Public websites show low accuracy in predicting pollen counts in five U.S. cities
Click Here to Manage Email Alerts
Key takeaways:
- Pollen forecasts varied significantly among websites used in the study.
- The forecasts showed high variability and low overall accuracy.
Pollen predictions on public websites across five U.S. cities showed significant inaccuracies, according to a study published in The Journal of Allergy and Clinical Immunology: In Practice.
None of the websites that were analyzed consistently provided accurate pollen counts when compared with data from the National Allergy Bureau (NAB), Divya Shah, MD, resident in the department of internal medicine, The University of Arizona College of Medicine, and colleagues wrote.
The study looked at tree, grass and/or ragweed pollen levels from NAB stations in Washington, D.C.; Atlanta; Scottsdale, Arizona; Seattle; and Omaha, Nebraska, from March 1 to May 1, 2023. It then compared these levels with data from four prominent websites, identified in the study as A, B, C and D, chosen through a Google search for “pollen forecasts.”
Data from the websites were recorded based on numerical counts to establish a classification system of “none, low, medium, high” to align with NAB guidelines. Data analysis involved comparing tree, grass and ragweed pollen levels from NAB with 1-day pollen forecasts from the websites alongside a persistence model forecast method that used an NAB count from the day before to predict the next day. The days were then classified as either a match, a near match or a mismatch.
Data analysis of tree pollen forecasting accuracy in the five cities revealed that the persistence model continuously outperformed the chosen websites. Among the websites, C and D were more accurate in reporting tree pollen levels than the others. C was most accurate in Washington, D.C., Seattle and Scottsdale, whereas D worked best in Atlanta and Omaha.
The persistence model was the most accurate in tree pollen forecasting for exact matches. The model’s predictions were 73%, 65%, 66%, 54% and 53% for Washington, D.C., Atlanta, Scottsdale, Seattle and Omaha, respectively, compared with prediction rates for the websites of 54%, 44%, 34%, 25% and 34%.
In assessing grass pollen forecasting, website A did not report on grass pollen, and B, C and D did not include grass pollen in their reports on Seattle and Omaha.
The persistence model showed the highest accuracy in grass pollen forecasting for Washington, D.C., Atlanta and Scottsdale, including match counts of 18 of 19, 27 of 34 and 26 of 43. Website D had the highest grass pollen count match percentages for these cities at 80%, 71% and 23%.
For ragweed, website A did not provide reports, whereas B, C and D did not have consistent reports. Websites C and D were most accurate in reporting ragweed pollen forecasts in Washington, D.C., with two of two. Websites B and C had more matches in Atlanta than the persistence model with 25 of 33 vs. 11 of 17. The persistence model was most accurate in Scottsdale with 50 of 61 matches. In Omaha, the persistence model along with websites B, C and D had accurate reports of ragweed pollen counts of none (12 of 12; 21 of 21; 20 of 20; and 20 of 20).
Authors of the study emphasized the significant variability and low accuracy in these publicly accessible websites. They speculated that the factors impacting this inaccuracy could be unpredictable weather conditions and a lack of existing data to serve as a basis for forecasting. Also, they encouraged a reevaluation of current forecasting methods.
Perspective
Back to Top
The findings are significant because they show that pollen forecasts on public websites may not be as accurate as we thought. This will have significant public health implications, as many individuals usually refer to these websites to plan for activities and medication use.
In Australia, similar findings have been reported (Emmerson et al, 2022). According to the authors, variations in forecast accuracy were observed across different locations and years, potentially linked to factors such as forecasting expertise, the extent of historical pollen data, and the complexity of the pollen season.
The accuracy of pollen forecasts relies on a combination of local pollen aerobiological understanding and insights drawn from meteorological and environmental conditions affecting grass phenology and pollen dynamics.
Given the impact of climate change and the heightened occurrence of extreme weather phenomena such as droughts, wildfires and thunderstorms, the variability in pollen aerobiology is expected to increase, posing challenges for accurate forecasting.
Therefore, there is a pressing need for further research and the development of pollen forecasting methods tailored to local conditions. It would be useful for countries to have a national pollen monitoring network so the results can be collated, and this can aid model development and increase confidence in the value of pollen forecasts distributed to the community.
In my opinion, health care professionals can use these results to advise patients not to be too reliant on public forecasting websites when making decisions on allergen avoidance and treatment. It is important for patients to be adherent to medications and remain cautious of the adverse health impacts of pollen throughout the pollen season.
Collapse
Read more about
Click Here to Manage Email Alerts