arXiv:2608.30795v1 Announce Type: cross Abstract: End-to-end weather forecasting systems produce skillful global gridded and station forecasts directly from raw Earth observations, replacing the nume...
arXiv:2604. 09041v2 Announce Type: replace-cross Abstract: AI-based weather forecasting now rivals traditional physics-based ensembles, but state-of-the-art (SOTA) models rely on specialized architectures and massive computational budgets, creating a high barrier to entry.
arXiv:2606. 26421v1 Announce Type: new Abstract: State-of-the-art medium-range AI weather models can outperform traditional Numerical Weather Prediction (NWP) but require massive training budgets.
The paper introduces DIAL, a method that dynamically adjusts the native stochasticity of a frozen weather forecasting model to improve ensemble reliability across variables, scales, and lead times. DIAL learns from past forecasts to predict necessary spread adjustments and applies small noise updates during each forecast step, avoiding retraining or full backpropagation. Experiments on FourCastNet 3 show that DIAL outperforms post‑processing and reward‑based baselines, achieving an 8.27% improvement in multiscale CRPS over the unmodified model.
Rescene is a 0.4 M‑parameter wrapper that transforms a frozen 1.5° weather model into a stochastic climate emulator. By anchoring the model to a calendar climatology with a deterministic slow clock and adding a band‑limited AR(1) perturbation tuned to the observed anomaly spectrum, Rescene produces century‑long runs with no detectable drift and realistic variability. A 100‑year, 8‑member experiment shows stable temperature trends, calibrated ensemble spread/skill, and accurate reproduction of day‑to‑day variability and blocking frequency.
The paper introduces SCROLL, a method for forecasting multiple observables in stochastic dynamical systems by composing each observable’s likelihood into per‑task free‑routed last‑layer beliefs on a shared backbone. This approach learns unit‑dependent loss scaling directly from data, enabling accurate predictive variance estimation without separate tuning. Experiments on the Ornstein–Uhlenbeck process, stochastic Lorenz‑63, and real air‑quality data show that SCROLL recovers analytic kernels, achieves superior negative log‑likelihood on state and regime tasks, and maintains calibration while reducing hyper‑parameter search costs.
KNOU‑Flood extends the open AIFL recipe for global streamflow forecasting by retaining a single‑layer LSTM and two‑stage transfer learning while adding a longer 424‑day input window, lead‑aware forecast inputs, and GraphCast as an additional forecast source. The model also introduces a latent‑noise ensemble trained with the fair CRPS, achieving a median KGE′ of 0.65 at lead time 1 and 0.48 at lead time 10 across 2,099 test basins, outperforming AIFL and a 1,024‑unit baseline. The probabilistic head, GraphCast channels, longer window, and mixed fine‑tuning collectively reduce lead‑time decay, improve calibration, and enhance detection of multi‑year floods.
The paper compares two main strategies for producing probabilistic forecasts of physical systems: generative models (e.g., diffusion or flow matching) and ensembles of deterministic models trained with the continuous ranked probability score (CRPS). It introduces a framework that evaluates both methods on diverse 2‑D spatiotemporal systems, measuring empirical coverage of predictive intervals, accuracy, and computational efficiency. The study finds that CRPS‑trained ensembles generally provide more reliable uncertainties and faster inference, while generative models trained in ambient space can match coverage but at higher latency. The authors also release AutoCast, a modular framework for both approaches, and AutoSim, a dataset generation tool.
The paper presents a probabilistic post‑processing framework that uses a conditional Variational Autoencoder (cVAE) with a non‑parametric decoder to adjust seasonal Arctic sea‑ice forecasts. By replacing the Gaussian decoder with a generator and optimizing with Continuous Ranked Probability Score (CRPS), the method produces larger, bias‑corrected ensembles that are better calibrated, more consistent with observations, and exhibit higher resolution and sharper spectral power than standard cVAEs. The approach demonstrates improved forecast skill and reduced errors compared to benchmark predictions.
State-of-the-art medium-range AI weather models can outperform traditional Numerical Weather Prediction (NWP) but require massive training budgets. This restricts usage for under-resourced groups and severely limits fast model iteration.
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.
arXiv:2606. 30821v1 Announce Type: new Abstract: Probabilistic downscaling is the task of modeling the conditional distribution of high-resolution fields given coarse inputs, and is a central challenge to atmospheric science, climate modeling, and other multiscale physical systems.