More Like this

1. Atmospheric Predictability beyond 30 Days with Machine Learning

   https://doi.org/10.1175/AIES-D-26-0009.1  

   Vonich, P Trent; Hakim, Gregory J (July 2026, Artificial Intelligence for the Earth Systems)

   Abstract Atmospheric predictability research has long held that rapid error growth at small spatial scales imposes an intrinsic limit of roughly 2 weeks on deterministic weather forecast skill. We challenge this limit using GraphCast, a machine learning weather model, by optimizing initial conditions for twice-daily forecasts spanning 2020. This approach yields an average error reduction of 86% at 10 days relative to control forecasts from reanalysis initial conditions, with skill lasting beyond 30 days. Mean optimal initial-condition perturbations reveal large-scale, spatially coherent corrections primarily reflecting an intensification of the Hadley circulation. Forecasts using GraphCast-optimal initial conditions in the Pangu-Weather model achieve a 21% error reduction, peaking at 4 days, indicating that analysis corrections reflect adjustments that target both model and analysis error. These results demonstrate the existence of initial conditions producing skillful deterministic forecasts far beyond 2 weeks. Whether such initial conditions can be identified in real time for improving operational weather forecasts remains a topic of future research. Significance StatementOur results contest the long-standing view that rapid error growth at small spatial scales fundamentally limits reliable weather forecasts to about 2 weeks. By applying machine learning–based optimization to atmospheric initial conditions, we achieve skillful deterministic hindcasts to at least 33 days—more than twice the conventional limit. Validating these optimized initial conditions in an independent weather model leads to significant but smaller improvements, highlighting the importance of model error in the procedure. On average, the optimal corrections reveal physically consistent large-scale patterns that align with an intensified tropical overturning circulation. These results demonstrate the existence of initial conditions that considerably extend the current established limit of predictability. Real-time determination of such initial conditions for operational forecasting remains an open question. 

   more » « less
2. A Comparison of AI Weather Prediction and Numerical Weather Prediction Models for 1–7-Day Precipitation Forecasts

   https://doi.org/10.1175/WAF-D-24-0081.1  

   Radford, Jacob T; Ebert-Uphoff, Imme; Stewart, Jebb Q (April 2025, Weather and Forecasting)

   Abstract Pure artificial intelligence (AI)-based weather prediction (AIWP) models have made waves within the scientific community and the media, claiming superior performance to numerical weather prediction (NWP) models. However, these models often lack impactful output variables such as precipitation. One exception is Google DeepMind’s GraphCast model, which became the first mainstream AIWP model to predict precipitation, but performed only limited verification. We present an analysis of the ECMWF’s Integrated Forecasting System (IFS)-initialized (GRAP_IFS) and the NCEP’s Global Forecast System (GFS)-initialized (GRAP_GFS) GraphCast precipitation forecasts over the contiguous United States and compare to results from the GFS and IFS models using 1) grid-based, 2) neighborhood, and 3) object-oriented metrics verified against the fifth major global reanalysis produced by ECMWF (ERA5) and the NCEP/Environmental Modeling Center (EMC) stage IV precipitation analysis datasets. We affirmed that GRAP_GFSand GRAP_IFSperform better than the GFS and IFS in terms of root-mean-square error and stable equitable errors in probability space, but the GFS and IFS precipitation distributions more closely align with the ERA5 and stage IV distributions. Equitable threat score also generally favored GraphCast, particularly for lower accumulation thresholds. Fractions skill score for increasing neighborhood sizes shows greater gains for the GFS and IFS than GraphCast, suggesting the NWP models may have a better handle on intensity but struggle with the location. Object-based verification for GraphCast found positive area biases at low accumulation thresholds and large negative biases at high accumulation thresholds. GRAP_GFSsaw similar performance gains to GRAP_IFSwhen compared to their NWP counterparts, but initializing with the less familiar GFS conditions appeared to lead to an increase in light precipitation. Significance StatementPure artificial intelligence (AI)-based weather prediction (AIWP) has exploded in popularity with promises of better performance and faster run times than numerical weather prediction (NWP) models. However, less attention has been paid to their capability to predict impactful, sensible weather like precipitation, precipitation type, or specific meteorological features. We seek to address this gap by comparing precipitation forecast performance by an AI model called GraphCast to the Global Forecast System (GFS) and the Integrated Forecasting System (IFS) NWP models. While GraphCast does perform better on many verification metrics, it has some limitations for intense precipitation forecasts. In particular, it less frequently predicts intense precipitation events than the GFS or IFS. Overall, this article emphasizes the promise of AIWP while at the same time stresses the need for robust verification by domain experts. 

   more » « less
3. Evaluating the Performance of AI-Based and Traditional NWP Guidance for High-Impact Winter Storms over the U.S. East Coast

   https://doi.org/10.1175/WAF-D-25-0150.1  

   Spallone, Christian A; Greybush, Steven J (June 2026, Weather and Forecasting)

   Abstract Over the last several years, artificial intelligence–based numerical weather prediction (AI-NWP) has emerged to become a promising tool in weather forecasting. Many models, like the ECMWF AI Forecasting System (AIFS) or Google DeepMind’s GraphCast, have been shown to outperform traditional NWP for particular metrics and variables. However, questions remain over their reliability in day-to-day use and when applied to extreme weather events. Further analysis is critical to properly compare these systems with traditional methods. This study will analyze the performance of three leading AI-NWP systems for high-impact winter weather along the U.S. East Coast with a goal of identifying systematic errors or improvements in predictions. Several key variables, including the quantitative precipitation forecast, are examined from both deterministic and ensemble-based perspectives. Evaluation of a collection of storms against reanalysis and other observations affirms the ability for AI-NWP to outperform traditional methods by up to 20%–30% for standard metrics like root-mean-square error (RMSE) over a regional domain. Additionally, AI-NWP is shown to be capable of recognizing key high-amplitude winter storm features, like conveyor belts, even at long forecast lead times (7–10 days). Analysis of an ensemble sensitivity test applied to GraphCast track errors displays physically realistic upper-level drivers when compared against a physics-based ensemble. By examining the performance of traditional NWP and AI-NWP at predicting winter storm features as a function of lead time, the practical predictability of snowstorms is reevaluated. As a result, this work can allow for meteorologists to have higher confidence when utilizing AI-NWP in an operational setting. Significance StatementArtificial intelligence methods for weather prediction are rapidly expanding in popularity among weather forecasters, research meteorologists, and the general meteorological community. Given their rise in use, it is paramount that a robust understanding of the accuracy of these systems is developed in various cases. Many studies from model developers have shown positive performance for these systems on global scales compared to traditional baselines. However, these results often leave out many areas of interest in weather prediction, including for anomalous weather events and when considering smaller spatial scales. This study will attempt to provide clearer insight into some of these questions with respect to one type of high-impact weather event: winter storms in the eastern United States. 

   more » « less
4. Dynamical Tests of a Deep Learning Weather Prediction Model

   https://doi.org/10.1175/AIES-D-23-0090.1  

   Hakim, Gregory J; Masanam, Sanjit (July 2024, Artificial Intelligence for the Earth Systems)

   Abstract Global deep learning weather prediction models have recently been shown to produce forecasts that rival those from physics-based models run at operational centers. It is unclear whether these models have encoded atmospheric dynamics or simply pattern matching that produces the smallest forecast error. Answering this question is crucial to establishing the utility of these models as tools for basic science. Here, we subject one such model, Pangu-Weather, to a set of four classical dynamical experiments that do not resemble the model training data. Localized perturbations to the model output and the initial conditions are added to steady time-averaged conditions, to assess the propagation speed and structural evolution of signals away from the local source. Perturbing the model physics by adding a steady tropical heat source results in a classical Matsuno–Gill response near the heating and planetary waves that radiate into the extratropics. A localized disturbance on the winter-averaged North Pacific jet stream produces realistic extratropical cyclones and fronts, including the spontaneous emergence of polar lows. Perturbing the 500-hPa height field alone yields adjustment from a state of rest to one of wind–pressure balance over ∼6 h. Localized subtropical low pressure systems produce Atlantic hurricanes, provided the initial amplitude exceeds about 4 hPa, and setting the initial humidity to zero eliminates hurricane development. We conclude that the model encodes realistic physics in all experiments and suggest that it can be used as a tool for rapidly testing a wide range of hypotheses. 

   more » « less
5. Initialized Seasonal Prediction with the NCAR Models in the North American Multimodel Ensemble (NMME)

   https://doi.org/10.1175/WAF-D-24-0123.1  

   Becker, Emily; Kirtman, Ben P; Min, Dughong (June 2025, Weather and Forecasting)

   Abstract This study investigates the seasonal prediction capabilities of three models, all developed by the National Science Foundation (NSF) National Center for Atmospheric Research (NCAR) and implemented by the University of Miami, within the North American Multimodel Ensemble (NMME) framework. All three models, Community Climate System Model, version 3 (CCSM3), CCSM4, and Community Earth System Model, version 1 (CESM1), are initialized using the Climate Forecast System Reanalysis (CFSR) and have a common period of 1991–2018. The models’ performance in predicting key climate variables including surface temperature, precipitation, and El Niño–Southern Oscillation (ENSO) teleconnections is assessed. The models’ prediction skill is assessed using the sign test, a robust nonparametric method for comparing forecast errors. CCSM4 succeeded CCSM3 in 2014, bringing a much more accurate representation of global temperature trends and improved prediction of precipitation extremes and 2-m temperature over land. CESM1, introduced in 2023, shows further improvement relative to CCSM4 in the prediction of sea surface temperature in the tropical Pacific and precipitation extremes over land. The improvement in precipitation prediction skill is encouraging, as this field has seen little improvement over the life of the NMME. The modeled similarity to observed ENSO teleconnection patterns of 2-m temperature is somewhat less in CESM1 than in CCSM4, although precipitation teleconnection patterns are similar. CCSM4 and CESM1 show stronger surface temperature trends in the tropical Pacific and Southern Ocean than observed trends over the same period, a common problem for current state-of-the-art climate models with implications for prediction and for climate projections. Significance StatementThis study documents the improvements in seasonal climate prediction across three generations of coupled models developed by the National Science Foundation National Center for Atmospheric Research (NCAR) and implemented within the North American Multimodel Ensemble (NMME) by the University of Miami. Model upgrades are an important aspect of the NMME and have contributed to incremental increases in forecast skill. A thorough and ongoing assessment of individual models is critical to our understanding of the NMME system’s evolution and to future model improvements. 

   more » « less