NASA · NASA-879 · record R0481

Data-Mining Similar Scenarios for Uncertainty Quantification of Solar Wind Predictions at L1

National Aeronautics and Space Administration · GSFC: Goddard Space Flight Center

Overview

Development stage: In development or acquisitionSource: a) Pre-deployment – The use case is in a development or acquisition status.
High-impact designation: Not high-impactSource: c) Not high-impact
Impact justification: Not reported
Start date: Not reported
Withheld from public reporting: NoSource: a) No

Mission

Topic area: Science
Operational functions: Forecasting and prediction Derived by keyword rules; see methodology

Problem the AI is intended to solve

Accurate Uncertainty Quantification (UQ) for space weather forecasts is an ever-important supplementary variable to enable accurate risk response. Modeling uncertainty is itself a “model of a model”, and one of the best datasets to describe a model’s performance is a past database of it’s predictions and after-the-fact observations. In this work, we develop a method based on k-NN and kernel regression to quantify uncertainty in the WSA solar wind model and it’s predictions of the solar wind speed at L1. By constructing state vectors that describe the current forecasting context— recent observations, recent predictions, and future predictions, we build a catalog of “similar scenarios” from past data. With a set of similar scenarios at each timestep, we can base our uncertainty on the performance in those cases. This approach—suitable for low-dimensional datasets such as time series—is extremely fast and interpretable. We find that the resulting uncertainty estimates naturally capture structured patterns in forecast error, such as shifts between solar minimum and maximum, and periodic features on the scale of half a solar rotation.

Expected benefits

Being integrated into WSA software for operational use in Moon 2 Mars and NOAA Space Weather Prediction Center

System outputs

Uncertainty Estimation (sigmas)

Technology

AI classification: Classical / predictive MLSource: Classical/Predictive Machine Learning: Models trained on data to make predictions or classifications based on identified patterns or relatio…
System name(s): Not reported
Custom-developed code: Not reported
Public source code: Not reported