Applying machine learning to forecast daily Ambrosia pollen using environmental and NEXRAD parameters

Area of Science:

• Environmental Science
• Allergy Research
• Data Science

Background:

• Allergic diseases affect millions, with airborne plant pollen as a major trigger.
• Ragweed (Ambrosia) pollen is a potent allergen prevalent in North America.
• Accurate daily pollen concentration forecasts are vital for public health and allergy management.

Purpose of the Study:

• To develop and evaluate machine learning models for estimating daily atmospheric ragweed pollen concentration.
• To assess the effectiveness of meteorological, land surface, and radar data in pollen prediction.
• To identify key predictors for ragweed pollen concentration.

Main Methods:

• Utilized supervised machine learning algorithms: random forests, neural networks, and support vector machines.
• Integrated meteorological data, land surface parameters, and Next-Generation Radar (NEXRAD) measurements.
• Validated model performance using a holdout cross-validation method with 10% of the data.

Main Results:

• Random forests achieved the highest prediction accuracy (R=0.61, R²=0.37), followed by support vector machines (R=0.51, R²=0.26) and neural networks (R=0.46, R²=0.21).
• The study successfully demonstrated the utility of machine learning and diverse data sources for pollen estimation.
• Identified and ranked the relative importance of predictors using random forests, correlation coefficients, and interaction information.

Conclusions:

• Machine learning models, incorporating meteorological and radar data, can effectively estimate daily ragweed pollen concentrations.
• These predictive models offer a valuable tool for individuals with allergies and healthcare professionals.
• Further research can refine these models for improved allergy forecasting and management.