This study evaluates large language models (LLMs) for predicting weather‑related forced outage risk in a distribution grid using a zero‑shot approach without labeled training data. The task is framed as binary severity classification over 3h, 6h, and 12h horizons, leveraging six years of outage records and high‑resolution weather data from central Texas. Four zero‑shot LLMs are compared to two supervised classifiers under two input settings—current weather observations and forecast data—showing that supervised models lead on macro‑F1 and precision, while newer LLMs achieve competitive scores and offer complementary strengths in reasoning and geographic scalability.
arXiv:2608.24665v1 Announce Type: new Abstract: Power outage prediction models are increasingly used in assessments of climate-driven infrastructure risk, yet current evaluation practices obscure whe...
OutageDiT is a foundation model designed to generate seven‑day power‑outage trajectories at 15‑minute resolution. It uses a condition encoder to process historical context and future covariates, and a shallow flow‑matching decoder to sample complete outage trajectories, enabling point forecasting, uncertainty quantification, and conditional scenario generation. The model improves forecast accuracy and scenario quality over strong baselines and can transfer zero‑shot to held‑out regions, linking outage forecasting to operational planning under uncertainty.
EvoBench is a diagnostic benchmark framework designed to reveal where and when machine‑learning models for environmental variables fail, rather than relying on a single average accuracy score. It tracks model performance over time and across spatial extrapolation, highlighting issues such as non‑stationarity, poor performance at unmonitored sites, and under‑prediction of rare events like floods. In a case study on daily streamflow prediction across 9,000 U.S. basins, EvoBench shows that no model dominates all scenarios and that targeted fixes can mitigate identified weaknesses.
OutageDiT is a generative foundation model that produces seven‑day power‑outage trajectories at quarter‑hour resolution, trained on nationwide outage and weather data. It uses a condition encoder to process historical context and future covariates, and a shallow flow decoder to generate full trajectories, enabling point forecasting, uncertainty quantification, and conditional event simulation. The model outperforms strong baselines on forecasting benchmarks and can transfer zero‑shot to unseen regions, linking outage simulation to operational planning under uncertainty.
arXiv:2606. 04143v1 Announce Type: cross Abstract: Accurate flood forecasting is essential for mitigating disaster risks and protecting communities.
The study evaluates how forecast lead time influences the effectiveness of outage predictions for crew deployment using AI weather models GraphCast and FourCastNet. It finds that the ability to capture outage burden decreases with longer lead times, with a notable drop between three and five days. Early deployment with all requested crews outperforms late deployment with fewer crews, and the study also examines the impact of antecedent land surface predictors and calibration of prediction intervals for tropical cyclone events.
arXiv:2607. 21785v1 Announce Type: cross Abstract: Roadblocks in Bolivia are a social conflict phenomenon with devastating economic impacts, estimated at losses equivalent to 4% of the national Gross Domestic Product.
arXiv:2608.23058v1 Announce Type: new Abstract: Large language models (LLMs) now support forecasting systems that combine language-based reasoning with temporal data, evidence retrieval, external too...
The paper introduces a learned atmospheric critic that discriminates between real weather data and model outputs to produce a realism score. Unlike fixed metrics, the discriminator adapts to the specific failure modes of a given model, effectively detecting various synthetic corruptions in ERA5 data. Experiments show the learned critic outperforms existing metrics and reveals that realism decreases with longer forecast lead times, favoring numerical over machine‑learning models.
arXiv:2609.27197v1 Announce Type: new Abstract: Minimum Risk Training (MRT) enables neural machine translation models to directly optimize sequence-level evaluation metrics instead of relying only on...
The study evaluates crop‑yield forecasting methods for the 2012 Midwestern US drought, comparing non‑deep learning machine learning models with a deep learning model (VITA) using 16 meteorological predictors. It highlights challenges such as distributional dissimilarity between training and test data, spatial and temporal sparsity, and demonstrates that sample weighting and feature selection improve non‑deep learning models but not VITA. The work contrasts deep versus non‑deep learning approaches and shows how modifications can mitigate issues arising from extreme drought conditions.
The paper presents Laxmi, a retrained version of the AIFS weather model that uses satellite-based precipitation observations instead of ERA5 reanalysis data. Laxmi achieves a 19% improvement in global probabilistic accuracy, reduces drizzle overprediction by 33%, and boosts the 95th percentile Brier skill score by 57%. In a case study of 10 Indian tropical storms, Laxmi accurately forecasted 150 mm event-total precipitation in 7 events, outperforming both the original AIFS and the leading physical model IFS.