Assessing the impact of data fusion and data quality on spatiotemporal characteristics of exposure models and probability maps - a case study for hazardous mineral fibres

• Reproducible GIS-based workflow for fibre exposure probability mapping • Data fusion choices strongly alter predicted exposure hotspots • Data source and resolution affect model accuracy in space and time • Applied to naturally occurring asbestos with sparse legacy data • Targets emerging fibre pollutants including erionite Exposure modelling for lesser-studied hazardous mineral fibres is complicated because data is scarce and fundamental knowledge of their distribution, dispersion, and health effects is limited. Despite growing concern of the health impacts associated with inhaling airborne mineral fibres, there are no standardised approaches for selecting, integrating, and resolving the diverse environmental datasets needed to build reliable exposure models, particularly when required datasets are incomplete or unavailable. To address this gap, we evaluate how data fusion choices influence outputs using a constrained dataset, applying a GIS-based data fusion workflow to assess how data type, quality, and integration method shape exposure models. The workflow was applied to a case study in El Dorado Hills (USA), where exposure is linked to the disturbance of naturally occurring asbestos. Historical fibre counts were cleaned, georeferenced, and integrated with multiple environmental datasets of varying spatiotemporal resolution. Our analysis shows that both model performance and probability maps are highly sensitive to data fusion and resolution decisions. Model outputs improved when input layers were refined or supplemented, and different fusion methods led to substantial variation in predicted exposure patterns, even when using the same underlying data. These results highlight the complex interplay between data quality and the spatial patterns of predicted exposure, especially under data constraints. Rather than producing validated exposure estimates, this work demonstrates what can be learned from applying a reproducible workflow when available monitoring data is limited, including how modelling sensitivities emerge and how to identify where future data collection would most improve confidence in model output.