Search

In recent years, extreme severe weather has frequently occurred in China, causing heavy casualties and property losses. As a result, a growing focus has been on the early warning and forecasting of tornadoes and other small to medium-scale severe convective weather events as part of national disaster prevention and mitigation efforts. However, the current tornado detection algorithm exhibits limitations in terms of accuracy and warning time. This algorithm relies on the threshold judgment of the new generation weather radar base data in multiple elevation angles and volume scans to obtain the tornado vortex signature (TVS). The TVS algorithm's tornado warning forecasts are characterized by low accuracy, and the early warning time is short due to the sudden onset of tornadoes and their short generation and lysis duration. The integration of a machine learning algorithm with the multiple features of the tornado, including radar reflectivity, radial velocity, and velocity spectrum width, can effectively improve the accuracy and warning time of tornado recognition. In the present study, the tornado detection algorithm based on random forest (TDA-RF) utilizes CINRAD radar historical tornado data as the training set, classifies the training set through the random forest algorithm to generate a tornado prediction model, and employs the prediction model to detect tornadoes in real-time radar data. The test results show that the TDA-RF algorithm can effectively identify tornadoes of varying intensities. Compared with the TVS tornado detection algorithm, it can calculate the classification probability value of the tornado area. TDA-RF does not require the judgment of the spatiotemporal continuity of tornado features. It is less affected by false and invalid echoes, and the early warning time can reach up to 18 minutes. [ABSTRACT FROM AUTHOR]

Improving the resolution of degraded radar echo images of weather radar systems can aid severe weather forecasting and disaster prevention. Previous approaches to this problem include classical super-resolution (SR) algorithms such as iterative back-projection (IBP) and a recent nonlocal self-similarity sparse representation (NSSR) that exploits the data redundancy of radar echo data, etc. However, since radar echoes tend to have rich edge information and contour textures, the textural detail in the reconstructed echoes of traditional approaches is typically absent. Inspired by the recent advances of faster and deeper neural networks, especially the generative adversarial networks (GAN), which are capable of pushing SR solutions to the natural image manifold, we propose using GAN to tackle the problem of weather radar echo super-resolution to achieve better reconstruction performance (measured in peak signal-to-noise ratio (PSNR) and structural similarity index (SSIM)). Using authentic weather radar echo data, we present the experimental results and compare its reconstruction performance with the above-mentioned methods. The experimental results showed that the GAN-based method is capable of generating perceptually superior solutions while achieving higher PSNR/SSIM results. [ABSTRACT FROM AUTHOR]