Data-driven reconstruction of wind speed randomness in an urban area

This paper addresses the problem of characterizing the randomness of wind speed in urban areas. The study proposes two diffusion models to characterize the behavior of the random component of wind speed in an urban area from time series measured hourly by fourteen meteorological stations during the time period 2022–2023. In particular, a basic criterion is established to categorize these stations based on the variance of the underlying stochastic process for the stationary case in order to reduce the number of mathematical models. Kernel-based regression (KBR) was used to estimate the Karmers–Moyal (KM) coefficients associated with the drift and diffusion terms. The numerical solution of the proposed Langevin equation was employed to calculate the statistical properties of the process, taking into account the variance values for station classification. The results show that from the variance values expressed in $$m^{2}/s^{2}$$ , the weather stations can be classified into two types-the former with a variance of less than one and the latter with a variance greater than one. Therefore, only two Langevin models are obtained. This shows that model reduction using basic statistical properties of the time series is feasible when a geographical classification is not possible.