GMAO's GEOS Model Performs Well With Back-to-Back Winter Storms

As early as two weeks before it occurred, meteorologists at NASA's Global Modeling and Assimilation Office (GMAO) detected the sort of atmospheric pattern conducive for a large winter storm over the eastern United States using the GEOS-FP (Forward Processing) modeling system. Through analysis of model-generated forecasts and simulated observations, GMAO scientists tracked the evolution of this high-impact weather event, providing a unique opportunity to assess the predictability of major winter storms and validate NASA's Earth system modeling capabilities.

Up until this system, this winter had largely featured weaker, moisture-starved low-pressure systems, often referred to as 'Alberta Clippers' as they frequently originate from the Great Plains in western Canada. This type of storm track is typical of the La Niña phase of the El Niño Southern Oscillation (ENSO) presently ongoing, and systems typically have relatively low precipitation totals. However, a major change in the circulation pattern was about to unfold—one that NASA's models detected well in advance.

Simulated GOES-16 Band 09 brightness temperature (°C) showing atmospheric river moisture transport from the eastern Pacific penetrating deep into a developing winter storm over the eastern U.S. Yellow-orange regions indicate dry mid-level air; white-green regions show moist upper-level clouds. The continuous moisture plume demonstrates efficient vertical transport of water vapor from surface to mid-tropospheric levels (600-400 mb) feeding the cyclone's precipitation shield.

The GMAO's GEOS-FP model consistently showed extremely cold air over Canada would break loose and charge southward deep into the eastern United States, just as the southern sub-tropical jet stream branch was set to direct a large plume of moisture towards the same region. These two air masses – sometimes referred to as the Siberian express and Pineapple express, respectively – were consistently modeled to converge around the weekend of January 24-25, 2026 as far as two weeks prior. GMAO scientists monitored multiple model runs and ensemble forecasts to assess the confidence and uncertainty in this prediction. By the afternoon of Wednesday, January 21st, model consensus had strengthened significantly, and winter storm watches were raised from Arizona, to the Northeast Megalopolis region, and even as far south as Houston, Texas.

Predictability Challenges: The Warm Layer Problem

Though NASA's models clearly captured the massive dome of frigid Arctic high pressure being overridden by the expansive plume of moisture-rich tropical air, significant uncertainty remained regarding exactly which areas would be most heavily impacted, and by what forms of precipitation. This represented a key predictability challenge: forecasting the depth and intensity of a warm layer roughly 2-3 km aloft in the atmosphere. Various model runs showed a wide range of potential precipitation types for the Mid-Atlantic I-95 corridor several days before the storm. GEOS-FP simulated soundings and vertical temperature profiles indicated this warm-layer uncertainty would be critical—less warm air aloft would keep precipitation as snow longer, while greater warm air intrusion would lead to more sleet and freezing rain.

Simulated Observations and Storm Evolution

To capture the fine-scale details of this complex winter storm, GMAO scientists utilized a global downscaling of GEOS-FP analyses from 12.5km to 2km resolution over CONUS. This high-resolution configuration enabled explicit representation of mesoscale precipitation bands, sharp gradients in precipitation type, and terrain-induced effects that are critical for winter weather prediction.

GMAO scientists generated simulated satellite observations from the GEOS 2km output to compare with actual GOES-R ABI imagery as the storm unfolded. Simulated brightness temperatures closely matched observed infrared and water vapor channels, while simulated radar reflectivity fields captured the distinctive "bright band" signature at 2-3 km altitude—the key indicator of the warm layer intrusion that would lead to prolonged sleet. The 2km model's precipitation-type diagnostics accurately predicted the transition from an initial period of moderate to heavy snow early on January 25 to an uncommonly long period of moderate to heavy sleet lasting ten to twelve hours—a feature that appeared in forecasts 3-4 days in advance. No other winter storm had produced such prolific and continuous sleet since the 2007 Valentine's Day winter storm.

Four-panel sequence showing 12-hour (top left), 36-hour (top right), 60-hour (bottom left), and 84-hour (bottom right) forecasts all valid January 25, 2026 12:00Z. The product displays simulated radar reflectivity (5-75 dBZ) color-coded by precipitation type using 1000-500mb thickness and surface temperature diagnostics. Green-yellow-red shades indicate liquid rain, with intensity increasing from light green (5 dBZ) through yellow to dark red (75 dBZ). Simulated radar reflectivity (dBZ) color-coded by precipitation type: green-yellow-red = rain (increasing intensity), blue-purple = snow, pink-magenta = ice pellets/mix, amber-orange = freezing rain. NWS winter weather watches and warnings overlaid. The time sequence demonstrates GEOS's skillful prediction of the storm's development and eastward progression out to 84 hours, accurately forecasting the expanding precipitation shield and complex phase boundaries that created significant winter weather impacts across multiple regions.

Despite the intrusion of warm air aloft, the Arctic air mass that had been in place over the region kept temperatures below freezing near and just above the surface, ensuring precipitation in the greater Washington, DC region would fall as sleet or freezing rain after the change from snowfall. The 2km downscaled simulation maintained realistic surface temperature gradients and captured the shallow cold air damming against the Appalachians that was essential for the prolonged freezing precipitation event.

The winter storm of January 24-25, 2026, will be remembered for the widespread area that it impacted; from ice accretions measured in inches over the Deep South, to snow totals measured in feet for big Northeast cities like Boston and Providence. The GEOS 2km downscaled simulation demonstrated exceptional skill in forecasting both the broad spatial extent and mesoscale details, with simulated observations across multiple satellite channels and radar reflectivity validating NASA's capability to provide satellite-realistic output from high-resolution global downscaling.

Extended Predictability: The Second Storm with GenCast AI

With a frigid airmass remaining in place over the eastern US, GMAO scientists turned to GenCast, an advanced AI-based ensemble forecasting system driven by GEOS-FP initial conditions, to assess the potential for additional winter storm development in the following days. GenCast's ensemble predictions consistently indicated the synoptic pattern would remain favorable for cyclogenesis off the Carolina coast, highlighting the extended predictability of this persistent cold regime.

GenCast_GEOS-FP 108-hour artificial intelligence ensemble forecast valid February 1, 2026 12Z, showing 500mb cyclonic relative vorticity (shaded, 10⁻⁵ s⁻¹) and geopotential height (black contours, 30 m intervals) for a developing coastal low pressure system off the Carolinas and Mid-Atlantic coast. Vorticity values range from 0-60 (×10⁻⁵ s⁻¹) with warmer colors indicating stronger cyclonic circulation in the upper-level trough pattern. Stippling indicates ensemble uncertainty, with denser patterns showing regions where the 32-member ensemble exhibits greater spread in geopotential height (50-200 m standard deviation).

Sure enough, a substantial low-pressure system formed off the Carolina coast the following weekend, exactly as GenCast ensemble forecasts had indicated with high confidence 5-7 days in advance. The AI-based system captured the rapid intensification of this mid-latitude cyclone, with ensemble members showing strong agreement on the development of classic bombogenesis characteristics—a rapidly deepening surface low, strong temperature gradients, and intense precipitation. GenCast's probabilistic forecasts provided valuable guidance on the likelihood of extreme precipitation rates and high wind speeds across coastal regions of the Carolinas westward to the Appalachians, with significant impacts for a region that does not encounter major winter storms annually.

The combination of GEOS-FP deterministic forecasts and GenCast ensemble predictions demonstrated the value of integrating physics-based and AI-based modeling approaches for assessing the predictability and uncertainty of back-to-back high-impact winter weather events.

Implications for NASA Earth Science

These back-to-back winter storms provided valuable case studies for assessing the predictability limits of high-impact winter weather events and demonstrated the utility of NASA's Earth system models and simulated observation capabilities for advancing weather and Earth system prediction science.