CLEAR Space Weather Center of Excellence

• CLEAR is a NASA-funded initiative that integrates empirical algorithms, machine learning, and physics-based modeling to provide robust solar energetic particle (SEP) forecasts.
• It delivers end-to-end SEP predictions including probabilistic pre-eruption forecasts, post-eruption nowcasts, and all-clear alerts using a modular plug-and-play workflow.
• The validated SOFIE model demonstrates high fidelity and real-time performance, ensuring timely space weather predictions to safeguard both human and robotic missions.

The CLEAR (Coordinated Leading‐edge Advancements for Reliable SEP) Space Weather Center of Excellence is a five-year NASA-funded initiative within the Space Weather Center of Excellence (SWxC) program. Its primary mandate is to deliver an operational, end-to-end system for solar energetic particle (SEP) forecasting. This system is designed to provide robust probabilistic predictions of SEP events, post-eruption nowcasts (including onset, peak, and decay), and, critically, reliable “all-clear” forecasts specifying intervals—up to 24 hours—when SEP fluxes remain below mission-specific thresholds. CLEAR’s integrated approach combines empirical algorithms, ML, and physics-based first-principles modeling, directly addressing the highest-priority knowledge “gap” identified in NASA’s 2021 Space Weather Science & Observation Gap Analysis: forecasting SEP occurrence and properties at arbitrary heliospheric locations (Zhao, 2023, Liu et al., 12 Nov 2025).

1. Programmatic and Organizational Context

CLEAR is a five-year project (NASA Award 80NSSC23M0191) motivated by the imperative to protect human and robotic assets, both near-Earth and deep-space, from SEP-induced radiation hazards. The center leverages infrastructure from the Community Coordinated Modeling Center (CCMC), the NOAA-operational Geospace Modeling Framework (SWMF), and collaborations with operational and research actors such as NOAA/SWPC, NASA SRAG, CCMC, and the Moon-to-Mars Space Weather Analysis Office (SWAO). The team comprises heliophysics observers, physical modelers, statisticians, and computer scientists, providing a multidisciplinary foundation for advancing the SEP prediction “tall pole.”

CLEAR’s primary mission objectives are:

• Probabilistic pre-eruption forecasts for the likelihood and characteristics of SEP events.
• Post-eruption nowcasts/forecasts of the dynamically evolving SEP environment (onset, rise time, peak flux, decay, and full energy spectra) throughout the inner heliosphere (0.1 AU–1 AU, conceptually including out to Jupiter’s orbit).
• “All-clear” forecasts: alerts of intervals—up to 24 hours—where the time- and energy-dependent proton flux $j(E, t)$ across $E \geq E_0$ remains below user-prescribed thresholds $j_\text{threshold}(E)$.

The “all-clear” operational criterion is:

$$\max_{E > E_0,\, t \in [t_0, t_0 + \Delta t]} j(E, t) < j_\text{threshold}(E)$$

2. Scientific and Technical Foundations

The CLEAR forecasting platform employs a modular “plug-and-play” workflow integrating three principal modeling strata:

A. Empirical Algorithms

• SEPSTER and SEPSTER2D: Analytical relations mapping observed CME properties—velocity ($V_\text{CME}$), direction, and magnetic connectivity—to SEP peak flux via formulas such as

$$\log_{10} j_\text{peak}(E) = a(E) + b(E) \log_{10} V_\text{CME} + c(E)\cos\Delta\phi$$

where $\Delta\phi$ is the angular separation between CME nose and the observer’s magnetic footpoint.

• REleASE: Utilizes in-situ MeV electron measurements at L1 as early indicators for impending $>10$ MeV proton enhancements.

B. Machine Learning Modules

• Solar Eruptive Event Prediction: Convolutional and dense neural networks process HMI/MDI/GONG magnetogram image patches or SHARP/SMARP vector features to predict CME occurrence and kinematics.
• SEP Occurrence & Peak Flux Regression: ML models integrate photospheric magnetic field parameters and ML-inferred CME outputs to yield probability of SEP occurrence ($P_\text{event}$) and peak flux $\hat{y} = f_\text{ML}(x)$. Classification and regression are optimized via binary cross-entropy

$$\mathcal{L}_\text{class} = -\sum_n [y_n \log p_n + (1-y_n) \log(1-p_n)]$$

and mean squared log error

$$\mathcal{L}_\text{reg} = \sum_n ( \log j_\text{obs, n} - \log j_\text{pred, n} )^2$$

The ML workflow incorporates oversampling, cost-sensitive losses, and stability analyses to mitigate rare-event biases.

C. Physics-Based Modules

• Steady and Time-Dependent Solar Wind: AWSoM/AWSoM-R solves extended MHD equations (with anisotropy and Alfvén-wave closure), driven by ADAPT-GONG magnetograms, in a computationally efficient manner ($\approx$512 cores for real-time operation).
• CME Initiation: EEGGL and TiDeS-G inject parameterized magnetic flux ropes (Gibson–Low, Titov–Démoulin) into the modeled corona.
• Particle Acceleration & Transport: Includes M-FLAMPA (field-line Parker diffusion), AMPS (3D Monte Carlo), and iPATH/SEPCaster (SDE approach). The Parker transport equation is central:

$$\frac{\partial f}{\partial t} + (u + v\mu \mathbf{b}) \cdot \nabla f + (1 - \mu^2) \frac{v}{2L} \frac{\partial f}{\partial \mu} = \frac{\partial}{\partial \mu}\left[(1-\mu^2) D_{\mu\mu} \frac{\partial f}{\partial \mu}\right] + \frac{\partial}{\partial s}(\kappa_\parallel \frac{\partial f}{\partial s}) + Q$$

• Seed Population Specification: Incorporates thermal tails from AWSoM temperatures, retrodicted suprathermal spectra, and HYPERS hybrid simulations (shock-injection as a function of $\mathcal{M}_A$ and $\theta_{BN}$).

3. The SOFIE Model and Real-Time Validation

CLEAR’s prototype physics-based operational model, SOFIE (SOlar wind with FIeld lines and Energetic particles), was demonstrated during the May 2025 Space Weather Prediction Testbed (SWPT) at NOAA/SWPC (Liu et al., 12 Nov 2025). SOFIE is architected within SWMF and comprises:

• AWSoM-R for stream-aligned MHD solar wind
• EEGGL for CME flux-rope initiation
• M-FLAMPA for focused transport and acceleration of SEPs along field lines

The model employs the focused transport equation for phase-space density $f(\mathbf{r}, p, \mu, t)$, with pitch-angle diffusion coefficient parameterized by upstream mean free path $\lambda_\parallel$ (default 0.1 au) and seed injection at the evolving shock front:

$$D_{\mu\mu}(p,\mu) = \frac{v}{2\lambda_\parallel}(1-\mu^2)$$

$$Q_\text{acc}(s,p,t) = S_0\;\delta[s - s_s(t)]\,p^{-q}$$

where $S_0$ was empirically tuned per event (e.g., $S_0=1.0$ for 2017, $S_0=10.0$ for 2001). The hybrid grid strategy enables a balance of simulation accuracy and operational feasibility via block-adaptive meshing, focusing high resolution along critical structures (Heliospheric Current Sheet, Earth-directed cone, shock fronts).

Table: SOFIE Operational Performance (4 Nov 2001 Event, 1,000 Cores)

Setup	SC grid resolution	4-day run time (h)	Catch-up to real time (h)	Spearman $\rho$ ($>10$ MeV)	% within 1 dex	Median log err
1	default (0.7°/1.4°)	21.1	10.9	0.93	99.9%	–0.18
2	coarse SC ×2	4.9	1.2	0.84	92.7%	+0.05
3	hybrid (coarse + refined HCS/cone)	18.6	4.1	0.87	84.7%	–0.04

Coarsening the SC grid enabled SOFIE to complete a 4-day simulation in 4.9 hours wall time with moderate correlation ($\rho=0.84$) and high order-of-magnitude accuracy. Higher fidelity comes at increased computational cost; a two-stage forecast (Setup 2 for rapid warnings, Setup 1 for detailed analysis) was adopted based on forecaster feedback.

4. Data Assets, Computational Infrastructure, and Workflow

The CLEAR operational workflow is anchored by:

• Model Coupling: SWMF (Fortran/C++) provides the framework for integrating modules, operational at NOAA/SWPC since 2016.
• Solar Data Inputs: ADAPT-GONG and SDO/HMI magnetograms, SOHO/LASCO and STEREO/SECCHI CME imagery, GOES flare and proton channel data, ACE/SOHO in-situ electrons, and PAMELA proton spectra.
• ML Libraries: TensorFlow/PyTorch (neural net architectures), scikit-learn (feature vector models).
• Particle and CME Simulations: M-FLAMPA, AMPS, iPATH/SEPCaster for transport; EEGGL, TiDeS-G for flux-rope CMEs; HYPERS hybrid simulations for shock-particle injection efficiency.

Routine operational use involves daily AWSoM-R solar wind runs based on near-real-time magnetograms. Upon CME detection, corresponding EEGGL and M-FLAMPA modules are executed, with forecast products delivered through SWPC analysis tools and real-time data streams.

5. Model Validation and Performance Metrics

Validation relies on a new benchmark dataset assembled from multi-spacecraft SEP event lists, cross-calibrated GOES/IMP8/STEREO/PAMELA fluxes, and well-documented flare, CME, and radio source associations (1973–present). Modules and the integrated system are evaluated via:

• Timing and Flux Accuracy: Mean error (ME) of event onset and peak, mean log error (MLE), mean absolute log error, RMSLE.
• Probabilistic Skill: Probability of detection (POD), false alarm rate (FAR), True Skill Score (TSS = POD – FAR), Heidke Skill Score, Brier Skill Score (BSS).
• All-Clear Reliability: Reliability diagrams, ROC curves, with forecast lead-times of 6 and 24 hours.
• Uncertainty Quantification: Bootstrap ensembles, Bayesian dropout, rare-event weighting, and stability checks.

By the end of year 3, individual modules are targeted to reach TRL 4. TRL 5 is planned for year 5, with system hand-off to operational centers and provision of publicly accessible SEP forecasting services.

6. Timeline, Deployment, and Operational Integration

CLEAR’s staged deployment involves:

1. Years 1–2: Assembly of the benchmark dataset, release of SEPSTER2D and REleASE upgraded with stream-aligned MHD connectivity, operational AWSoM-R solar wind.
2. Years 2–3: Integration of ML-based CME kinematics and SEP occurrence modules; coupling of M-FLAMPA/AMPS with CME drivers.
3. Years 3–4: End-to-end 0–24 hr forecasts demonstrated in shadow mode at NOAA SWPC/CCMC; refinement of uncertainty quantification protocols.
4. Years 4–5: Full operationalization of iPATH/SEPCaster, completion of user interfaces and data delivery pipelines, community workshops.
5. End of Year 5: Handover of operational SEP forecasting capabilities to NOAA SWPC and NASA CCMC at TRL/RL 5.

SOFIE, within this framework, achieved a milestone during the May 2025 SWPT exercise ($\sim$5 hours wall-time for 4-day simulation on 1,000 cores; factor-of-2 agreement with GOES), overturning assumptions regarding the impracticality of physics-based real-time SEP prediction models for operations.

7. Broader Implications and Future Prospects

CLEAR exemplifies a new generation of space weather prediction systems—characterized by modularity, rigorous empirical and physics foundations, and operational agility. The demonstrated success of SOFIE in real-time SEP forecasting (both in accuracy and performance) suggests the feasibility of routine physics-based SEP hazard alerts for human spaceflight, including Artemis-class missions and beyond.

Further development is focused on full automation (from flare detection through forecast product generation), workflow integration with operational centers, and continuous benchmarking against historical and contemporary events. The center’s commitment to scientific transparency, modular extensibility, and rigorous validation aims to advance both foundational understanding and practical forecasting of heliospheric particle environments.