Accuracy of commercial pollen forecasts in the United States

Abstract

Approximately 80 million Americans suffer from seasonal pollen allergies, and many rely on commercial pollen forecasts to guide exposure reduction and medication timing. However, these forecasts are proprietary, and their accuracy has never been comprehensively evaluated. Here, we compared pollen forecasts from four widely used commercial platforms (Pollen.com, AccuWeather, Google/BreezoMeter, and WeatherTrends360) against empirical pollen measurements from 25 National Allergy Bureau (NAB) monitoring stations across the contiguous United States from 2022 through 2025. Forecasts and observations were standardized to a common pollen index for comparison. Overall agreement between forecasts and observations was modest, with pooled Spearman correlations of 0.52 (Pollen.com), 0.48 (AccuWeather), 0.40 (WeatherTrends360), and 0.32 (Google/BreezoMeter) across all forecast-observation pairs. All platforms overpredicted low-pollen days and underpredicted high-pollen days. Accuracy varied by pollen group, with tree and weedpollen forecasts generally outperforming grass forecasts. Forecast accuracy also declined with increasing lead time for AccuWeather and Pollen.com, while Google/BreezoMeter and WeatherTrends360 showed little degradation from day 0 to day 4. Forecast accuracy varied substantially by geography, with higher accuracy in the South-Central United States and lower, sometimes near-zero or negative, accuracy along the Pacific Coast and in parts of the Northeast. These results indicate that current commercial pollen forecasts fall short of the accuracy needed to reliably inform individual health decisions, and that accuracy is highly platform-, taxon-, and location-dependent. These findings motivate the development of open and improved pollen forecasting systems to better support public health decision-making.