Short-distance dispersion of birch pollen

Yamane, Towa*; Inatsu, Masaru*; Kawano, Jun*; Sato, Takuto  ; Kusaka, Hiroyuki*

This study aims to obtain fundamental information on birch pollen deposition data by field observation for the high-resolution, accurate pollen modeling. On the peak dispersal day in 2024, simple pollen collectors were installed just below and at three downwind points of an isolated birch tree line in Ebetsu, Hokkaido, Japan. Meteorological observation were also conducted at the site during the days. The birch pollen captured on slide glasses was imaged by a microscope. We first developed the automatic pollen counting technique by applying a machine learning algorithm YOLO to the images. This technique was validated by comparison with subjective counting, and we successfully achieved the automatic counting that has never been done before. The results suggested that the pollen count was highest in the point 200 m downstream from the tree line and diurnal variations were observed at all distances. A simple linear regression analysis of pollen count and meteorological factors revealed a significant positive correlation with temperature. Additionally, a positive correlation with wind speed was also found only at the point just below the tree line. The large-eddy simulation with the pollen advection supported the observation results, though the pollen deposition position was more concentrated near the tree in the simulation.