Development of probabilistic assessment framework for pedestrian wind environment using Bayesian technique

Abstract This study extends the assessment framework proposed by Murakami et al. (1986) for the pedestrian wind environment to a fully probabilistic method by implementing Bayesian modeling. The method quantifies the uncertainties in constructed models based on the measured wind data and the results of the assessment. To model the probabilities of the daily maximum mean wind speed and wind direction, we employed the Weibull distribution and categorical distribution, respectively. The parameters defining the probability distributions were probabilistically modeled using Bayesian techniques. Using the wind data measured at the Meteorological Observatory of Tokyo, we demonstrated the effectiveness of the proposed method. The results showed that the wind direction probability and each parameter of the Weibull distribution could be estimated in the form of a posterior probability density function. Using the constructed models, we predicted the exceedance probability of the daily maximum instantaneous wind speed and evaluated the wind environment index (rank) in a city model. We provided a discrete rank scale in the form of a probability distribution, which enables us to quantify the evaluation uncertainties. Additionally, we clarified the effect of varying the amount of data used for the model construction. The uncertainty of the exceedance probability decreased with the amount of data. When only 1 year of data was used, some evaluation points possibly changed over three ranks. Even when 5-year observation data was used, the evaluated rank of some points varied within the range of uncertainty, thereby highlighting the importance of uncertainty quantification in the wind environment assessment.