Field measurement data reveals the complexity and natural variability of urban flow phenomena. Validation of our models with field measurements is of fundamental importance to our overall research goal. In addition, the rapid increase in urban sensing network capabilities creates exciting opportunities for using data assimilation methods to reduce uncertainty in the predictions.
The objective of this project is to implement an array of sensors for wind, temperature and air quality on Stanford’s engineering quad. Initial simulations and sensitivity analyses have guided the selection of the sensor locations. A computational fluid dynamics (CFD) model of the Engineering Quad was generated to investigate (1) the potential of using an ensemble Kalman filter for data assimilation, and (2) optimal sensor placement for a field experiment.
The analysis was motivated by the observation that a single measurement of the wind speed and direction at a remote weather station does not necessarily allow for the accurate definition of boundary conditions for CFD simulations. Therefore, our investigation focused on whether data from wind sensors inside the urban canopy can be used to infer statistics for the inflow boundary conditions, using the process shown in Figure 1. First, a polynomial chaos expansion (PCE) surrogate model was constructed based on several Reynolds-averaged Navier-Stokes (RANS) simulations. Subsequently, the ensemble Kalman filter is used to estimate the probability distribution for the uncertain parameters in the inflow boundary conditions. This represents the natural variability in the atmospheric boundary layer, such that the probability distribution can be used for a forward analysis to predict the mean flow field in the urban canopy. Numerical testing of this procedure was performed using artificial data generated by adding noise to the initial simulation results. As shown in Figure 2, optimal selection of the sensor location is essential to minimize errors in the retrieved probability distribution and the corresponding mean flow field prediction when only one sensor is available. These optimal locations are primarily identified further away from buildings. Another solution is to perform data assimilation using multiple sensors, which was shown to quickly improve the prediction, and offers more flexibility in terms of sensor placement.
Measurements using 6 anemometers on the engineering quad started in August 2017. In ongoing work the data will be used for extensive validation of our models, and we will explore the challenges and opportunities that arise when using field measurements for data assimilation to improve the predictions .