Machine learning for biodiversity - UAV-based flower detection as an indirect proxy for bee abundance

Aug 2, 2026

Upload by Aurelia

Pollination supports agriculture and ecosystem functioning, but bee populations face mounting risks from habitat fragmentation, intensive agriculture, and climate change. Effective monitoring of key habitat factors such as flower cover is therefore essential, while traditional field approaches remain costly and time-consuming. A machine-learning workflow is proposed using optical RGB imagery from an unoccupied aerial vehicle (UAV) to estimate flower cover and infer wild bee pollinator abundance numerically. A single model is trained to recognize flower cover across diverse grassland ecosystems, enabling efficient estimates of bee abundance and diversity over time; Gradient Boost Machine results show strong correlation.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages12

Reading Time30 min

AccessFree to download

Abstract

Introduction

Background: biodiversity decline and key threats

Bees and pollination ecosystem services

Need for efficient monitoring and remote sensing alternatives

| Machine learning for biodiversity: UAV-based flower detection as an indirect proxy for bee abundance
Ludovico Chieffallo a ,∗, Michele Torresanib, Piero Zannini a, Jan Peter Reinier de Vries d, Marharyta Blahad,e, Alessio Monacchia a, David Kleijnf, Duccio Rocchini a,c
a BIOME Lab, Department of Biological, Geological and Environmental Sciences, Alma Mater Studiorum University of Bologna, via Irnerio
42, 40126, Bologna, Italy
b Free University of Bolzano/Bozen, Faculty of Agricultural, Environmental and Food Sciences, Piazza Universitá/Universitätsplatz 1, 39100, Bolzano/Bozen, Italy c Czech University of Life Sciences Prague, Faculty of Environmental Sciences, Department of Spatial Sciences, Kamýcka 129, Praha - Suchdol, 16500, Czech Republic
d Center Agriculture Food Environment, University of Trento, S. Michele all’Adige, Italy e Research and Innovation Centre, Fondazione Edmund Mach, S. Michele all’Adige, Italy
f Plant Ecology and Nature Conservation Group, Wageningen University, Droevendaalsesteeg 3a, Wageningen 6708PB, The Netherlands | | | --- | --- | | A R T I C L E I N F O | A B S T R A C T
Pollination plays a crucial role in supporting agriculture and ecosystem functioning, making it an essential ecosystem service provided by bees and other insects. However, bee populations are increasingly threatened by habitat fragmentation, intensive agriculture, and climate change, among other threats. Improved monitoring of critical habitat factors, such as flower cover, is crucial to restore pollinator populations. Traditional approaches, such as field analyses, are often time-consuming and expensive, prompting the adoption of alternative methods to achieve greater efficiency and cost-effectiveness. Here, we introduce a novel approach that incorporates machine learning algorithms and optical images obtained from an unoccupied aerial vehicle (UAV). Using machine learning methods on RGB UAV imagery enabled us to estimate flower cover in UAV-monitored areas and make numerical inferences of wild bee pollinator abundance from those estimates.
Unlike our previous study, which relied on separate machine learning models for each study area, our new method develops a single model that can automatically and efficiently recognize flower cover in various grassland ecosystems to successively estimate bee abundance and diversity. In addiction to this main objective, we also sought to determine which machine learning model would perform this important task best. The machine learning models used, particularly the Gradient Boost Machine (GBM), highlighted the capability of UAV RGB images combined with artificial intelligence to predict flower cover over time, which was highly correlated with bee abundance and diversity.
This development represents an additional starting point for the use of machine learning and deep learning techniques in biodiversity studies within AES systems. | | Dataset link: [https://zenodo.org/records/1479](https://zenodo.org/records/1479)[ ](https://zenodo.org/records/1479)3347, [https://github.com/Ludovico-Chieffallo/](https://github.com/Ludovico-Chieffallo/)[ ](https://github.com/Ludovico-Chieffallo/)[Machine_Learning_for_biodiversity.git](Machine_Learning_for_biodiversity.git) | | | Keywords:
Machine learning
Applied ecology Bees
Remote sensing UAV | | 1. Introduction Over the decades, global biodiversity has steadily decreased. This undermines a crucial resource that supports ecosystem processes, maintains ecological balance, and underpins a wide range of essential goods and services for human survival (David et al., 2015; Bradley et al., 2012). The most widespread threats to biodiversity are habitat fragmentation and intensive agriculture (Denis et al., 1991). Natural habitats have been turned into large monocultures with high fertilizer and pesticide use, drastically reducing the biological diversity of agricultural ∗ Correspondence to: Street: Via Irnerio, Bologna, 4

FAQs

How does the method estimate bee abundance using UAV imagery?

The workflow applies machine-learning models to RGB images captured by an unoccupied aerial vehicle (UAV) to estimate flower cover, then uses those estimates to make numerical inferences of wild bee pollinator abundance.

What is the main advantage of the new approach compared with the previous study?

Instead of training separate models for each study area, the new method develops a single model that can automatically and efficiently recognize flower cover across multiple grassland ecosystems and then estimate bee abundance and diversity.

Which machine-learning model is highlighted as performing particularly well?

The Gradient Boost Machine (GBM) is emphasized for its capability to predict flower cover over time, with high correlation to bee abundance and diversity.

Aug 2, 2026

Upload by Aurelia