Low-cost optical detection of pollen bioaerosols with machine learning for human health

Mills, Sophie A. ORCID: 0000-0001-6072-6549 (2024). Low-cost optical detection of pollen bioaerosols with machine learning for human health. University of Birmingham. Ph.D.

Preview

Mills2024PhD.pdf
Text - Accepted Version
Available under License All rights reserved.
Download (11MB) | Preview

Abstract

Pollen and associated sub pollen particles are responsible for up to 40% of populations suffering from allergies and the situation in many countries is becoming more severe with changing lifestyles, environment, and climate. Meanwhile, pollen can interact with cloud processes, affecting cloud albedo, lifetime, and precipitation patterns. However, our means to characterise airborne pollen, and bioaerosols in general, are severely limited, making it difficult to answer many important questions and assess risk to public health. Conventional pollen monitoring instruments are generally manual samplers with crucial limitations of labour, time, and cost. There are recent advancements towards developing automated pollen monitoring instruments, however, these are expensive and not economically viable to populate large monitoring networks with high spatial resolution.

The objective of this work is to address the limitations of conventional methods by investigating alternative methods that can provide useful information on pollen bioaerosols for fundamental understanding and public health. This thesis presents and evaluates novel, low-cost methods for monitoring airborne pollen, using optical particle counters (OPCs) and machine learning, and investigating physical properties of pollen under varying atmospheric conditions, using an acoustic levitator, macroscope and computer vision techniques. The superior ability of supervised machine learning models to distinguish pollen trends from OPC data is demonstrated, as well as their potential to provide useful, high spatiotemporal resolution data in unique locations and for public health. This work provides comprehensive detail on how to train, interpret and employ such models for purpose, including scrutinising how the models learn to distinguish between different pollen types. Collectively, these studies demonstrate the potential for these low-lost techniques to provide novel information on pollen bioaerosol characteristics that was previously inaccessible. This novel information could be vital for our comprehension of the bioaerosol component of atmospheric aerosols, climate models, pollen forecasts, and public health advice and warnings.

Type of Work:

Thesis (Doctorates > Ph.D.)

Award Type:

Doctorates > Ph.D.

Supervisor(s):