Constructing a pollen proxy from low-cost Optical Particle Counter (OPC) data processed with Neural Networks and...

Area of Science:

• Environmental Science
• Aerobiology
• Sensor Technology

Background:

• Pollen allergies impact a large global population, with increasing prevalence.
• Automated pollen monitoring systems are needed to provide timely data.
• Low-cost Optical Particle Counters (OPCs) offer real-time, high-resolution particulate matter data.

Purpose of the Study:

• To evaluate the effectiveness of low-cost OPC sensors for monitoring airborne pollen.
• To develop and assess machine learning models for constructing pollen proxies from OPC data.

Main Methods:

• Supervised machine learning (Neural Network and Random Forest) models were trained using data from a Hirst-type sampler.
• OPC data, including particle size, temperature, and relative humidity, were used as input features.
• Model performance was evaluated using correlation coefficients, coefficient of determination, and F1 Scores, with 40% of data held out for testing.

Main Results:

• Neural Network and Random Forest models significantly outperformed simpler proxies.
• The best performing model, a Neural Network for Poaceae (grass) pollen, achieved a Spearman correlation of 0.85 and R-squared of 0.67.
• Models successfully captured monthly and diurnal pollen trends, with F1 Scores up to 0.83 for high pollen event detection.

Conclusions:

• Low-cost OPC sensors coupled with machine learning can provide meaningful and useful information on airborne pollen.
• This approach offers a viable alternative for automated pollen monitoring, benefiting allergy sufferers with timely data.
• Further research and development can enhance the application of OPCs in aerobiology.