A multi-stage probabilistic infrastructure for outage forecasting in the ERCOT grid using GFDL-CM4 climate projections

This paper presents a climate-informed outage forecasting pipeline (ranging from 2025 to 2030) for the ERCOT electric grid as part of a broader series examining climate resilience and grid reliability in Texas infrastructure. Synthesizing projections from the CMIP6 GFDL-CM4 climate model, we developed a three-stage pipeline integrating random forest classification, bilateral logistic regression, and histogram-based gradient boosting. Outage severity was projected to rise 5.1% by 2030, with annual economic losses increasing from $1.4 billion to $1.58 billion. Monte Carlo simulations were introduced to quantify uncertainty. The pipeline performed exceptionally, reaching an F1 of 0.86 in the storm classifier and ROC-AUC scores of 0.86 and 0.89 for non-stormy and stormy cases, respectively, in the bilateral logistic regressions. Geospatial risk analysis identified the Houston metropolitan area and Dallas as major vulnerability hotspots. This study aims to inform long-term resilience planning, and develops a set of model-supported mitigation strategies. These recommendations include grid automation, storm hardening, and parametric insurance mechanisms, tailored for effective integration in ERCOT. The study’s methodology and risk analysis aims to contribute to the transition of grid planning from reactive to proactive, long-term resilience.