An Operational Deep Learning System for Satellite-Based High-Resolution Global Nowcasting

TL;DR

Global MetNet introduces an operational, satellite-based precipitation nowcasting system designed to close the global accuracy gap in data-sparse regions. It integratesCORRA-derived training targets, geostationary satellite mosaics, and global NWP data within an encoder–decoder architecture, delivering probabilistic nowcasts at approximately 0.05° spatial and 15-minute temporal resolution for the next 12 hours with sub-minute latency. The approach outperforms industry baselines (HRRR, HRES) across lead times and precipitation rates, particularly enhancing skill in the tropics and Global South, and demonstrates robustness without radar data in many regions. Case studies on deep convective systems, ITCZ MCS, and tropical cyclones illustrate practical benefits for hazard forecasting, while limitations of CORRA as a proxy and biases in extreme events motivate future work to further refine probabilistic forecasts and expand data sources. The system is already deployed for millions of users via Google Search, signaling readiness for broad dissemination and real-world impact on weather risk management.

Abstract

Precipitation nowcasting, which predicts rainfall up to a few hours ahead, is a critical tool for vulnerable communities in the Global South frequently exposed to intense, rapidly developing storms. Timely forecasts provide a crucial window to protect lives and livelihoods. Traditional numerical weather prediction (NWP) methods suffer from high latency, low spatial and temporal resolution, and significant gaps in accuracy across the world. Recent machine learning-based nowcasting methods, common in the Global North, cannot be extended to the Global South due to extremely sparse radar coverage. We present Global MetNet, an operational global machine learning nowcasting model. It leverages the Global Precipitation Mission's CORRA dataset, geostationary satellite data, and global NWP data to predict precipitation for the next 12 hours. The model operates at a high resolution of approximately 0.05° (~5km) spatially and 15 minutes temporally. Global MetNet significantly outperforms industry-standard hourly forecasts and achieves significantly higher skill, making forecasts useful over a much larger area of the world than previously available. Our model demonstrates better skill in data-sparse regions than even the best high-resolution NWP models achieve in the US. Validated using ground radar and satellite data, it shows significant improvements across key metrics like the critical success index and fractions skill score for all precipitation rates and lead times. Crucially, our model generates forecasts in under a minute, making it readily deployable for real-time applications. It is already deployed for millions of users on Google Search. This work represents a key step in reducing global disparities in forecast quality and integrating sparse, high-resolution satellite observations into weather forecasting.

Figures (14)

• Figure 1: Critical Success Index (CSI) at a 1° resolution for the HRES and Global MetNet model at 1 hour lead time for 1.0 mm/hr of precipitation.
• Figure 2: Critical Success Index (CSI) globally and for several regions (Brazil, India, Africa, and the USA), using the GPM CORRA dataset as ground truth at precipitation rates of 0.2 mm/hr (drizzle), 2.4 mm/hr (light rain), 7.0 mm/hr (heavy), and 25.0 mm/hr (very heavy).
• Figure 3: Forecasting Accuracy Gap: Critical Success Index (CSI) of Global MetNet vs. HRES in the Global South and Global North (top), and Tropics and Mid-Latitudes (bottom), validated against the GPM CORRA dataset at rates of $0.2, 1.0, 2.4, 7.0, \text{and } 25.0\,\text{mm/hr}$. Global North includes areas covering USA, Canada, Europe, Japan, and Australia. Global South includes regions covering India, South-east Asia, Middle-east, Africa, Brazil, Mexico, Central America and South America
• Figure 4: Comparison of Critical Success Index (CSI) for HRES and Global MetNet nowcasts at different lead times (3, 6, 9, and 12 hours) for light (1.0 mm/hr) and moderate (2.4 mm/hr) precipitation.
• Figure 5: Critical Success Index (CSI) for Global MetNet models vs. NWP baselines in the US (vs. MRMS), Europe (vs. Opera), and Japan (vs. JMA) at precipitation rates of 0.2, 2.4, 7.0, and 25.0 mm/hr.