Research Paper Analysis
Skillful Subseasonal-to-Seasonal Forecasting of Extreme Events with a Multi-Sphere Coupled Probabilistic Model
Bin Mu†, Yuxuan Chen†, Shijin Yuan¹*, Bo Qin¹,²*, Hao Guo¹
Abstract: Accurate subseasonal-to-seasonal (S2S) prediction of extreme events is critical for resource planning and disaster mitigation under accelerating climate change. However, such predictions remain challenging due to complex multi-sphere interactions and intrinsic atmospheric uncertainty. Here we present TianXing-S2S, a multi-sphere coupled probabilistic model for global S2S daily ensemble forecast. TianXing-S2S first encodes diverse multi-sphere predictors into a compact latent space, then employs a diffusion model to generate daily ensemble forecasts. A novel coupling module based on optimal transport (OT) is incorporated in the denoiser to optimize the interactions between atmospheric and multi-sphere boundary conditions. Across key atmospheric variables, TianXing-S2S outperforms both the European Centre for Medium-Range Weather Forecasts (ECMWF) S2S system and FuXi-S2S in 45-day daily-mean ensemble forecasts at 1.5°×1.5° resolution. Our model achieves skillful subseasonal prediction of extreme events including heat waves and anomalous precipitation, identifying soil moisture as a critical precursor signal. Furthermore, we demonstrate that TianXing-S2S can generate stable rollout forecasts up to 180 days, establishing a robust framework for S2S research in a warming world.
Keywords: subseasonal-to-seasonal forecast, extreme event, deep learning, optimal transport
Executive Impact: Revolutionizing S2S Forecasting
TianXing-S2S offers a groundbreaking approach to subseasonal-to-seasonal (S2S) forecasting, particularly for extreme events. By integrating a vast array of Earth system variables with advanced AI, it delivers superior accuracy and extended predictability, crucial for strategic planning and disaster preparedness in a changing climate.
0
CRPS Improvement
0
Stable Forecast Horizon
0
Daily Ensemble Forecasts
0
Multi-Sphere Variables
Deep Analysis & Enterprise Applications
Select a topic to dive deeper, then explore the specific findings from the research, rebuilt as interactive, enterprise-focused modules.
The Challenge of Subseasonal-to-Seasonal Forecasting
Accurate prediction of extreme weather events on subseasonal-to-seasonal (S2S) timescales (2 weeks to 2 months) is critical for societal resilience and resource management. Current deep learning models excel at short-range weather but struggle with S2S due to complex multi-sphere interactions (atmosphere, ocean, land) and inherent atmospheric uncertainty. Existing probabilistic models often fall short in capturing extreme tail distributions, which are most impactful.
TianXing-S2S: A Multi-Sphere Coupled Probabilistic Model
TianXing-S2S employs a two-stage Latent Diffusion Model (LDM) architecture. Stage 1: VQVAE Embedding compresses 81 diverse multi-sphere predictors into a compact latent space using Vector-Quantized Variational Autoencoders (VQVAEs). This allows for sphere-specific feature extraction while preserving crucial modal characteristics. Stage 2: Diffusion-based Probabilistic Modeling then uses a diffusion model to generate daily ensemble forecasts in this latent space. A key innovation is the Optimal Transport Block (OTB) within the denoiser network, which leverages optimal transport theory to optimize physically meaningful information exchange between atmospheric variables and multi-sphere boundary conditions, ensuring robust ensemble spread.
Demonstrated Skill and Interpretability
TianXing-S2S significantly outperforms ECMWF S2S and FuXi-S2S in 45-day daily-mean ensemble forecasts across key atmospheric variables, showing up to 12% improvement in CRPS. It achieves superior Brier Skill Scores (BSS) for extreme events like heat waves and anomalous precipitation, maintaining positive BSS beyond 10 days where benchmarks degrade. The model's ensemble reliability (SSR) is notably better calibrated. Furthermore, attribution analysis reveals that TianXing-S2S effectively identifies soil moisture as a critical precursor signal for East Asian extreme events, aligning with physical understanding. The OTB also demonstrates physically consistent inference processes, with attention expanding from local to global scales as lead time increases.
Towards Seamless Weather-Climate Prediction
TianXing-S2S represents a significant step towards more interpretable and skillful S2S forecasting systems. Its ability to generate stable rollout forecasts up to 180 days, accurately reproducing seasonal cycles, opens new avenues for developing seamless weather-climate prediction systems. The innovative OTB and multi-sphere integration provide a robust framework for understanding complex Earth system interactions. Future work will focus on specialized models for individual Earth system components and further enhancing ensemble forecast skill by incorporating different types of perturbations.
TianXing-S2S Core Forecasting Process
Multi-Sphere VQVAE Embedding
→
Compact Latent Space Representation
→
Diffusion Model with Optimal Transport Block (OTB)
→
Daily Ensemble Forecast Generation
→
Autoregressive Long-Term Rollout
12%
CRPS Improvement over Benchmarks (beyond 10 days)
180
Days of Stable Rollout Forecasts Achieved
| Feature | TianXing-S2S | ECMWF S2S | FuXi-S2S |
|---|---|---|---|
| Total Variables Integrated |
|
|
|
| Core Probabilistic Method |
|
|
|
| Extreme Event Prediction (BSS) |
|
|
|
| Ensemble Reliability (SSR) |
|
|
|
| Key Coupling Mechanism |
|
|
|
| Max Stable Forecast Length |
|
|
|
Case Study: Predicting East Asian Heat Waves
TianXing-S2S demonstrates exceptional skill in forecasting the 2018 Sichuan Basin heat wave at 4-week lead times. It accurately captured the spatial patterns and intensity, outperforming both ECMWF S2S and FuXi-S2S. Crucially, the model identified soil moisture anomalies as a key precursor signal, highlighting the importance of land-atmosphere feedback in extreme event prediction.
Key Takeaway: TianXing-S2S's ability to identify and leverage physically meaningful precursor signals like soil moisture is vital for early extreme event warnings.
Calculate Your Potential AI ROI
Estimate the tangible benefits of integrating advanced AI forecasting into your enterprise operations.
Estimated Annual Savings
$0
Productive Hours Reclaimed Annually
0
Your AI Implementation Roadmap
A structured approach to integrating cutting-edge AI for maximum impact and minimal disruption.
Phase 1: Discovery & Strategy Alignment
Collaborative workshops to identify key forecasting needs, data availability, and strategic objectives. Define success metrics and a phased implementation plan.
Phase 2: Data Integration & Model Adaptation
Secure and preprocess your multi-sphere data. Adapt TianXing-S2S architecture and train the model on your specific datasets, including fine-tuning the OTB for optimal local coupling.
Phase 3: Pilot Deployment & Validation
Deploy the model in a controlled environment. Conduct rigorous backtesting and real-time validation against historical data and current benchmarks. Gather initial user feedback.
Phase 4: Full-Scale Integration & Operationalization
Integrate TianXing-S2S into your existing operational workflows. Develop intuitive dashboards and alert systems. Provide comprehensive training for your teams.
Phase 5: Continuous Optimization & Expansion
Monitor model performance, implement iterative improvements, and explore opportunities to expand forecasting capabilities to new regions or variables.
Ready to Transform Your Forecasting?
Connect with our experts to explore how TianXing-S2S can enhance your enterprise's predictive capabilities and mitigate risks from extreme events.
Ready to Get Started?
Book Your Free Consultation.
Let's Discuss Your AI Strategy!
Lets Discuss Your Needs
Select a Date
Sun
Mon
Tue
Wed
Thu
Fri
Sat