Skip to content

About

SOLARIS is an AI-driven solar flare forecasting system that analyzes magnetic time-series data to provide early risk warnings, helping improve preparedness for disruptions to satellites, GPS, communication networks, and other space-dependent infrastructure.

Resources

Stars

1 star

Watchers

0 watching

Forks

Latest commit

 

History

8 Commits

Folders and files

Repository files navigation

Solaris AI

Solar intelligence. Earthly impact.

Understand the Sun. See what’s at stake. Solaris AI turns solar measurements into an understandable view of space weather: what is happening now, how likely a strong flare may be in the next 24 hours, and which connected services could be affected.

Built for students, researchers and curious explorers, it combines live monitoring, calibrated machine learning and measurement-based impact indicators in a local dashboard. It helps people investigate risk and uncertainty; it does not predict exact satellite failures or control critical infrastructure.

Watch the demo

Solaris AI — animated demo preview

⬇ Download the 10-second MP4 · 1080p · Actual application screens

The video is a recorded snapshot from 4 November 2026. Displayed measurements and forecasts were captured from the running application; they are not current readings when you replay the video. The animated Sun is an artistic illustration. This user-selected edit is approximately 10 seconds long. The original video was made locally with brag and Hyperframes. Media credits.

What you can explore

Page What it helps you understand
Overview Why solar activity matters to a world connected by satellites and electrical systems.
Solar activity Current X-ray class and flux, interactive history, and an experimental 24-hour M/X-class flare probability.
Impact Possible effects on HF radio, GPS/navigation, power grids, undersea communications and spacecraft environments.
Alerts Recent issued watches, warnings, alerts and summaries. An issued bulletin is not necessarily still active.
Guide Scientific terminology, NOAA R/G/S scales, threshold effects, data sources and Version 6 model details.

The interface includes light/dark modes, a collapsible sidebar, responsive cards, solar animations and reduced-motion support. Missing or stale measurements are labelled; no random values are substituted for unavailable data.

Run locally on Windows

Requirements

  • Python 3.10–3.13; the commands below use Python 3.13.
  • Node.js 20.19+ or 22.12+ and npm.
  • Internet access to NOAA for current conditions. CelesTrak is used by the optional orbital-data backend integration.
  • The trusted V6 model and supporting artifacts in backend/models/, included in this repository.

1. Get the project and install dependencies

Run in PowerShell:

git clone https://github.com/Nikhila-Dhevi-R/Solaris-AI.git
Set-Location Solaris-AI
py -3.13 -m venv .venv
.\.venv\Scripts\python.exe -m pip install -r backend\requirements.txt
npm ci --prefix frontend

No API key is required for the public feeds. The default frontend uses Vite’s /api proxy. Optional configuration is documented in .env.example; copy it to frontend/.env only if you need a custom API base URL.

2. Start the backend

.\.venv\Scripts\python.exe -m uvicorn app:app --app-dir backend --host 127.0.0.1 --port 8000

3. Start the frontend in a second terminal

From the same project directory:

npm run dev --prefix frontend

Open http://127.0.0.1:5173/. Interactive API documentation is at http://127.0.0.1:8000/docs. Keep both terminals running; press Ctrl+C to stop each service. This project runs locally and requires no hosting or deployment.

How it works

flowchart LR
    A[NOAA measurements] --> B[Fetch and cache]
    B --> C[UTC alignment and data-quality checks]
    C --> D[38 Version 6 features]
    D --> E[Saved preprocessing and Random Forest]
    E --> F[Probability calibration]
    F --> G[24-hour forecast and risk band]
    C --> H[Environmental threshold comparisons]
    G --> I[FastAPI → React dashboard]
    H --> I
Loading

Measured conditions come from NOAA SWPC GOES X-ray and proton observations and planetary Kp. Current observations and history are cached for two minutes; Kp, protons and alert products use five-minute caches. Orbital elements use a one-hour cache and are returned by the backend; an orbital-elements card is not displayed.

Forecasts use completed-hour bins and trailing windows. The live adapter requires fresh observations, a single identified satellite and at least 80% usable coverage in both X-ray channels over 1-, 24- and 72-hour windows. The saved preprocessing pipeline and probability calibrator are applied during inference. Inputs with excessive missing features or invalid coverage are rejected.

Impact indicators compare measurements against environmental scales. They describe possible effects, not observed service outages. HF indicators reflect X-ray-driven sunlit-side conditions; navigation and power-grid indicators use global Kp context. Local GPS error, transformer current, cable damage and satellite failure probabilities require additional information and are not calculated.

The frozen Version 6 model

Property Value
Artifact backend/models/solaris_ai_random_forest_phase2.joblib
Estimator Random Forest with saved preprocessing and probability calibrator
Inputs Exactly 38 features, validated against the supplied schema
Target M/X-class flare occurrence over a 24-hour horizon
Forecast alert cutoff 0.20594365951619512
Low → Moderate boundary 0.02756797277329327
Moderate → High boundary 0.36999108243721823
Runtime scikit-learn 1.6.1

The loader verifies the artifact’s SHA-256, feature order, estimator type and frozen thresholds. The supplied model is used without retraining. Load only trusted joblib artifacts: their serialization can execute Python code.

Historical evaluation — not a live accuracy promise

The supplied descriptive 2020–2024 test table reports 79.21% accuracy, 88.13% recall, 59.99% precision, and 0.901 ROC AUC at the frozen alert cutoff. These describe hourly forecast windows, not independently counted flare events. Results vary by year; test data had previously been observed.

All forecasts remain experimental informational shadow outputs. Operational NOAA feeds lack the science-quality flags used in training. Live feature equivalence, probability calibration under current conditions and operational reliability have not been established. A risk band is not a guarantee, and a flare probability alone cannot establish spacecraft impacts.

API

Method Endpoint Response
GET /api/health Service health
GET /api/conditions Current X-ray flux, class, timestamp and freshness
GET /api/history?hours=24 Flux history; accepts 1–168 hours
GET /api/forecast Experimental live-input forecast or a reason it is unavailable
POST /api/forecast/features Inference from exactly 38 caller-supplied named features
GET /api/satellites Service and spacecraft environmental indicators, measurements and optional orbital elements
GET /api/alerts Recent issued space-weather bulletins
GET /api/model Verified V6 metadata, cutoffs and historical evaluation

The feature-input endpoint accepts {"features": {"feature_name": 0.0}} as a shape illustration only: supply all 38 exact names from the schema. It cannot independently verify a caller’s data provenance or quality. All displayed timestamps use UTC.

Project structure

backend/
  app.py                 API, source retrieval and caching
  features.py            Science-quality feature transformation
  realtime.py            Operational-feed feature adapter
  inference.py           Input checks and calibrated inference
  model_service.py       Verified artifact loading and model details
  impacts.py             Measurement-based environmental assessments
  models/                V6 model, schema and evaluation artifacts
  tests/                 Backend regression tests
frontend/
  src/                   React pages, charts, guide and themes
  tests/                 Threshold, freshness and glossary tests
docs/demo/               Video demo, poster and credits
.env.example             Optional local configuration

Test and build

$env:PYTHONPATH='backend'
.\.venv\Scripts\python.exe -m pytest backend\tests -q
npm test --prefix frontend
npm run build --prefix frontend

Latest executed application checks: 20 backend tests passed, 6 frontend tests passed, and Vite build passed. The build reports a JavaScript chunk above 500 kB; backend tests emit a Starlette/httpx deprecation warning. These focused checks do not establish operational forecast accuracy. No GitHub CI workflow is currently configured.

Troubleshooting

  • Page does not open: keep the frontend terminal running and use the URL Vite prints. Check that port 5173 is available.
  • API unavailable: confirm the backend is running on port 8000 and open /api/health.
  • Forecast unavailable: inspect its reason in Guide or the API response. Feed failures, stale observations and insufficient coverage can correctly block inference.
  • Model verification fails: restore the trusted V6 artifact and install the pinned requirements. Do not substitute a different model to bypass the checks.
  • Feed temporarily unavailable: retry Refresh after a short interval. External data can be delayed, revised or missing.

Sources and scientific scope

Solaris AI is built for research, education and situational awareness. Use authoritative operational products and mission-specific procedures for safety-critical decisions. It issues no autonomous commands.

About

SOLARIS is an AI-driven solar flare forecasting system that analyzes magnetic time-series data to provide early risk warnings, helping improve preparedness for disruptions to satellites, GPS, communication networks, and other space-dependent infrastructure.

Resources

Stars

1 star

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages