FuXi-Nowcast: Meet the longstanding challenge of convective initiation in nowcasting

TL;DR

FuXi-Nowcast presents a deep-learning nowcasting framework that jointly forecasts multiple hazards at 1-km resolution by integrating large-scale 3-D atmospheric fields with high-resolution observations. It leverages a multi-task Swin Transformer, a convective signal enhancement module, and a distribution-aware hybrid loss to preserve intense convection and reduce intensity decay. Compared to CMA-MESO 3-km, it achieves higher CSI for reflectivity, precipitation, and wind gusts up to 12 h, with notable gains for heavy rainfall, and case studies demonstrate improved timing and structure of convective initiation. This work shows that coupling 3-D machine-learning forecasts with fine-scale observations can deliver robust, long-lead, multi-hazard nowcasts that outperform current operational systems.

Abstract

Accurate nowcasting of convective storms remains a major challenge for operational forecasting, particularly for convective initiation and the evolution of high-impact rainfall and strong winds. Here we present FuXi-Nowcast, a deep-learning system that jointly predicts composite radar reflectivity, surface precipitation, near-surface temperature, wind speed and wind gusts at 1-km resolution over eastern China. FuXi-Nowcast integrates multi-source observations, such as radar, surface stations and the High-Resolution Land Data Assimilation System (HRLDAS), with three-dimensional atmospheric fields from the machine-learning weather model FuXi-2.0 within a multi-task Swin-Transformer architecture. A convective signal enhancement module and distribution-aware hybrid loss functions are designed to preserve intense convective structures and mitigate the rapid intensity decay common in deep-learning nowcasts. FuXi-Nowcast surpasses the operational CMA-MESO 3-km numerical model in Critical Success Index for reflectivity, precipitation and wind gusts across thresholds and lead times up to 12 h, with the largest gains for heavy rainfall. Case studies further show that FuXi-Nowcast more accurately captures the timing, location and structure of convective initiation and subsequent evolution of convection. These results demonstrate that coupling three-dimensional machine-learning forecasts with high-resolution observations can provide multi-hazard, long-lead nowcasts that outperforms current operational systems.

Figures (5)

• Figure 1: Comparison of the Critical Success Index (CSI) for FuXi-Nowcast (blue line) and CMA-MESO (orange line) in 12-hour forecasts. Rows 1 to 3 correspond to composite reflectivity (CR), total precipitation (TP), and wind gust (WS), respectively. Columns represent three increasing intensity thresholds for each variable (from left to right). Higher CSI values indicate better forecast performance.
• Figure 2: Composite radar reflectivity (dBZ) over East China (29.01°N - 36.68°N, 114.67°E - 123.62°E) for a case initialized at 09 UTC 16 June 2025. Rows from top to bottom show observations, FuXi-Nowcast forecasts, and CMA-MESO forecasts, respectively. Columns from left to right correspond to forecast lead time of 1 to 5 hours, valid from 10 to 14 UTC. The color bar denotes reflectivity values from 0 to 65 dBZ.
• Figure 3: Spatial distribution of ground-based weather stations across Jiangsu Province. The station network consists of fixed station locations, with occasional data gaps at some stations during certain time periods.
• Figure 4: Schematic of the FuXi-Nowcast model. (a) Overall architecture of FuXi-Nowcast, which integrates large-scale atmospheric fields, high-resolution observations, and temporal embeddings as inputs. (b) Structure of the prediction module. (c) Design of the convective signal enhancement module.
• Figure 5: Statistical distributions of key forecast variables in the entire dataset. Columns 1 to 4 in the top row are 2-meter temperature (T2M), 2-meter specific humidity (Q2M), total precipitation (TP), 10-meter u wind component (U10M), 10-meter v wind component (V10M), and columns 1 to 4 in the bottom row are wind speed (WS), wind gusts (GS), composite reflectivity (CR), and total precipitation (TP). The y-axis shows probability density (top row) and log probability density (bottom row).