Flow variability and stochastic dispersion in street networks

• Ano de defesa: 2025
• Autor(a) principal: Klippel, Karine
• Orientador(a): Goulart, Elisa Valentim
• Banca de defesa: Furieri Bruno , Santos, Jane Meri , Soulhac, Lionel , Carpentieri, Matteo
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Instituição de defesa: Universidade Federal do Espírito Santo; Doutorado em Engenharia Ambiental
• Programa de Pós-Graduação: Programa de Pós-Graduação em Engenharia Ambiental
• Departamento: Centro Tecnológico
• País: BR
• Palavras-chave em Português: Ar - Poluição; Turbulência atmosférica; Fluidodinâmica computacional; Processo estocástico; Stochastic processes
• Palavras-chave em Inglês: Urban areas; Turbulence; Scalar dispersion
• Área do conhecimento CNPq: Engenharia Sanitária
• Link de acesso: http://repositorio.ufes.br/handle/10/20832

This study investigates flow variability and stochasticity across different scales and their impact on passive scalar dispersion in a street network. Direct numerical simulations (DNS) data from the DIPLOS project were used for a regular array of rectangular buildings with inflow at a 45° angle to the streets. Puff and continuous sources were positioned near the ground in three locations: short street (S1), intersection (S2), and long street (S3). Simulations were compared to wind tunnel (WT) measurements for flow and scalar dispersion from source S2. The research addresses: (i) street-network scale variability, focusing on sources of flow variability, their effects on scalar releases, and uncertainties from experimental setups; (ii) intersection-scale variability, analysing flow switching and its influence on continuous releases; and (iii) concentration fluctuations, investigating spatial dependence and statistical modelling. Key findings include: (i) flow variability was observed at multiple scales, from small-scale intra-street variations linked to local flow topology, to inter-street differences, street-network-scale variability, and larger-scale variations associated with above-canopy structures. DNS and WT comparisons agreed for flow statistics and mean concentration profiles in continuous releases but differed significantly for puff releases, attributed to flow variability, setup discrepancies, and experimental uncertainties. An implication of these results is that for singular events like accidental releases, characterizing uncertainties is more meaningful than computing ensemble averages. (ii) At intersections, bistable horizontal wind direction switching was observed near the ground (z = 0.125h and z= 0.25h), with peaks at approximately -10° and 110°, offset from the expected alignment with the short (0°) and long (90°) streets. The average switching timescale was 10.5T, where T is the eddy turnover time, with waiting times of 5T and 3T per state. Strong wind direction correlation in diagonally neighbouring intersections was observed with a time lag below 1T. Wind direction switching significantly influenced scalar concentrations from continuous sources, with nearfield anticorrelation in short streets and weaker far-field correlations, highlighting the role of street-network structures and mixing processes. (iii) Clustering analysis of concentration fluctuations identified three regions within the plume with distinct distributions: (1) plume edge, with exponential-like distributions, high intermittency, and extreme skewness and kurtosis; (2) transition region, with asymmetric distributions and reduced intermittency; and (3) plume centre, exhibiting Gaussian-like distributions with negligible intermittency and near-zero skewness. Gamma, Beta, Lognormal, and Weibull distributions were evaluated for modelling these fluctuations. Gamma was the most consistent, capturing distribution shapes and performing well for the 50th and 98th percentiles of the inverse cumulative density function. While Gamma excelled in variance predictions in high-fluctuation regions (cluster 1) and Beta performed better for skewness and kurtosis in low-fluctuation areas (cluster 3), all models struggled with higherorder moments in cluster 1, highlighting challenges in modelling fat-tailed distributions.