Built on InnoVisions™ · NOAA SBIR Phase I and II · U.S. patent pending
LEARN² (LEARN2, the Landfalling Event Atmospheric River Neural Network) is INNOVIM’s hybrid AI and numerical weather prediction tool for atmospheric river and extreme precipitation forecasting. It reads the world’s leading ensemble forecasts and predicts whether rainfall will exceed the thresholds that matter to decision makers, up to 10 days ahead.
Fast path to award
Because LEARN² was developed with NOAA Small Business Innovation Research (SBIR) funding, any federal agency can award INNOVIM a sole-source SBIR Phase III contract for work that derives from, extends, or completes it. Phase III uses the agency’s own program funds, has no dollar limit, and the SBIR program is authorized through September 30, 2031. NOAA lists LEARN² in its Technology Marketplace as eligible for Phase III awards.
| Phase I | NOAA award NA20OAR0210336 (2020) |
|---|---|
| Phase II | NOAA award NA21OAR0210296 (July 2021 to December 2023) |
| NOAA SBIR topic | 9.5, Next Generation Observation and Modeling Systems |
| Patent | U.S. patent pending, application 17/572,534 |
| Awardee | INNOVIM, LLC · UEI PFKNCNAF6MD3 · CAGE 3TTR0 · Woman-Owned Small Business |
Why it matters
Atmospheric rivers carry tropical moisture to the West Coast and drive most of its extreme rain. Between 1978 and 2017, flooding cost states west of Kansas $50.8 billion in insured losses, and $42.6 billion of that came from atmospheric-river floods. The hardest part is the 1 to 10 day window, when reservoir releases, flood warnings and evacuations must be decided.
Loss figures: Corringham et al. (2019), cited in the LEARN² paper.
How it works
NOAA GEFS and ECMWF ensemble forecasts, plus large-scale climate indices (PNA, AO, NAO, El Niño) that add longer-range skill.
Dense and convolutional networks for each forecast day. A custom genetic algorithm selects each network’s structure and inputs.
A novel voting algorithm combines the AI networks with the physics-based forecasts, so each strengthens the other.
Maps of the likelihood of exceeding fixed rainfall amounts from 0.5″ to more than 4″, or local percentiles, for each of the next 10 days.
See it
Adapt it to your mission
LEARN² was built for atmospheric rivers on the U.S. West Coast, but its method is general: learn from years of ensemble forecasts, then predict whether a quantity that matters will cross a threshold that matters. INNOVIM can retrain and extend it for new regions, hazards, and decisions.
Reservoir operations, flood control releases, and early flood risk communication.
Heavy rain from tropical systems, flash flooding, and other extreme precipitation regimes.
Weather impacts on operations, installations, training, and logistics.
Energy, transportation, insurance, and agriculture decisions that hinge on extreme rainfall.
Each extension is scoped with the customer. Contact us to discuss whether your requirement fits SBIR Phase III.
Platform and research
LEARN² runs on InnoVisions™, INNOVIM’s commercial data analytics and workflow platform, which handles data ingestion, processing, and the decision-support interface. The same platform supports other mission applications, with a working prototype on customer data in 3 to 6 months.
Landfalling Event Atmospheric River Neural Network (LEARN²): an AI Machine Learning Numerical Weather Prediction Tool for Extreme Precipitation Forecasting. Ardanuy, P. E., S. Isaacman, D. Mister, E. Carton, B. R. Colman, and J. Hicks.
Presented at the 103rd American Meteorological Society Annual Meeting, 22nd Conference on Artificial Intelligence for Environmental Science, Denver, January 10, 2023. Also presented at the AGU Fall Meeting, Chicago, December 14, 2022.
Common questions
LEARN2 (LEARN²) is an AI and machine learning tool for extreme precipitation forecasting. It combines neural networks with the NOAA Global Ensemble Forecast System (GEFS) and the ECMWF ensemble to predict, up to 10 days ahead, whether rainfall from atmospheric rivers and other storms will exceed thresholds that matter for flood risk and water management.
LEARN2 was developed under NOAA SBIR Phase I (NA20OAR0210336) and Phase II (NA21OAR0210296). Any federal agency can award INNOVIM a sole-source SBIR Phase III contract for work that derives from, extends, or completes that research, without a new competition.
Yes. The method learns from years of ensemble forecasts to predict threshold exceedance, so INNOVIM can retrain and extend it for other regions, other extreme weather hazards, other forecast models, and other decisions, such as reservoir operations or defense weather support.
Validated on fully independent data using the F1 score, LEARN2 outperformed both the NOAA GEFS and the ECMWF ensemble, adding 1 to 2 days of skillful lead time for extreme rainfall. National Weather Service Western Region forecasters beta tested it during the 2022 to 2023 rainy season.
Contracts and Phase III: Marianna Murphy, President, mmurphy@innovim.com, 781-640-5526.
Technical: Dr. Shahin Samadi, Principal Investigator, ssamadi@innovim.com.