Global Evaluation of AI and NWP Precipitation Forecasts During Atmospheric River Events

Marina Vicens-Miquel; Taylor Mandelbaum; Amy McGovern; Aaron J. Hill; Daniel Rothenberg

Home/Archive/arXiv

arXiv · 2610.03758

Global Evaluation of AI and NWP Precipitation Forecasts During Atmospheric River Events

Marina Vicens-Miquel·Taylor Mandelbaum·Amy McGovern·Aaron J. Hill·Daniel Rothenberg

View original source ↗Read source document ↗Export metadata ↓

Abstract

Atmospheric rivers (ARs) produce many of the world's most extreme precipitation events and hydrometeorological hazards. Although artificial intelligence weather prediction (AIWP) models have demonstrated skill comparable to or exceeding numerical weather prediction (NWP) systems for large-scale atmospheric variables, their ability to forecast AR-related precipitation remains insufficiently characterized globally. Here, we evaluate 24-hour precipitation forecasts from the Global Forecast System (GFS), Global Ensemble Forecast System (GEFS), GraphCast, and Artificial Intelligence Forecasting System (AIFS) from Day 1 through Day 10 globally and across North America, Europe, and Australia and New Zealand. Using the Extreme Weather Bench framework, forecasts are evaluated against Integrated Multi-satellitE Retrievals for GPM (IMERG) observations using measures of precipitation magnitude, spatial structure, and localization. GraphCast and AIFS exhibit greater spatial skill than GFS and GEFS, particularly for heavy precipitation and at longer lead times, and better preserve the spatial organization of AR-related precipitation through Day 10. However, this improved spatial skill does not translate into accurate precipitation magnitudes. AIWP models tend to overpredict moderate-to-heavy accumulations while underpredicting the heaviest precipitation at longer lead times, whereas NWP systems develop pronounced dry biases. These results reveal distinct strengths and limitations of AIWP for high-impact precipitation forecasting and provide a reproducible benchmark.

Explore related subjects

physics.ao-ph cs.AI

Keep this discovery

Explore connections, maps & timelines

BibTeX RIS

Marina Vicens-Miquel, Taylor Mandelbaum, Amy McGovern, Aaron J. Hill, Daniel Rothenberg. 2026-09-26. Global Evaluation of AI and NWP Precipitation Forecasts During Atmospheric River Events. https://arxiv.org/abs/2610.03758

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related papers

arXiv2026

Arctic bimodality reflects enhanced variability in the clear-cloudy transition rather than bistability

The wintertime Arctic boundary layer is characterized by two distinct states, radiatively clear (cloud-free) and radiatively opaque (cloudy), which is reflected in observations as a bimodal distribution. Previous studies hypothesized that these regimes are related to an underlying bistability of the system's dynamics. We apply a drift-diffusion framework based on the two leading terms of the Kramers-Moyal expansion to observations from 15 winter periods in Ny-Alesund, Norway. Our model captures the bimodality; we find a single stable point for the deterministic drift, however, which questions the existence of bistable deterministic dynamics. Instead, the observed regimes can be attributed to state-dependent stochastic fluctuations with maximum amplitude close to the clear-cloudy transition. The position of the stable point depends on the meteorological conditions. Controlling for synoptic variability does not remove the state-dependence of diffusion, however. The cloudy regime is associated with southwesterly winds, high specific humidities, and uplifting motion, the latter two being conducive for cloud formation, while the cloud-free regime occurs under drier conditions and northeasterly winds. Through an illustrative example, we suggest a possible process-based interpretation of our statistical analysis. The nonlinear onset of cloud formation, once atmospheric humidity reaches saturation, could map state-independent fluctuations in humidity to state-dependent fluctuations in cloudiness.

physics.ao-ph↗

arXiv2026

Improved Precipitation Variability in the Community Atmosphere Model through Implementation of a Neural-Network Deep Convection Scheme

Conventional deep convection schemes remain a primary source of uncertainty in global climate models, yet replacing them with data-driven alternatives while maintaining physical consistency and numerical stability is an open challenge. We present a hybrid configuration of the Community Atmosphere Model version 6 (CAM6), in which a fully connected deep neural network replaces the Zhang-McFarlane deep convection scheme (ZM), while radiation, turbulence, and other cloud processes are left unchanged. The network is trained on output from a storm-resolving global model - the global System for Atmospheric Modeling (gSAM) - using a coarse-graining approach to derive subgrid-scale convective fluxes and tendencies. This data-driven parameterization is ported to CAM6 and found to run stably for more than a decade while conserving energy and moisture. It produces a broadly realistic climatology without retuning the remaining parameterizations. Relative to the default Zhang-McFarlane scheme, the neural network parameterization substantially improves some long-standing model deficiencies. In particular, the biases in light precipitation and extreme precipitation are markedly reduced, and the diurnal cycle of tropical land precipitation shifts from a spurious midday peak over many regions to the observed late-afternoon peak. These results support targeted ML-based convection schemes as a viable path toward reducing systematic model errors in next-generation climate models.

physics.ao-ph↗

arXiv2026

MAG-Net: Radar-Satellite Fusion for Severe Convective Nowcasting with Matched Evaluation of Satellite Contributions

Radar-based nowcasting remains challenging for intense convective echoes because extrapolation cannot represent storm growth and decay, while regression-trained deep networks tend to smooth small-scale high-reflectivity structures. We present MAG-Net, a radar-satellite nowcasting framework that combines composite radar reflectivity with three Fengyun-4A (FY-4A) Advanced Geostationary Radiation Imager (AGRI) infrared fields through separate encoders and a Dual-Head decoder with Gradient-Preserving Fusion (GPF). We evaluate 73,928 spatial forecast windows, each covering a 90-minute forecast, from hourly initialization times across 122 severe-convective outbreaks over southeastern China in 2023. Matched variants separate the effects of output design and satellite input. For radar-only input, the critical success index (CSI) at 40 dBZ increases from 0.092 with regression output to 0.252 with Dual-Head/GPF. Adding AGRI fields under the same output design raises CSI at 40 dBZ to 0.273. The paired gain is 0.021, with a 95% confidence interval of 0.018-0.024 based on outbreak-level resampling. Object-based lifecycle evaluation shows satellite gains across growth, mature, and decay stages. At the 90-minute lead, the largest relative gain occurs before the birth of a new radar core, when radar-only skill is lowest. These findings show that the output formulation and the satellite observations contribute jointly, with the larger share of the severe-echo improvement coming from the output formulation, and identify when the satellite contribution is most pronounced.

physics.ao-ph↗