Data-Driven Detection of Nocturnal Pollen Fragmentation Triggered by High Humidity in an Urban Environment

Abstract

Biological particulate matter (BioPM) in the urban environment can affect human health and climate. Pollen, a key BioPM component, produces smaller particles when fragmented, significantly impacting public health. However, detecting pollen fragmentation and identifying the meteorological thresholds that trigger it remain largely hypothetical and uncertain. Here, we develop a novel data-driven approach integrating deep learning, efficient clustering methods, and automatic machine learning with explainable methods to identify BioPM components and quantify their environmental drivers. For the first time, we demonstrate the ability to routinely detect pollen fragmentation using only meteorological and online BioPM spectral data. Our findings resolve the previously unclear humidity threshold, confirming that fragmentation is triggered when relative humidity exceeds 90%. Our results find that this humidity-induced fragmentation occurs at night─a critical, yet previously overlooked, time, resulting in the highest pollen concentrations of the day. This critical yet previously unidentified fragmentation phenomenon may have significant health impacts on urban cohorts.