Related Experiment Videos
Investigation of short-range cedar pollen forecasting
J-J Delaunay1, C Seymour, V Fouillet
1NTT Energy and Environment Systems Laboratories, 3-I Morinosato Wakamiya, Atsugi, 243-0198 Kanagawa, Japan. jean@mech.t.u-tokyo.ac.jp
Physical Review. E, Statistical, Nonlinear, and Soft Matter Physics
|February 9, 2005
Summary
Accurate pollen forecasting remains challenging. While some determinism exists in pollen data, chaotic system dynamics were not confirmed, limiting prediction accuracy for large pollen bursts.
Area of Science:
- Environmental science
- Atmospheric science
- Biometeorology
Background:
- Pollen forecasting is crucial for allergy sufferers seeking to avoid exposure.
- Recent research suggested pollen concentration dynamics resemble low-dimensional chaotic systems, offering new forecasting approaches.
Purpose of the Study:
- To investigate the underlying dynamics of hourly cedar pollen concentration time-series.
- To evaluate the effectiveness of nearest-neighbor methods and other time-series techniques for pollen forecasting.
Main Methods:
- Analysis of two seasons of hourly cedar pollen concentration data.
- Application of the nearest-neighbor method with local constant prediction and a 1-h lead time.
- Evaluation of nonlinear filtering and standard time-series techniques like neural networks.
Main Results:
- Evidence of a small degree of determinism in pollen time-series dynamics was found.
- The nearest-neighbor model effectively predicted small to medium pollen variations but struggled with large, intermittent bursts.
- A nonlinear filter significantly improved the nearest-neighbor model's performance; neural networks did not offer further improvement.
Conclusions:
- The pollen time-series dynamics are likely not strongly governed by a low-dimensional chaotic system.
- Nonstationarity and unpredictable large pollen bursts complicate accurate pollen forecasting.
- Advanced filtering techniques can enhance predictive models for pollen concentration.