Short-distance dispersion of birch pollen

• We successfully counted birch pollen grains using machine learning. • The total number of pollen counts was highest at 200 m among the sites every 100 m from 0 m to 300 m. • Birch pollen counts had a significant positive correlation with air temperature on a 3-h scale. • Birch pollen counts had a marginally significant positive correlation with wind speed. • The City-LES model overestimated the near-source pollen deposition, but it also reflected accurately the optimum deposition distance. 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 observations were also conducted at the site during the days. The birch pollen captured on slide glasses was imaged by a microscope. We automatically counted birch pollen grains by applying a machine learning algorithm You Only Look Once (YOLO) v5 to the images. 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. The pollen counts in the downstream was correlated with air temperature with a statistical significance, but was correlated with wind speed with a marginal significance. The large-eddy simulation with the pollen advection supported the observation results, though the pollen deposition was more concentrated near the tree in the simulation.