Uncertainty Quantification of Wind Gust Predictions in the Northeast US: An Evidential Neural Network and Explainable Artificial Intelligence Approach

2025-02-01

AI Summary

Forecasting extreme, right-skewed gusts associated with extratropical cyclones in the U.S. Northeast suffers from large systematic biases and poorly quantified uncertainty. Method: This study introduces evidential neural networks (ENN) for gust forecast uncertainty quantification (UQ), enabling physically interpretable probabilistic prediction intervals without ensemble modeling. Integrating WRF-based feature engineering with SHAP and LIME explainable AI techniques, we identify storm intensity and spatial gust gradients as key drivers of predictive uncertainty. Contribution/Results: Evaluated on 61 historical cases, the framework significantly improves forecast reliability—yielding well-calibrated prediction intervals—and provides operationally actionable, confidence-aware decision support for emergency response.

Technology Category

• Machine Learning: Calibration & Uncertainty Quantification
• Reasoning under Uncertainty: Uncertainty Representations
• Knowledge Representation and Reasoning: Geometric, Spatial, and Temporal Reasoning

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• User Modeling, Personalization and Recommendation: Explainable and interpretable methods for personalization

Abstract

Machine learning has shown promise in reducing bias in numerical weather model predictions of wind gusts. Yet, they underperform to predict high gusts even with additional observations due to the right-skewed distribution of gusts. Uncertainty quantification (UQ) addresses this by identifying when predictions are reliable or needs cautious interpretation. Using data from 61 extratropical storms in the Northeastern USA, we introduce evidential neural network (ENN) as a novel approach for UQ in gust predictions, leveraging atmospheric variables from the Weather Research and Forecasting (WRF) model as features and gust observations as targets. Explainable artificial intelligence (XAI) techniques demonstrated that key predictive features also contributed to higher uncertainty. Estimated uncertainty correlated with storm intensity and spatial gust gradients. ENN allowed constructing gust prediction intervals without requiring an ensemble. From an operational perspective, providing gust forecasts with quantified uncertainty enhances stakeholders' confidence in risk assessment and response planning for extreme gust events.

Problem

• Wind Prediction
• Accuracy Improvement
• Uncertainty Quantification

Innovation

• Evidence Neural Networks (ENN)
• Interpretable AI
• Uncertainty Quantification