Problem Statement Session 2 | Bharatiya Antariksh Hackathon 2026

Scientific concepts, discoveries, and nature/space phenomena presented

Space technology & image/wave/atmospheric sensing (Hackathon problem statements)

Adaptive optics & wavefront reconstruction (PS9)

• Wavefront sensing and phase retrieval using a Shack–Hartmann wavefront sensor.
• Turbulence characterization of Earth’s atmosphere, including deriving:
  • Coherence length
  • Coherence time
• Ground-based astronomy link: Atmospheric turbulence distorts incoming starlight wavefronts, producing blurry images. A deformable mirror is shaped to correct aberrations.
• Reconstruction methods mentioned/allowed:
  • Modal methods
  • Direct gradient control
  • Machine learning / deep learning approaches
• Workflow (as described):
  1. Use a time series of Shack–Hartmann sensor frames.
  2. Apply centroiding to find spot positions.
  3. Compute spot deviations from reference positions to estimate local slopes.
  4. Reconstruct the distorted wavefront.
  5. From the reconstructed wavefront, compute turbulence metrics (coherence properties).
  6. Convert wavefront correction into an actuator map (deformable mirror stroke lengths), including inter-actuator coupling.

Infrared thermal image colorization & enhancement (PS10)

• Super-resolution of thermal infrared (TIR) satellite imagery to recover structural detail.
• RGB colorization / translation to generate realistic-looking color images.
• Dataset context: Landsat 8/9 co-registered thermal band and RGB bands.
• Workflow (as described):
  1. Input: single-channel TIR at lower resolution (e.g., 200 m).
  2. Super-resolve to higher resolution (e.g., 100 m).
  3. Colorize the super-resolved output to produce RGB at matching resolution.
  4. Compare to ground truth RGB derived from Landsat imagery.
• Evaluation metrics mentioned: PSNR, SSIM, FID
• Additional checks:
  • Visual inspection for hallucination
  • Measuring inference time

Temporal super-resolution for satellite imagery via optical flow (PS12)

• Addresses a temporal resolution gap in geostationary weather imaging.
• Uses optical flow for:
  • Motion vector estimation
  • Inter-frame interpolation
• Motivation (space weather / rapidly evolving events):
  • Cyclones, wildfires, thunderstorms, floods
  • Evolution scales: minutes to tens of minutes

• Workflow (as described):

  1. Start from frames taken at known time intervals (e.g., 10 min or 30 min cadence).

  2. Train an optical-flow-based deep learning model to interpolate an intermediate frame (e.g., predict 12:10 using 12:00 and 12:20).

  3. Validate against the ground truth intermediate-time frame using metrics such as SSIM, MSE, PSNR (and FSIM mentioned).

  4. Apply the trained model to INSAT-3DR imagery to generate synthetic 15-minute frames. - Deliverables: - Interpolated frames - Trained model - Web dashboard - Animations - Validation report - Data sources mentioned: - GOES-19 (thermal infrared ABI channel 13) for training/validation - INSAT-3DR for final application - Other possible data sources mentioned: Himawari, Meteosat

Cross-modal satellite image retrieval using multi-sensor remote sensing data (PS11)

• Retrieval across modalities: find semantically similar regions between:
  • Optical
  • SAR
  • Multispectral imagery
• Core ML idea: learn a shared/unified embedding space so a query from one modality retrieves top-K matches in another.
• Evaluation targets/measurements mentioned:
  • Top-5 and Top-10 retrieval evaluation
  • F1 score (same-modality and cross-modality accuracy)
  • Efficiency: average retrieval time per query

Air-gap predictive co-pilot for secure MPLS operations (PS13)

(Network/security; not nature science but “AI for operations”.)

• Focus: offline/air-gapped autonomous predictive fault analytics for MPLS network operations.
• Uses:
  • Network simulation
  • Fault precursor features such as:
    • Route flapping
    • Queueing
    • Tunnel state
    • Tunnel flapping
  • Offline NLP with open-weight LLMs for RAG over local playbooks/runbooks and logs
• Output requirement: explain predicted issues in plain English, with no internet/API dependency.
• Network log technologies referenced: SNMP, syslog
• Model approach referenced: RAG and LLM (open-weight)

Forecasting energetic particle radiation environment for ISRO geostationary satellites (PS14)

• Space weather / energetic radiation at geostationary orbits.
• Physical phenomena described:
  • Solar storms: the Sun emits energetic charged particles that interact with Earth’s magnetosphere.
  • Earth’s magnetic field can be punctured during strong solar wind conditions, causing harsh radiation environments.
  • Particle populations include MeV electrons, with emphasis on “killer electrons” (MeV+), which penetrate deep and cause electronic malfunctions.
• Forecasting goal: short-term and long-term forecasts of electron fluxes at geosynchronous orbit for Indian longitude.
• Data mentioned:
  • GOES satellite electron fluxes (NOAA; >2 MeV) in CDF format
  • Solar wind conditions parameters from public sources (CDAWeb mentioned)
  • A validation dataset described as GRACE/India-related (public domain source mentioned)
• Modeling outline:
  • Data preprocessing (remove spikes, interpolate gaps)
  • Feature selection
  • Time-series networks such as LSTM and transformers
  • Multi-step forecasting:
    • Example nowcast: 30–45 min
    • Extra forecast: 6–12 h
• Deliverable:
  • Real-time GUI forecasting electron flux evolution for >2 MeV electrons.

Nowcasting/forecasting solar flares using soft & hard X-ray data from Aditya-L1 (PS15)

• Space phenomenon: Solar flares are energetic magnetic explosions releasing radiation and energetic particles.
• Space weather impacts on Earth:
  • Geomagnetic storms
  • Power grid damage
  • GPS disruption
  • Satellite communication effects
• Instrumentation (Aditya-L1, launched 2023):
  • SoLEXS (Solar Low Energy X-ray Spectrometer): soft X-rays (thermal plasma heating); gradual enhancements
  • HELIX/HELIOS (High Energy L1 Orbiting X-ray Spectrometer): hard X-rays (non-thermal particle acceleration); impulsive signatures
• Fine structures mentioned:
  • Pre-flare sudden increases
  • Quasi-periodic pulsations in hard X-rays
• Prediction targets:
  • Solar flare detection/nowcasting
  • Build a unified flare catalog combining SoLEXS and HELIOS
  • Forecast probability that a flare occurs in the next N minutes
  • Multi-class probabilities for flare intensity/energy class
• Evaluation criteria mentioned:
  • Detection accuracy across low/high-class flares
  • High true positive rate and low false positive rate
  • Forecast with 15–30 minutes lead time
• Data source mentioned: Aditya-L1 SoLEXS and HELIOS Level 1 data via the ISSDC portal

---

Researchers/sources featured (named in subtitles)

• ISRO (Indian Space Research Organisation) — hackathon organizer; referenced for problem-context missions/instruments
• Hack to Skill — organizer of the explainer session (mentioned by a mentor)
• NOAA — referenced for providing GOES electron flux data
• CDAWeb — referenced for solar wind conditions/parameters
• USGS — referenced for Landsat imagery (thermal/RGB data)
• Aditya-L1 — ISRO solar observatory mission
• Landsat 8 / Landsat 9 — referenced satellite data
• GOES-19 — referenced geostationary satellite
• Himawari and Meteosat — mentioned as possible data sources for PS12
• SNMP, syslog — referenced network log technologies for PS13
• RAG and LLM (open-weight LLMs) — model approach/source type referenced for PS13
• Vikram Sarabhai Space Centre (VSSC) — mentor affiliation named for PS14
• Pradhan — referenced (website/source) for geostationary validation dataset mentioned under PS14
• ACE (ACEwind) — referenced indirectly for real-time solar wind/space weather retrieval for PS14 (in Q&A)