This article reviews advances in monitoring and nowcasting of severe convective weather (SCW), along with developments in operational nowcasting systems. It focuses on deep learning (DL)-based techniques using multisource data, highlighting associated challenges and opportunities. Based on multisource observations including those from dual-polarization weather radars and geostationary satellites, the monitoring capabilities of SCW types and intensities, convective initiation, and identification and tracking of convective storm cells have been significantly improved using advanced technologies, including storm structural feature recognition, fuzzy logic, and DL. Among these approaches, deep generative models have proven particularly effective, substantially improving the accuracy and extending the lead time of SCW nowcasting. The performance of the China Meteorological Administration’s Severe Weather Analysis and Forecasting (SWAN) 3.0 system continues to advance, with widespread operational adoption across China. Future efforts will leverage higher-resolution observations and numerical weather prediction products at the hundred-meter resolution to enhance the understanding of the underlying mechanisms of SCW development at meso-γ- and microscales. Current purely data-driven AI models are transitioning toward physics-informed frameworks for SCW nowcasting. Integrating forecasters’ operational expertise with state-of-the-art AI technology will further enhance operational capabilities in monitoring and nowcasting extreme SCW events.

Based on the three-dimensional radar mosaic data from April to June in 2016-2019 and the differences in triggering conditions, a classification for warm-sector squall lines with vastly different formation mechanisms during the pre-flood season was proposed. They were divided into two types: squall lines that developed in the warm sector after frontal triggering(Type Ⅰ warm-sector squall lines), and squall lines that were generated and developed in the warm sector(Type Ⅱ warm-sector squall lines). On this basis, the distributions of short-term heavy rainfall and thunderstorm winds were compared in terms of large-scale environmental conditions and mesoscale radar echo characteristics during the occurrence of squall lines. The results showed that: Type Ⅰ warm-sector squall lines were mostly formed in plain areas, while Type Ⅱ warm-sector squall lines were mostly generated on the windward slopes of mountains and coastlines. Both squall lines were accompanied by obvious short-term heavy precipitation, which was easy to cause rainstorm and flood; Type Ⅰ warm-sector squall lines were prone to producing regional thunderstorm winds, while Type Ⅱ warm zone squall lines produce scattered thunderstorm winds. The Type Ⅰ warm-sector squall lines had longer lifecycle, more vigorous convective development, and a faster moving speed, while the Type Ⅱ warm-sector squall lines had shorter lifecycle and a slower moving speed with weaker echo intensity. Both types of squall lines had good water vapor conditions and vertical shear conditions of low-level winds at 0-3km, and the dynamic and thermal conditions are better when Type Ⅰ warm zone squall lines occurred.

In order to improve the capability of precipitation nowcasting of the SWAN (Severe Weather Automatic Nowcasting) operational system, research on the improvement and evaluation of precipitation nowcasting algorithms is carried out, and a new QPF (Quantitative Precipitation Forecast) module is implemented. Firstly, minute rainfall observations are used to increase the frequency of QPE (Quantitative Precipitation Estimation) correction and a correction technology based on rain clusters is proposed to improve the rainfall field. The echo motion vector is then optimized by DIS (Dense Inverse Search) optical flow technology. Finally, precipitation nowcasting is produced by both the improved rainfall field and the optimized echo extrapolation vector. Through the national 1 km resolution echo and precipitation nowcasting test evaluation and case analysis in July 2021, it is found that: (1) The optimized echo motion vector based on DIS optical flow technology can capture different scales of echo motion. The results have better consistency and smoothness, which leads to an improvement of extrapolation. (2) Compared with the SWAN operational system, the TS score of the new QPF module is relatively improved by more than 50% in the 0—1 h forecast, and the bias is much closer to 1; in the 1—2 h forecast, except for heavy rainfall like that of 20 mm/h, the TS score increases by about 20%, and the bias decreases by 1—3. In summary, the proposed QPF module based on the SWAN algorithm development kit has significantly improved the forecast skill compared to the SWAN in both 0—1 h and 1—2 h leading time, and can be applied to operational nowcasting directly.