SDO (Solar Dynamics Observatory)

NASA’s flagship solar imaging + magnetic observatory. 15 years of continuous full-sun observations across 10 EUV/UV channels at 12-second cadence, plus magnetograms — the largest single-mission dataset in NASA’s archive. The training data for the Surya foundation model.

What it sees

• AIA: 10 channels imaging the solar atmosphere at different temperatures — from the photosphere (1700 Å) to the corona (94 Å, ~6 MK). 12-second cadence. 4096×4096 pixel images per channel per observation.
• HMI: line-of-sight magnetic field, plasma velocity (dopplergrams), and continuum intensity of the photosphere. 45-second cadence for fast products; 720-second averages for high-precision.
• EVE: full-disk EUV spectral irradiance — captures the spectral profile of solar radiation reaching Earth.

Why it matters

• The single most cited dataset in heliophysics. Underpins ~2000+ research papers since launch.
• Space-weather forecasting — solar flare, coronal mass ejection (CME), and solar energetic particle event prediction depend on SDO observations.
• Surya (NASA-IMPACT + IBM, Aug 2025) — 366M-parameter foundation model trained on 218 TB of SDO imagery. Demonstrates that solar physics is now firmly in the foundation-model era.
• Practical importance: a major CME could cost trillions of USD in critical-infrastructure damage; SDO is the warning system.

Where to get the data

• JSOC (Joint Science Operations Center) at Stanford: jsoc.stanford.edu — the official archive. SDO data is NOT distributed via NASA’s standard 12 DAACs.
• VSO (Virtual Solar Observatory): virtualsolar.org — federated search across many heliophysics archives including JSOC.
• SunPy: the canonical Python toolkit — pip install sunpy — wraps VSO/JSOC search + read.
• Surya pretrained weights: huggingface.co/nasa-ibm-ai4science/Surya-1.0 — for downstream fine-tuning + inference.

What it enables

• Solar flare classification + early warning (10-60 min ahead of significant flare events)
• CME launch detection + initial trajectory estimation (when fused with SOHO LASCO coronagraph data)
• Coronal hole tracking → high-speed solar wind prediction
• Active region magnetic-flux evolution → flare productivity studies
• Long-term solar cycle variation (Cycle 24 + Cycle 25 fully captured by SDO)
• Foundation-model-driven downstream tasks (CME impact prediction, SEP forecasting, flare classification)

Gotchas

• Not on an EOSDIS DAAC. Heliophysics has its own data system (JSOC, VSO, SPDF for in-situ space-physics) separate from Earth-science DAACs. earthaccess won’t help — use SunPy or direct JSOC queries.
• Data volume. Full-resolution AIA archive is petabytes. Most workflows downsample either temporally (use 1-minute or 1-hour averages, not 12-sec native) or spatially.
• Surya is a research model. Pretrained for downstream tasks but not turn-key for “predict tomorrow’s CME.” Requires fine-tuning + careful evaluation.
• JSOC’s REST API has rate limits — for bulk downloads use SDO data center mirrors or pre-staged Hugging Face datasets.
• Pixel orientation conventions vary between products. WCS metadata is mostly trustworthy but verify before stacking.

Related missions

• Parker Solar Probe (NASA, 2018+): in-situ solar wind / coronal observations close to the sun.
• Solar Orbiter (ESA + NASA): out-of-ecliptic + close-approach solar imaging.
• GOES-R SUVI: NOAA’s geostationary solar EUV imager — operational space-weather workhorse, complementary to SDO’s research-grade AIA.
• STEREO-A: side-view of the sun + CMEs.
• Surya (NASA-IMPACT + IBM): the FM trained on SDO; downstream fine-tuning is an active research area at AJ’s institution (Sujit Roy is lead author).

Related datasets