Satellite change detection for Indian hazards
A remote-sensing laboratory for comparing pre-event and post-event satellite imagery and mapping selected hazard-related surface changes in India.
Project definition
Problem statement
Differences between satellite images can come from a hazard, season, sensor, viewing geometry, cloud, shadow, water level, or imperfect registration.
The engineering problem is to prepare comparable image pairs, isolate selected change categories, quantify uncertainty, and validate mapped areas without treating every pixel difference as hazard damage.
Project objectives
- Prepare public pre-event and post-event imagery for selected Indian flood, landslide, fire, or coastal-change cases.
- Align imagery, harmonise resolution, and apply cloud, shadow, no-data, and quality masks.
- Compare index, threshold, classical, and selected learned change-detection methods.
- Validate change maps with prepared reference polygons or independently interpreted samples.
- Calculate affected area, confidence, and error by administrative or physical analysis unit.
System design
System modules
Imagery registry
Records sensor, product, date, resolution, bands, orbit, licence, quality, and event metadata.
Preprocessing pipeline
Reprojects, co-registers, resamples, normalises, clips, and masks each image pair.
Change engine
Runs selected spectral, statistical, and learned methods through a common experiment interface.
Validation workspace
Manages reference samples, reviewer labels, confusion matrices, area error, and disagreement.
Hazard map
Displays paired imagery, change classes, confidence, masks, validation points, and aggregate area.
Methodology
System workflow
- 01Select case
The student opens a prepared hazard event, study boundary, image pair, and target change class.
- 02Prepare images
The pipeline aligns spatial resolution and geometry and removes invalid or cloud-affected observations.
- 03Run methods
Baseline and selected advanced methods process the same valid comparison area.
- 04Validate
Predicted change is compared with held-out reference samples and area estimates are adjusted or qualified.
- 05Publish map
The accepted experiment exports a map with source dates, masks, confidence, metrics, and limitations.
Demonstration scenario
Pre-flood and post-flood images for a prepared Indian study area are aligned and masked. A water-index baseline and a learned change method produce inundation maps. The dashboard compares them against held-out reference samples and exports the selected map with confidence and valid-area coverage.
Engineering
Technical architecture
- Web application
- Nuxt and Vue for image pairing, swipe comparison, layer controls, validation, charts, and export.
- Geospatial API
- FastAPI for imagery metadata, raster jobs, model inference, validation, summaries, and tiles.
- Spatial data layer
- PostgreSQL and PostGIS for study areas, events, footprints, reference samples, changes, and metrics.
- Raster pipeline
- Python and Rasterio for registration checks, masks, indices, feature stacks, zonal calculations, and exports.
- Model pipeline
- Transparent baselines and optional PyTorch change models with event-separated tests and saved configurations.
Testing
Evaluation
Evaluation measures
- Registration error and valid comparison-area coverage
- Per-class precision, recall, F1 score, and intersection over union
- Affected-area error against prepared reference polygons
- Performance across sensors, seasons, clouds, terrain, and event types
- False changes caused by misregistration, shadow, or seasonal differences
- Processing time across resolution, band count, and study-area size
System boundaries
- The project uses public or explicitly permitted satellite imagery and prepared hazard cases.
- Detected surface change does not establish damage, cause, ownership, compensation, or legal responsibility.
- Cloud, resolution, acquisition timing, and reference-data limitations are shown with every map.
- Outputs are educational analysis products and not operational emergency-response information.
Included
- 01Satellite-pair and study-area interface
- 02Image preparation, change detection, validation, and mapping modules
- 03Before-and-after, confidence, error, and area dashboards
- 04Prepared public imagery, hazard cases, labels, and results
- 05Complete source code in a private GitHub repository
- 06Complete project documentation with synopsis, abstract, methodology, remote-sensing workflow diagrams, change maps, screenshots, and conclusion
- 07Setup and usage guide
Project record
No buyer information is collected on this page.
- Permanent project ID
- GP-AE-14ZUPCJ
- Catalogued
- 21 Aug 2026
- Completed
- Pending
- Verified
- Pending
- Demonstration
- Added when ready
Handover
After purchase
- 01Payment is confirmed
The project is marked unavailable and cannot be purchased again.
- 02Repository access is granted
The buyer's submitted GitHub account receives access to the private repository.
- 03The purchase record is delivered
The certification sheet is prepared from the reviewed buyer details and sent privately by email.