraging (e.g. whether bees avoid/prefer the treated plants), and also allow us
to look for interactions between the treatments.
The project requires the development of three tools:
(1) the tracking tool we use works by taking photos with a flash of bees foraging while
wearing retroreflective tags. The tags appear as bright spots in the photos, allowing the bees
to be tracked. Different bees are impossible to distinguish with this technique so we will be
extending the method with coloured tags to allow multiple bees to be uniquely tracked.
(2) the 3d path of the bee might provide important information about the impact of the
pesticides, so we will combine multiple tracking cameras and then use recently developed
mathematical tools (a method for Bayesian inference) to reconstruct the 3d flight path of
each bee.
(3) To make sense of this huge dataset of 3d flight paths we will consider several
approaches to extract relevant features from the raw data, which will then be used to train a
classifier to distinguish between the different treatment groups.
The research will provide a novel, sensitive tool for detecting behavioural changes in flying
insects, and explore the impact of specific pesticides (and their interactions) on key
pollinators. We anticipate the new assay will allow manufacturers to produce more targeted
pesticides (and thus support growers) and, through its use by other researchers, provide
legislators with the evidence base required for regulatory decision making. Medium term
impacts are expected for the wider public through improved biodiversity
