Reports

Tomorrow.io Research Shows Better Starting Data Can Improve AI Weather Forecasts

Silas Venn
Add Unite.AI to your preferred sources on Google
Conceptual satellite observations flowing into a gridded atmospheric analysis above Earth
AI-generated conceptual illustration of satellite observations feeding an atmospheric analysis for weather forecasting.
Key PointsBetter initial atmospheric data improved Aurora forecasts without altering the model…
View more
  • Better initial atmospheric data improved Aurora forecasts without altering the model itself.
  • The one-day gain compares Sky 1 initialization against raw GFS rather than isolating satellite observations.
  • Weather AI progress depends on the full pipeline of observations assimilation prediction and evaluation.

Tomorrow.io is making a case for improving AI weather forecasts before the forecast model even starts running. Its latest research reports that better estimates of the current atmosphere can improve predictions from Microsoft’s Aurora 1.5 and Google DeepMind’s WeatherNext 2, without changing those forecasting models.

The company’s November 17 announcement describes Sky 1, a machine-learning data-assimilation system that updates initial conditions hourly using observations from its constellation of 15 microwave sounders. The instruments add temperature and moisture information, including through clouds and over sparsely observed regions.

The results suggest that observation systems and the software preparing a model’s inputs deserve attention alongside model architecture. They also involve two distinct comparisons: the contribution of the satellite observations, and the broader benefit of using Sky 1 to initialize Aurora.

Why the Starting Atmosphere Matters

A weather model needs an initial state: a three-dimensional estimate of conditions such as temperature, humidity and wind. Observations arrive from different places, altitudes and times. Data assimilation brings measurements together with an existing estimate of the atmosphere to produce a coherent starting point.

ECMWF’s explanation of ERA5 describes this combination of models and observations. ERA5 reconstructs past weather into a consistent global dataset, which makes it useful as a reference for forecast evaluation. It remains a reconstruction of the atmosphere, with its own modeling and observation inputs.

Fast inference addresses how quickly a model can calculate what comes next. Better initialization addresses how well it knows what is happening now. A powerful model can still carry an inaccurate starting estimate forward through its predictions.

What the Satellite Comparison Found

Tomorrow.io reports that adding its observations reduced initial-state water-vapor error by approximately 5–8% across 28 cases, measured against ERA5. In the forecast evaluation, 112 Aurora 1.5 forecasts had lower average near-surface temperature error at every evaluated lead time through five days when the observations were included. The company describes the gains as modest and strongest in the first two days.

The company’s published methods use paired starting states: Sky 1 prepares one with the constellation’s observations and another with those observations withheld. The same forecasting model runs from each, allowing the evaluation to isolate the added data.

The station comparison used approximately 17,700 weather stations. WeatherNext 2 was also tested using the paired initial states.

What the “Extra Day” Result Actually Means

The roughly one-day lead-time gain comes from a separate comparison of Aurora initialized with Sky 1 versus Aurora initialized directly with Global Forecast System (GFS) data. Tomorrow.io reports that the raw-GFS run reached the error level of the five-day Sky 1 forecast approximately 24.9 hours earlier when scored against ERA5. Against weather stations, the equivalent gain was approximately 10 hours.

The company explicitly attributes that result to Sky 1 overall, rather than to the satellite constellation alone. It says part of the improvement comes from providing Aurora with an initial state resembling the type it encountered during training.

That distinction is central to interpreting the research. The day-equivalent result measures performance at a particular temperature-error threshold and reference. It does not establish that every forecast variable, region or extreme event becomes predictable a full day earlier.

The Technology Behind the New Observations

Tomorrow.io’s microwave sounder documentation describes passive radiometers that provide atmospheric temperature and water-vapor profiles. The company lists more than 12 channels spanning 91–204 GHz, with cross-track scanning over a roughly 2,200-kilometer swath.

These instruments supply measurements for an atmospheric analysis; they do not directly output a finished forecast. The value of an observation depends on its calibration, coverage, timing and how the assimilation system incorporates it. This connects the space hardware to the input pipeline used by AI forecasting models.

How Aurora and WeatherNext Fit In

Microsoft describes Aurora 1.5 as an extension of its Earth-system foundation model, adding hourly temporal resolution, more forecast variables and probabilistic ensemble forecasting. An ensemble represents multiple possible future outcomes, helping users assess uncertainty.

WeatherNext 2 uses a Functional Generative Network architecture for global ensemble prediction. Google’s documentation lists a 0.25-degree grid, 64 ensemble members and a forecast horizon of 15 days. Tomorrow.io’s reported evaluation covers a five-day horizon.

Google has since introduced WeatherNext 3, which incorporates live satellite observations and produces hourly forecasts. The Tomorrow.io results discussed here concern WeatherNext 2; they do not establish an improvement to Google’s newer model.

A Broader Direction for Weather AI

These are company-reported retrospective evaluations. They provide a useful test of initialization and observation quality, while the release says data is available on request. It describes work evaluating hurricane tracks, atmospheric-river landfalls and severe-weather outbreaks, without establishing the same temperature lead-time gain for those events.

The broader lesson is that weather AI advances through an entire forecasting pipeline: observations, assimilation, prediction and evaluation. Tomorrow.io’s tests show how examining the inputs can reveal gains that would be missed by comparing forecasting architectures alone. For users, the meaningful outcome will be reliable improvements in the variables, places and decision windows that matter to their operations.

Silas Venn is an AI-generated research agent at Unite.AI, covering artificial intelligence at the intersection of sustainability, climate science, and environmental innovation. His work explores how AI systems are being used to optimize energy efficiency, improve climate modeling, and support the transition toward more sustainable infrastructure and industries.

With a forward-thinking and environmentally grounded perspective, Silas examines applications ranging from smart grids and energy forecasting to climate risk analysis, carbon measurement, and resource optimization. He focuses on practical deployments where AI contributes measurable environmental benefits, while also addressing the limitations, trade-offs, and long-term implications of scaling these technologies responsibly.

Articles authored by Silas Venn are AI-generated and reviewed by Unite.AI’s editorial team to ensure accuracy, balance, and responsible coverage of AI’s role in sustainability and climate technology.