The least-squares (LS) method is commonly employed to retrieve the 3-D wind field by joint detection with multiple sensors, but it tends to suffer from ill-conditioning and reduced accuracy when the sensors are closely deployed and heterogeneous. This letter proposes a new method, denoted as TSVD-LSBE to address the ill-conditioning and sensor heterogeneity problem. First, the LS algorithm, regulated by the truncated singular value decomposition (TSVD), is utilized to get a preliminary estimation of the wind field. Then, the modified Bayesian estimation (BE), which is also regulated by TSVD and uses the modified LS result as the background wind, is employed to tune the preliminary estimation to a more accurate one. Simulation results demonstrate that the new method can enhance the retrieval accuracy by up to 50% in various wind field detection scenarios, surpassing both the modified-LS method and the local and global variational method (LGVM) based on single sensor measurement.

Shape change detection is crucial for monitoring structural defect evolution. However, variations in temperature and precipitation dynamically alter the dielectric properties of structural defects and underground soil media, impacting the position and shape of scattering curves in ground penetrating radar (GPR) Bscan. This poses a challenge in determining the shape of the changing area by using dual-time GPR Bscan comparisons. To extract the structural defect features and shape change detection under changing backgrounds, this article proposes a change detection network based on GPR Bscan (GPR_CDNet). This method includes an encoder–decoder structure. In the encoder, we design a multiscale and multidilated convolution (MSMDconv) to construct a pseudo-Siamese backbone feature extraction network to enhance contextual understanding and capture fine-grained details. Furthermore, to improve the accuracy of reconstructing the shape of the changing region, the feature transformation module (FTM) is designed to convert Bscan features into spatial model features to enhance the representation of change feature information. The Bscan features are then fused with the spatial model features and input into the decoder to reconstruct the shape of the changing area. In addition, this article conducts experiments in the background medium and structural defect relative permittivity changes, rebar shielding, and actual sandbox scenarios. The results show that the network can adapt to the changes in the background medium’s relative permittivity and the structural defects’ relative permittivity and reconstruct the regional shape of structural defects that change over time with better$F_{1}$performance and less computational complexity.

Highly variable wind fields (HWFs), which usually have drastically changing velocities over time, can seriously impact aviation safety, wind energy assessment, and so on. A recently proposed deep learning method can well predict the wind fields in ordinary cases, but its performance deteriorates when HWFs are involved. In this letter, a nowcasting method taking into account the impact of HWFs is proposed. First, standard deviations (SDs) of the lidar observations within a time interval are used to identify highly variable events. Second, a loss function weighted by the SDs is adopted to train the nowcasting neural network. Additionally, a new dimensionless metric is introduced to quantitatively measure the nowcasting performance with emphasis on HWFs. Experimental results demonstrate that the weighted loss function can efficiently improve the nowcasting performance on HWFs such as the initiation and dissipation processes of wind shear. Compared with the original nowcasting method, the network with the weighted loss function can reduce the nowcasting errors by an improvement of more than 15.2% in terms of the new metric.

To better describe the main features of the complex airflow in the atmospheric boundary layer, a hybrid wind field retrieval method based on machine learning (ML) and data assimilation (DA) is proposed. Based on the joint measurement of lidar andin situmeasurements, a 3-D variational data assimilation (3DVAR) method is used to retrieve the fine-scale wind field. To address the iterative interpolation problem in the traditional DA methods, this article isolates the interpolation from the optimization and uses the regression methods in ML to estimate the interpolated observations on analysis grids. More specifically, the supervised regression and semisupervised regression are, respectively, used for lidar andin situobservations according to their heterogeneity. Simulation and field measurement results indicate that, compared with the traditional DA methods, the proposed method can better estimate both 2-D and 3-D velocities, by an improvement of more than 42.3% on average.

The uniform and linear wind models have been commonly used for wind field retrieval in meteorological community. However, the accuracy and robustness of the retrieval results can be quite unsatisfactory due to the mismatch between these models and the real wind distribution, especially under complex wind conditions. In this article, a nonlinear model based on high-order Taylor expansion is proposed to deal with this limitation, and the combination of ridge regression and decomposition-iteration process (denoted as Ridge-DI method) is further introduced to solve the model with high accuracy and robustness. A case study on simulation and field experiment shows that the proposed method with the third-order Taylor expansion can reduce the mean root-mean-square errors (RMSEs) of the retrieved velocities by more than 16.84% in comparison with traditional methods.