Development of a Random Forest model for forecasting allergenic pollen in North America

Area of Science:

• Environmental science and public health
• Computational biology and bioinformatics

Background:

• Pollen allergies significantly impact public health, necessitating better symptom management strategies.
• Accurate airborne pollen concentration data is crucial for managing allergies, but observations are sparse in the contiguous United States and southern Canada (CUSSC).
• Existing pollen observation networks are often seasonal and intermittent, limiting their utility.

Purpose of the Study:

• To develop and evaluate statistical models for predicting daily pollen concentrations using a machine learning Random Forest algorithm.
• To assess the models' performance in forecasting four key pollen types (Quercus, Cupressaceae, Ambrosia, and Poaceae) across different CUSSC locations.
• To explore the potential of modeling for pollen forecasting in areas lacking direct observations.

Main Methods:

• Utilized a Random Forest machine learning algorithm to predict daily pollen concentrations.
• Incorporated meteorological and vegetation data at city and regional scales as model inputs.
• Investigated a data augmentation technique to enhance model predictive skill.
• Developed models for forecasting pollen in data-scarce locations.

Main Results:

• The developed models demonstrated improved skill in forecasting pollen concentrations compared to previous methods.
• A data augmentation technique was found to significantly enhance model performance.
• Models successfully provided pollen forecasts for locations without direct observational data.
• Forecast skill was notably higher in models predicting for unobserved locations.

Conclusions:

• Machine learning models, particularly Random Forest, can effectively predict airborne pollen concentrations and fill observational data gaps.
• Data augmentation is a valuable technique for improving the accuracy of pollen prediction models.
• The study highlights the potential for enhanced pollen allergy management through improved forecasting, despite limitations in observational data resolution.