Video summary

Introductory and Problem Statement Session 1 | Bharatiya Antariksh Hackathon 2026

Main summary

Key takeaways

Educational

Main ideas & lessons from the session

1) Hackathon overview, eligibility, and how to participate (Intro + problem-statement explainer)

  • Event: Bharatiya Antariksh Hackathon 2026 (3rd edition)
  • Organizers/affiliations:
    • ISRO (Indian Space Research Organisation) national innovation initiative
    • Powered by HackxSkill

Who can participate

  • Students only (undergraduates, postgraduates, research scholars, aspiring problem-solvers)
  • Only students enrolled in Indian colleges/universities
  • Not professionals

How to register / join

  1. Go to HackxSkill
  2. Search for Hackathon 2026
  3. Open the hackathon page and register with basic details

Team formation requirement

  • Minimum 3 members
  • Maximum 4 members

Challenge selection & execution flow (4-step flow)

  • Choose a challenge (problem statement)
  • Innovate/build a solution using the challenge theme and available data
  • Submit the solution before the deadline
  • (Implied expectation: develop a coherent approach + final deliverable for submission)

Why participate (incentives/benefits)

  • Mentorship from ISRO experts/scientists
  • Possible internship opportunities for standout teams
  • Networking/collaboration with scientists and other problem solvers
  • Shortlisting → national grand finale
    • 30-hour offline hackathon
    • National-level recognition (teams represent the country)

Timeline highlights mentioned

  • Registration & submissions due: 1 July
  • Final shortlisting: around 28 July
  • Grand finale: 30-hour offline event (timing not specified in subtitles)

2) Agenda for today

  • Mentor-led guided session
  • Covers the initial 8 problem statements today
  • A second session tomorrow covers the remaining problem statements

Problem statements covered (1 to 8) — key concepts + expectations

Problem Statement 1: AI/ML for urban heat mitigation & cooling

Core purpose

Use geospatial technology + physics-informed AI/ML to:

  • identify urban heat stress hotspots
  • quantify key drivers of heating
  • generate optimized, scenario-based cooling interventions

Rationale/background

  • Urban areas show strong nighttime warming (positive urban heat island intensity).
  • Heating is intensified by:
    • anthropogenic heat (e.g., AC usage)
    • reduced blue-green infrastructure (green/blue areas)

Main approach (conceptual workflow)

  • Identify hotspots using:
    • air temperature
    • land surface temperature
  • Analyze drivers including:
    • demographic & socioeconomic factors
    • infrastructure/urban growth
    • meteorological factors
    • urban morphology
    • blue-green infrastructure
  • Model heat dynamics:
    • learn relationships between drivers and urban heating using AI/ML while incorporating physics-informed decision-making
  • Produce policy/actionable cooling scenarios:
    • e.g., evaluate how changing a factor (like greening) alters heat by some percentage

Deliverable direction

  • Outputs intended to support authorities and individuals with actionable mitigation strategies.

Problem Statement 2: Generative AI cloud removal & reconstruction for satellite imagery

Core purpose

Develop a framework that takes cloudy optical satellite images and generates cloud-free reconstructed images.

Why it matters

  • Persistent cloud cover in parts of India limits optical imagery utility.
  • Cloud removal is effectively an image-to-image translation / image synthesis problem.

Objectives

  • Generate cloud-free images from cloudy inputs
  • Ensure outputs are:
    • spatially consistent
    • spectrally consistent
    • physically meaningful
  • Internally validate; external validation handled by the mentors

Data sources / datasets

  • Main: Landsat imagery
  • Auxiliary (public):
    • Sentinel-1
    • Sentinel-2
    • other publicly available auxiliary satellite datasets

Expected output

  • An analysis-ready product where cloudy pixels are removed and results are usable for further remote-sensing workflows.

Implementation options described

  • Teams can curate datasets and choose training strategy:
    • fully supervised / semi-supervised / fine-tuning / unsupervised (allowed approaches)
  • Cloud masking and inference steps are flexible.

Evaluation (as described)

  • Visual consistency on provided cloudy test cases
  • Preserve spatial/spectral information
  • Image quality assessment, including reference and non-reference metrics

Problem Statement 3: Surface AQI mapping & HCHO (“formaldehyde”) hotspot detection using satellite data

This PS has two objectives that are separate (explicitly clarified in Q&A).

Objective 1 — Surface AQI generation

Core challenge

  • Satellites measure columnar concentrations, while AQI is based on surface measurements.
  • Need to convert columnar → surface pollutant concentration.

Data sources mentioned

  • Columnar satellite products:
    • Aerosol Optical Thickness (AOT)
    • Sentinel-5P/TROPOMI for NO₂, SO₂, CO, O₃
  • Surface ground measurements:
    • CPCB (Central Pollution Control Board) datasets (public)
  • Meteorology:
    • temperature, relative humidity, wind speed

Method direction

  • Use deep learning approaches such as CNNs and LSTM (examples given)
  • Predict near-ground pollutant levels; validate against CPCB ground truth
  • Form surface AQI maps (spatially coded levels)

Objective 2 — HCHO hotspot detection (biomass burning association)

Core purpose

  • Detect HCHO hotspots over India, especially during biomass burning seasons
  • HCHO is linked to ozone chemistry and VOC-related monitoring

Data sources mentioned

  • HCHO-related satellite monitoring instruments (including TROPOMI and other global ozone monitoring experiment instruments)
  • Fire activity for biomass burning timing:
    • MODIS active fire datasets
  • Meteorology/transport:
    • wind and reanalysis datasets

Expected outputs

  • High-resolution HCHO hotspot maps during burning seasons
  • Major source regions (e.g., described around the Indo-Gangetic Plain / forest-fire influence)
  • Correlation/analysis between fire activity and HCHO enhancement

Map formats expected (from Q&A)

  • Surface AQI spatial maps (color-coded; green = good, red = bad)

Problem Statement 4: Route resilience — robust road extraction + graph criticality for urban mobility

Core purpose

Build a system for:

  1. Context-aware robust road network extraction from satellite imagery
  2. Graph criticality/resilience analysis to quantify vulnerability and simulate failure impacts

Key issues in road extraction

  • Roads occluded by:
    • tree canopies
    • building shadows
    • cloud cover
  • Causes “spectral blindness”
  • Need topologically connected road graphs

Part A: Robust road extraction

  • Use transformer-based vision models (examples: “former/Former Master” mentioned)
  • Infer road continuity robustly across illumination and weather conditions

Part B: Topological reconstruction

  • After extraction, fragmented networks need “healing”/connection
  • Use graph/node representation and:
    • vector dot products / distance-based connections
    • to build unified connected components

Part C: Structural intelligence / resilience matrix

  • Identify “gatekeeper nodes” and bottlenecks using centrality
  • Build a framework to stress test networks by:
    • predicting impact of localized infrastructure failure (disaster/planning)

Expected outcome

  • High-fidelity routable topology
  • Quantitative criticality map
  • Predictive impact assessment for decision makers

Data/tools mentioned

  • Start with Sentinel-2 (limitation noted: small roads)
  • For higher spatial resolution: “resour set and card set three” (subtitles unclear)
  • Training annotations:
    • manual annotation not required; datasets/libraries provide labels/layers (examples: OpenStreetMap-based approaches)
  • Suggested stack:
    • Python, PyTorch
    • segmentation models from CNNs to transformers
    • graph analysis with NetworkX, possibly PyTorch Geometric

Evaluation

  • IOU / Dice scores
  • Generalization
  • Connectivity ratio / topological accuracy
  • Average path length error

Problem Statement 5: AI-powered digital twin of India’s climate using national data

Core purpose

Build an AI + geospatial analytics climate digital twin combining:

  • satellite observations
  • ground instruments
  • physics/numerical model inputs (described generally)

Enable “what-if” scenario analysis for climate variables and extremes.

What the digital twin enables

Evaluate impacts of:

  • temperature
  • rainfall
  • greenhouse gases
  • land use/land cover
  • extreme weather events …before implementing decisions in real time.

Objectives (4 explicitly listed)

  1. Design/develop a scalable AI-driven framework using national datasets
  2. Demonstrate predictive capability for climate variables (rainfall, temperature), extensible to other parameters
  3. Produce:
    • high-resolution analysis
    • short-term predictions
    • interactive geospatial visualizations/dashboard
  4. Perform “what-if” simulations, e.g.:
    • if temperature increases by 1°C, infer rainfall spatial changes and downstream consequences

Expected outcomes

  • Proof-of-concept AI prediction capability
  • Visualization dashboard (interactive + possible 3D/simulation capability)
  • Scalable framework suitable for national deployment

Data sources mentioned

  • IMD (Indian Meteorological Department) rainfall and temperature datasets (1967 onward referenced)
  • Maximum/minimum temperatures
  • Possible ingestion of LST (Land Surface Temperature) and other variables (implied via satellite datasets)
  • Example range claims in subtitles:
    • daily rainfall from 1951 to 2025
    • spatial resolutions mentioned: 0.25° and 1°

Workflow described

  • Problem understanding → data acquisition
  • Data preprocessing:
    • spatial masking, normalization, segmentation, interpolation
  • Model development:
    • choose deep learning models suitable for simulation/prediction
  • Visualization + interactive dashboard
  • Scenario modeling/simulation
  • Evaluation/validation + prototype capability

Problem Statement 6: AI-enabled crop type + moisture stress detection + irrigation advisories (MODERATE resolution / MODIS-style)

Core purpose

Create an automated precision agriculture system that performs:

  1. Crop type classification
  2. Phenology-aware stage identification
  3. Moisture stress detection
  4. Irrigation advisory generation (for command area and rainfed regions)

Agricultural motivation

  • Crop condition varies across time periods within a season (Karif/monsoon vs others)
  • Monsoon time involves clouds and heterogeneity
  • Stress detection and irrigation guidance support:
    • drought monitoring
    • forecasting
    • operational satellite-based agriculture monitoring
    • digital agriculture initiatives

Objectives (3-part)

  1. Crop type identification
    • multi-temporal spectral signatures from optical + microwave data
    • leverage previous crop season signatures as ground truth context
  2. Phenologically aware stage modeling
    • determine crop growth stage from satellite indicators
  3. Moisture stress → crop water deficit → irrigation advisory
    • estimate water deficit and produce irrigation advisory maps for:
      • command canal command area
      • tail-end rainfed regions

Expected outcomes

  • Automated methodology + user interface for high-resolution crop type maps
  • Dashboard-ready output package containing:
    • stress maps
    • growth stage interpretation
    • pixel-level irrigation advisory
  • Scalable approach from pilot to broader regions/multiple crops/seasons

Data mentioned

  • Optical:
    • Sentinel-2, Landsat 3 (mentioned as “list three”)
  • Microwave:
    • “US4” mentioned (unclear exact mission name)
    • Sentinel-1 referenced earlier (microwave all-weather emphasis)
  • Meteorological:
    • rainfall, temperature
  • Boundaries/command area layers and ground truth (provided)

Suggested tools

  • Google Earth Engine, Python, R/MATLAB, machine learning/deep learning

Evaluation criteria

  • Prototype must classify AND derive water deficit + irrigation advisory layers
  • Crop classification accuracy target: >80–85%
  • Moisture stress classification must be:
    • growth-wise specific
    • logically consistent with satellite indicators

Problem Statement 7: AI-enabled detection of exoplanets from noisy astronomical light curves

Core purpose

Detect and characterize exoplanet transits from noisy stellar light curves.

Physics/astronomy concept

  • During a transit, a planet passes in front of a star causing periodic brightness dips.
  • Key transit parameters:
    • Depth (linked to planet size)
    • Duration (time across the stellar disk)
    • Orbital period (periodicity of dips)

Methodology described (5 steps)

  1. Detrending
    • remove non-physical trends (instrument/systematic ramps) to avoid false signals
  2. Period detection / phase folding
    • identify periodic events; phase fold using a guessed period to reveal transit shape
  3. Transit shape characterization
    • fit/interpret transit curve (planet-like trapezoidal/U-shape vs V-shape)
  4. Classification approach
    • build an AI classifier to determine:
      • whether transiting signal exists
      • classify type:
        • planet vs star vs eclipsing binary vs detector/systematic mimic
  5. Parameter prediction & evaluation
    • after detecting events, predict parameters and verify against expected significance levels

Data mentioned

  • Curated datasets from missions:
    • Kepler and K2

Evaluation approach

  • robustness of AI classifier
  • recovered signatures accuracy (planetary or stellar transits)
  • final results assessed via the criteria above

Problem Statement 8: Discover hidden lunar subsurface ice (PRL Ahmedabad mentors)

Core purpose

Use Chandrayaan-like remote sensing data to:

  • detect subsurface ice
  • pick safe landing sites
  • plan rover route to ice sites
  • estimate ice volume for future missions

Motivation/background

  • Water is essential for lunar exploration (drinking/cooking/fuel).
  • On the Moon, water is “hidden.”
  • Focus: Doubly shadowed craters near lunar poles
    • extremely cold (~25 K inside, per subtitles)
    • never receiving sunlight → preserving ice patches

Workflow (explicitly stated)

Inputs (data mentioned):

  • DFSR (“dual frequency synthetic aperture radar”) data
  • OHRC (“orbiter high resolution camera”) data

Steps:

  1. Detect subsurface ice
  2. Select landing site (safe/hazard-free for landing)
  3. Design rover travel route to reach the ice-bearing site
  4. Estimate ice volume (usable water/ice estimate)

Expected deliverables

  • High-probability subsurface ice regions
  • Radar-based detection framework
  • Feasible landing site plan
  • Optimal rover path plan
  • Ice volume estimates

Detailed guidance on analysis

  • Identify permanently shadowed regions and double shadow craters using models + imagery
  • Analyze DFSR radar signal using:
    • circular polarization ratio
    • degree of polarization
    • infer whether signal indicates subsurface ice vs rough terrain
  • For landing/rover route safety:
    • crater morphology
    • slopes
    • boulder distribution
    • surface roughness
    • terrain safety near ice-bearing regions
  • For volume estimation:
    • compute ice volume at the site after route/landing is planned

Data/software access

  • Teams can access a data portal described as:
    • Pradan, via ISRO Space Science Data Center
  • Suggested software:
    • MIDAS (radar data detection analysis mentioned)

Mission-planning framing

  • Students should convert results into an actionable exploration strategy:
    • where to land
    • where to drive
    • where to excavate
    • how much ice is available

Methodology / instruction lists captured from the session

Participation workflow (general hackathon process)

  • Register on HackxSkill
  • Search Hackathon 2026 page
  • Fill basic details to register
  • Form a team of 3–4 members
  • Select one problem statement (challenge)
  • Build/innovate a solution
  • Submit before 1 July

Example methodological directions inside problem statements

  • PS1 (Urban heat): identify hotspots → quantify drivers → physics-informed AI/ML modeling → generate optimized cooling scenarios
  • PS2 (Cloud removal): curate dataset (cloudy/cloud-free) → choose training approach → perform inference → validate with visual/spatial/spectral consistency + quality metrics
  • PS3 (AQI + HCHO):
    • Objective 1: columnar satellite data + meteorology → deep learning mapping to surface concentrations → validate with CPCB → generate surface AQI maps
    • Objective 2: satellite HCHO + fire data (biomass burning timing) + met/transport → correlate and map hotspots + source regions
  • PS4 (Route resilience): robust road extraction (transformers) → graph reconstruction/healing → centrality-based criticality → resilience simulation + predictive impact assessment
  • PS5 (Digital twin): collect national climate datasets → preprocess/normalize → select deep learning models → short-term prediction + what-if simulation → interactive geospatial dashboard → validation + scalability
  • PS6 (Agriculture): crop classification (multi-temporal optical + microwave) → phenology-aware growth stage estimation → moisture stress → water deficit estimation → pixel-level irrigation advisory maps
  • PS7 (Exoplanets): detrend → phase-fold/period search → transit shape characterization → AI classification (planet vs other causes) → parameter estimation → evaluation for robustness and recovered signatures
  • PS8 (Moon ice): detect subsurface ice (radar) → identify safe landing zones → plan rover route → estimate ice volume

Original video