APPLIED ARTIFICIAL INTELLIGENCE FOR CLIMATE ADAPTATION, ENERGY FUTURES, AND SPACE EXPLORATION, FOR ALL HUMANKIND.

This work is supported through Grant #80NSSC24MO122 from SMD/NASA Heliophysics Division

FDL-X Heliolab 2024

In 2024, FDL-X Heliolab will include four exciting challenges in two different formats; Research Challenges and Research and Technology (R&T) Challenges

The two Research Challenges are low-TRL open research problems which will explore new scientific questions. 

The two R&T Challenges have a higher TRL outcome and place more emphasis on modularity, reproducibility, benchmarking and continuous learning - a key concept in ensuring reliability and safety. As in FDL-X in 2023, there still remains a requirement for useful integration between these AI pipelines.

FDL-X Heliolab 2024 AI SOTA Showcase

  • The use of AI to predict thermospheric density in response to solar activity is crucial for space traffic management, space sustainability, end-of-spacecraft life, and reentry calculations. Karman, an advanced AI tool developed by research teams in 2021 and 2023, has demonstrated a data-driven approach to thermospheric density estimation. This tool enables effective model comparison and prepares and ingests ML-ready space weather, solar, and thermospheric data. Prior teams have leveraged SDO images, NOAA space weather data, and EUV irradiance to reduce prediction errors in thermospheric density. Despite Karman's superior performance over current state-of-the-art estimates, this challenge will focus on building a continuous learning system to update model parameters and inputs as new data becomes available. Additionally, we will tackle the challenge of forecasting thermospheric density at future time windows.

    1. Can we enable an operational data-driven live streaming service for thermospheric density prediction?

    2. Can we continuously update the machine learning model as new data becomes available? 

    3. What are the additional solar and space weather inputs necessary to better understand and model the Sun-Earth interactions?

    4. Can we advance ‘state-of-the-art’ in the provision of ML tools for the community?

      Github - publicly available after results publication - January 2025

  • Geoeffective-CL transitions the twinned AI pipelines of DAGGER++ and SHEATH to operations. These pipelines track the geoeffectiveness of solar disturbances, modeling a chain of coupled systems that include observations of the solar disk, propagation via the solar wind, and interaction with the geospace environment. DAGGER++ remains the first and only ML pipeline able to predict localized geomagnetic field perturbations globally, while SHEATH-DAGGER provides actionable forecasts with uncertainty estimates and advanced lead-time. 

    The goal of this challenge includes developing additional capability and training on more data volumes and sources. Moving towards a continuous learning paradigm, online benchmarking and inference will allow for real-time assessment and seamless integration of future models/datastreams.

    1. What if we could reliably extend and evolve the forecast horizons for geospace environment models (namely DAGGER and SHEATH) by seamlessly incorporating data updates and new knowledge over time?

    2. Can we build a software pipeline to enable continual updating, and utilize this framework to build models that can effectively run a continual learning strategy? 

    3. Can we determine what solar and space weather inputs are necessary to better understand and model Sun-Earth interactions, and how these factors drive uncertainty in the modeling pipeline?

    4. Can we advance the provision of ML tools for the community and enhance their impact by defining the ‘state-of-the-art’ required for operational space environment modeling and future science, i.e., ensuring the ability to assimilate data sources and pipeline components, make online updates, and continuously validate capabilities?



    Github publicly available after results publication - January 2025

  • Building on previous work (FDL 2022 4PI Irradiance, FDL 2018, and FDL-X 2023 SDOML), we will create a virtual observatory probing the solar corona within the spherical region of the solar atmosphere stretching out to Mars, including the coronal plasma density and temperature, and the EUV spectral irradiance.  We will address gaps in SuNeRF models by enabling a temperature and density estimation pipeline that simultaneously uses multi-wavelength images from any available EUV imager for which the temperature response function of the instrument is well understood.  The goal is to produce near real-time estimates and a continuously-learning model.


    1. Can we predict the full EUV spectral irradiance at Mars?

    2. Can we extend the SuNeRF method to estimate the physical parameters in the solar atmosphere (i.e., temperature, density)?

    3. Can we combine the next generation of solar observing satellites to advance reconstructions of the solar atmosphere?

    4. Can we expand the reconstructed spectral irradiance to other parts of the electromagnetic spectrum?

    Github publicly available after results publication - January 2025

  • The goal of this project is to develop a virtual dosimeter to predict radiation exposure for human space flight beyond low Earth orbit (BLEO). The virtual dosimeter will provide early warnings (10-30 minutes) of solar energetic particle (SEP) events, allowing spacecraft crew to take action to minimize radiation exposure. It will integrate solar observation and radiation data to train a spatial time series model to make predictions of future radiation. The initiative leverages past projects to produce a radiation forecasting system, supporting future Artemis missions and beyond.


    1. What are the most promising machine learning approaches (e.g., see this recent survey on foundation models for time series) and dataset (e.g., RadLab) to predict the space radiation environment in locations beyond low Earth orbit (BLEO)?

    2. Can we advance the state-of-the-art in modeling the space radiation environment in order to forecast solar energetic particle (SEP) events and their biological impacts to human crew?

    3. How can the above approaches be used to provide early warnings of radiation exposure for human space missions 10 to 30 minutes ahead of major events?

    Github publicly available after results publication - January 2025