A multi-sensor interval fusion adaptive regularization data assimilation model for wind direction prediction

Real-time forecasting of wind fields is an essential prerequisite for computational fluid predictions of pollutant transport. In the domain of data assimilation for real-time weather forecasting, obtaining high-quality meteorological data measurements poses a challenge that significantly impacts prediction accuracy. Predicting wind direction through data assimilation presents an inverse problem, and low-quality wind direction data resulting from suboptimal sensor placement can lead to ill-posedness when constructing proxy models. Consequently, previous research has extensively investigated the optimal placement of meteorological sensors. However, the data assimilation experiment has thus introduced uncertainties associated with the positions of the sensors. To achieve this goal, this study proposes a adaptive data assimilation model. This model introduces the concept of local convergence intervals on reduced-order response model, and deconstructs ill-posed intervals into well-posed intervals, and obtains a unique solution interval by regularization through the convergence range distance fusing. Finally, the model selects sensors using adaptive local weights, and implements the data assimilation process using inverse Ensemble Kalman Filter. This paper employs data from the Huailai Test Station to design simulated wind direction experiments.The results indicate that the method is capable of overcoming the shortcomings of sensor placement and can enhance the accuracy of prediction.