Dataset of microscale atmospheric flow and pollutant concentration large-eddy simulations for varying mesoscale meteorological forcing in an idealized urban environment

By 2050, two-thirds of the world's population will live in urban areas under climate change, exacerbating the environmental and public health risks associated with poor air quality and urban heat island effects. Assessing these risks requires the development of microscale meteorological models that quickly and accurately predict wind velocity and pollutant concentration with high resolution, as the heterogeneity of urban environments leads to complex wind patterns and strong pollutant concentration gradients. Computational Fluid Dynamics (CFD) has emerged as a powerful tool to address this challenge by providing obstacle-resolved flow and dispersion predictions. However, CFD models are very expensive and require intensive computing resources, which can hinder their systematic use in practical engineering applications. They are also subject to significant uncertainties, particularly those arising from the mesoscale meteorological forcing and the internal variability of the atmospheric boundary layer, some of which are aleatory and thereby irreducible. Given these issues, the construction of CFD datasets that account for uncertainty would be an interesting avenue of research for microscale atmospheric science. In this context, we present the PPMLES (Perturbed-Parameter ensemble of MUST Large-Eddy Simulations) dataset, which consists of 200 large-eddy simulations (LES) characterizing the complex interactions between the turbulent airflow, the tracer dispersion, and an idealized urban environment. These simulations reproduce the canonical MUST dispersion field campaign while perturbing the model's mesoscale meteorological forcing parameters. PPMLES includes time series at human height within the built environment to track wind velocity and pollutant release and dispersion over time. PPMLES also includes complete 3-D fields of first- and second-order temporal statistics of the wind velocity and pollutant concentration, with a sub-metric resolution. The uncertainty of the fields induced by the internal variability of the atmospheric boundary layer is also provided. The computation of PPMLES required significant resources, consuming 6 million CPU core hours, equivalent to the emission of approximately 10 tCO2eq of greenhouse gases. This significant computational effort and associated carbon footprint motivates the sharing of the data generated. The added value of the PPMLES dataset is twofold. First, the perturbed-parameter ensemble of LES enables to quantify and understand the effects of the mesoscale meteorological forcing and the internal variability of the atmospheric boundary layer, which has been identified as a major challenge in predicting atmospheric flow and pollutant dispersion in urban environments. Secondly, PPMLES reference data can be used to benchmark models of different levels of complexity, and to extract key information about the physical processes involved to inform more operational modeling approaches, for example through learning surrogate models.