With support from ECT’s Small Grants Scheme, PhD student Sophie Mills from the University of Birmingham attended this year’s annual meeting of the European Geosciences Union (EGU23) to present findings from novel approaches to airborne tree pollen analysis at the BIFoR-FACE experiment in Staffordshire. We invited Sophie to write about her presentation at the conference, the wider applicability of her work to LTEs more generally, and to provide any reflections on the wider talks and sessions at EGU23. Readers are invited to note that this represents the last conference travel grant to be awarded under ECT’s Small Grants Scheme. In future, travel grant requests will only be eligible for collaborative research purposes at LTEs.
Pollen are a key component of the biosphere and atmosphere, transporting genetic information from plants and even affecting cloud processes and precipitation, yet what do we know about their existence in the atmosphere? Many people suffer from hay fever and pollen-related allergies, but how can we measure it? Current methods are severely limited by time, labour and cost and produce sparse snapshots at best. Alternative methods are needed that address these limitations for the benefit of public health and better understanding of plant phenology and colonisation.
At the 2023 conference of the European Geosciences Union (EGU23) in Vienna last month, I presented our recent work on low-cost monitoring of pollen using particle size data and machine learning methods. We demonstrated the application of our novel method to data collected at an ECT site: the BIFoR-FACE facility in Staffordshire. Using this data, we hope to gain better insight into airborne pollen trends and the effects of future climate and CO₂ levels.
We trained models to estimate concentrations of grass, oak, birch, pine, and total pollen from the outputs of an Optical Particle Counter (OPC) and, in some cases, temperature and relative humidity. This methodology would be useful to supply remotely accessible, real-time, highly time-resolved data on pollen concentrations at affordable costs across sensor networks that would offer spatial resolutions not previously possible. The sensors could supplement national pollen monitoring networks and provide insight into pollen trends across specific local environments, such as forests, grasslands, urban areas or indoors.
Employing explainable Artificial Intelligence (xAI) methods, we can see which particle size ranges (and meteorological variables) are positively (red) and negatively (blue) correlated with concentrations of different pollen types, visualised in the matrix plot below.
We collected OPC data from the CO₂ treatment arrays and flux tower at the BIFoR-FACE facility and applied our fitted models to estimate pollen concentrations at the site. Analysis of pollen concentrations in the treatment arrays is ongoing, but initial results from sensors positioned at different heights up the flux tower in the forest are interesting.
Using the methodology we developed, we can observe the local pollen trends for different taxa and, for example, oak pollen release events as the season progressed (see plot below). We found that at greater heights above the canopy, at 30m (green) and 40m (red), oak pollen concentrations were higher at peak events at the start of the season. As the season progressed, pollen concentrations became more dominant at lower heights at 10m (blue) and 20m (yellow). This can be explained by observations that catkins, which release pollen, mature first from the top of the canopy where there is more sunlight, while catkins lower in the canopy mature later in the season.
It was exciting to share our work involving the BIFoR-FACE facility at the EGU23 conference, and to network with many people from across the globe. I learned a lot and was inspired by various talks, from (bio)aerosols to land/vegetation-atmosphere interactions and machine learning for earth system modelling. In a great debate on what role scientists should take in politics, with inspirational speaker Katharine Hayhoe, societal polarisation was discussed and how best to communicate our knowledge and act for effective change.
We hope our ongoing research will demonstrate the novel wealth of information that can be gained from low-cost instruments paired with AI, to better understand processes and interactions occurring within ecosystems and long-term manipulation experiments.
