EPT2-RR: The First Hourly Updated Global AI Weather Model

Today marks a breakthrough in operational weather intelligence: EPT2-RR, the world's first hourly-updating, satellite-initialized global AI weather model. Where existing operational models (both traditional numerical weather prediction and AI-based) update every six hours, EPT2-RR delivers fresh forecasts every hour, with station-validated accuracy that consistently outperforms ECMWF HRES.

EPT2-RR represents a fundamental shift in how weather intelligence operates.

The Operational AI Weather Model Breakthrough

EPT2-RR represents three simultaneous firsts in global weather modeling:

Hourly Global Updates: While regional models like NOAA HRRR provide hourly updates for North America, EPT2-RR is the first global AI weather model to achieve hourly refresh cycles, delivering 24 forecasts per day compared to the industry standard of 4.

Direct Observational Initialization: Unlike other models that rely primarily on processed initial conditions from operational analysis systems or historical reanalysis data, EPT2-RR leverages real-time satellite observations as its primary input, providing a critical temporal advantage in rapidly evolving weather conditions.

Operational Superior Accuracy: Station-validated performance consistently exceeds ECMWF HRES across key variables including 10m wind speed, 100m wind speed, 2m temperature, and shortwave solar radiation.

EPT2-RR delivers 24 updates per day compared to traditional models' 4 updates, providing continuous weather intelligence.

The Latency Advantage: Every Minute Matters

In energy markets where millions of dollars trade on weather-dependent positions, forecast latency determines operational advantage. The contrast is stark:

ECMWF HRES: Runs 4 times daily (00, 06, 12, 18 UTC) with ~5 hours 45 minutes from observation time to first forecast field dissemination.

EPT2-RR: Runs hourly with ~65-85 minutes from observation time to forecast delivery.

Consider a 9 AM UTC energy trading scenario: The most recent available ECMWF HRES forecast is from the 00:00 UTC run (available since ~05:45 UTC), meaning you're working with 9-hour-old atmospheric conditions. EPT2-RR, however, provides forecasts from observations just 1-1.5 hours ago. In volatile weather situations, this 7-8 hour temporal advantage translates directly to trading edge and risk management capability.

Technical Architecture: Direct-to-Forecast Pipeline

EPT2-RR's breakthrough lies in its streamlined observational pathway:

Direct Satellite Initialization → AI Inference → Forecast Delivery

Traditional workflow involves multiple processing steps: Observations → Data Assimilation → Model Integration → Post-processing → Dissemination

By streamlining the data processing pipeline and prioritizing direct satellite observations, EPT2-RR achieves both speed and accuracy advantages. The model integrates real-time satellite radiances, temperature, and humidity profiles through an optimized end-to-end approach that minimizes the computational overhead of traditional processing chains.

EPT2-RR's direct observational pathway compared to traditional NWP processing chains.

Station-Validated Performance

EPT2-RR maintains superior accuracy while dramatically improving update frequency. Validation against International Surface Database (ISD) station measurements demonstrates consistent outperformance of ECMWF HRES:

10m wind speed validation showing EPT2-RR (red) outperforming ECMWF HRES (blue) across forecast lead times.

2m temperature validation demonstrating EPT2-RR's superior accuracy profile compared to ECMWF HRES.

The accuracy advantage extends across all key energy-relevant variables:

• 10m Wind Speed: Lower RMSE across 1-48 hour lead times
• 100m Wind Speed: Critical for wind energy applications
• 2m Temperature: Essential for demand forecasting
• Shortwave Solar Radiation: Key for solar energy prediction

Global station validation results showing EPT2-RR's consistent accuracy advantage.

Model Comparison: EPT2-RR in Context

EPT2-RR uniquely combines hourly global updates with operational deployment and superior accuracy, capabilities that existing models provide only in isolation.

Energy Operations Impact

The hourly update cycle unlocks new operational capabilities:

Intraday Energy Trading

Traditional 6-hour update cycles create information gaps during critical trading windows. EPT2-RR enables continuous strategy adaptation as atmospheric conditions evolve throughout the trading day.

Grid Management & Renewable Integration

Rapid renewable ramp rates (particularly solar photovoltaic) require forecast updates that match the speed of atmospheric changes. EPT2-RR's hourly updates enable grid operators to optimize renewable integration with minimal curtailment.

Risk Management & Hedging

Weather-dependent energy positions require continuous risk assessment. EPT2-RR's forecast freshness enables sophisticated hedging strategies that adapt to evolving weather patterns within the same operating day.

Technical Specifications & Integration

Model Characteristics:

• Update Frequency: Hourly (24 forecasts/day)
• Forecast Horizon: 2 days at 1-hour temporal resolution
• Spatial Resolution: Global coverage at operational resolution
• Latency Estimates: 65-85 minutes from observation time (for full dissemination times, see our documentation)
• Model Type: Deterministic (ensemble variant available separately in EPT-2 family)

Available Variables: For a comprehensive list of available weather parameters, see our parameter availability documentation.

EPT2-RR integrates seamlessly with existing EPT-2 family models through the same API endpoints and data formats, enabling drop-in replacement for existing workflows.

How EPT2-RR Compares to Other Weather Models

How does EPT2-RR differ from GraphCast and Microsoft Aurora? GraphCast and Aurora are research models that update only 4 times per day and are initialized with either GFS or ECMWF initial conditions. They primarily serve medium-range forecasting research. EPT2-RR is operationally deployed with a satellite-driven approach and hourly updates optimized for short-range, time-critical applications. Additionally, the EPT-2 family consistently outperforms both GraphCast and Aurora on station-validated accuracy metrics.

What's the difference versus ECMWF AIFS? ECMWF AIFS is operational but maintains the traditional 6-hourly update cycle with 6-hourly temporal resolution. EPT2-RR provides 6× more frequent updates with hourly temporal resolution and significantly lower latency.

How does EPT2-RR compare to NOAA HRRR? HRRR provides hourly updates but only covers North America and uses traditional NWP methods. EPT2-RR offers global coverage with AI-based accuracy advantages while maintaining comparable update frequency and latency.

Is this an ensemble model? EPT2-RR is deterministic. A separate ensemble version exists within the EPT-2 family for uncertainty quantification applications.

Operational Availability

EPT2-RR is available now through the Jua Earth Intelligence Platform with:

• Real-time access: Live forecasts updated every hour
• Historical validation: Complete archive for backtesting and validation
• API integration: RESTful API with comprehensive documentation
• SDK support: Available through our Python SDK for seamless integration
• Technical support: Implementation guidance and operational support

The model represents a new paradigm in operational weather intelligence where forecast updates match the speed of energy market operations.

Ready to transform your weather intelligence?

• Access EPT2-RR through the Jua Earth Intelligence Platform
• Integration support available for existing workflows
• Contact our team for implementation guidance and technical requirements